wdiff draft-ietf-nsis-qos-nslp-02.txt draft-ietf-nsis-qos-nslp-03.txt

Next Steps in Signaling S. Van den Bosch
Internet-Draft Alcatel
Expires: August 16, November 9, 2004 G. Karagiannis
University of Twente/Ericsson
A. McDonald
Siemens/Roke Manor Research
February 16,
May 11, 2004

NSLP for Quality-of-Service signaling
draft-ietf-nsis-qos-nslp-02.txt
draft-ietf-nsis-qos-nslp-03.txt

Status of this Memo

This document is an Internet-Draft

By submitting this Internet-Draft, I certify that any applicable
patent or other IPR claims of which I am aware have been disclosed,
and is any of which I become aware will be disclosed, in full conformance accordance with
all provisions of Section 10 of RFC2026.
RFC 3668.

Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as
Internet-Drafts.

Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."

The list of current Internet-Drafts can be accessed at http://
www.ietf.org/ietf/1id-abstracts.txt.
http://www.ietf.org/ietf/1id-abstracts.txt.

The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.

This Internet-Draft will expire on August 16, November 9, 2004.

Abstract

This draft describes an NSIS Signaling Layer Protocol (NSLP) for
signaling QoS reservations in the Internet. It is in accordance with
the framework and requirements developed in NSIS.

Together with the NTLP, GIMPS, it provides functionality similar to RSVP and
extends it. The QoS-NSLP is independent of the underlying QoS
specification or architecture and provides support for different
reservation models. It is simplified by the elimination of support
for multicast flows.

This version of the draft focuses on the basic protocol structure.
It identifies the different message types and describes the basic
operation of the protocol to create, refresh, modify and teardown a
reservation or to obtain information on the characteristics of the
associated data path.

Conventions used in this document

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119.

Table of Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 5 . 6
1.1 Scope and background . . . . . . . . . . . . . . . . . . . . 5 6
1.2 Model of operation . . . . . . . . . . . . . . . . . . . . . 5 6
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . 8 . 9
3. Protocol Overview . . . . . . . . . . . . . . . . . . . . . 9 . 10
3.1 QoS Models . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 NTLP GIMPS Interactions . . . . . . . . . . . . . . . . . . . . . 10 11
3.3 Authentication and authorization . . . . . . . . . . . . . . 10 11
3.4 Aggregation . . . . . . . . . . . . . . . . . . . . . . . . 11 12
3.5 Examples of QoS NSLP Operation . . . . . . . . . . . . . . . 11 12
3.5.1 Simple Resource Reservation . . . . . . . . . . . . . . . . 12 13
3.5.2 Sending a Query . . . . . . . . . . . . . . . . . . . 14
3.5.3 Basic Receiver Initiated Reservation . . . . . . . 13
3.5.3 . . 15
3.5.4 Bidirectional Reservations . . . . . . . . . . . . . . 17
3.5.5 Use of Local QoS Models . . . . . . . . . . . . . . . . . . 14
3.5.4 20
3.5.6 Aggregate Reservations . . . . . . . . . . . . . . . . . . . 15
3.5.5 21
3.5.7 Reduced State or stateless Stateless Interior Nodes . . . . . . 23
3.6 Authorization Model Examples . . . 16 . . . . . . . . . . . . 25
3.6.1 Authorization for the two party approach . . . . . . . 25
3.6.2 Token based three party approach . . . . . . . . . . . 26
3.6.3 Generic three party approach . . . . . . . . . . . . . 26
4. Design decisions . . . . . . . . . . . . . . . . . . . . . . 18 . 27
4.1 Message types . . . . . . . . . . . . . . . . . . . . . . . 18 27
4.1.1 RESERVE . . . . . . . . . . . . . . . . . . . . . . . . . . 18 28
4.1.2 QUERY . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 28
4.1.3 RESPONSE . . . . . . . . . . . . . . . . . . . . . . . . . . 19 29
4.1.4 NOTIFY . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 29
4.2 Control information . . . . . . . . . . . . . . . . . . . . 20 30
4.2.1 Message sequencing . . . . . . . . . . . . . . . . . . . . . 20 30
4.2.2 Requesting responses . . . . . . . . . . . . . . . . . . . . 21 31
4.2.3 Message scoping . . . . . . . . . . . . . . . . . . . 31
4.2.4 State handling . . . . . 22
4.2.4 State timers . . . . . . . . . . . . . . . 32
4.2.5 State timers . . . . . . . . . 22
4.2.5 Session binding . . . . . . . . . . . . 32
4.2.6 Session binding . . . . . . . . . . 23 . . . . . . . . . 33
4.3 Layering . . . . . . . . . . . . . . . . . . . . . . . . . . 23 34
4.3.1 Local QoS models . . . . . . . . . . . . . . . . . . . . . . 23 34
4.3.2 Local control plane properties . . . . . . . . . . . . . . . 24 35
4.3.3 Aggregate reservations . . . . . . . . . . . . . . . . . . . 25 35
4.4 Extensibility . . . . . . . . . . . . . . . . . . . . . . . 25 36
4.5 Priority . . . . . . . . . . . . . . . . . . . . . . . . . . 26 37
4.6 Rerouting . . . . . . . . . . . . . . . . . . . . . . . . . 26 37
4.7 State storage . . . . . . . . . . . . . . . . . . . . . . . 28 39
4.8 Authentication and authorization . . . . . . . . . . . . . . 29 40
4.8.1 Policy Ignorant Nodes . . . . . . . . . . . . . . . . . . . 29 40
4.8.2 Policy Data . . . . . . . . . . . . . . . . . . . . . . . . 30 41
5. QoS-NSLP Functional specification . . . . . . . . . . . . . 31 . 42
5.1 QoS-NSLP Message and Object Formats . . . . . . . . . . . 42
5.1.1 Common header . . . . . . . . 31
5.1.1 Common header . . . . . . . . . . . . 42
5.1.2 Object Formats . . . . . . . . . . . 31
5.1.2 . . . . . . . . . 42
5.2 General Processing Rules . . . . . . . . . . . . . . . . . 44
5.2.1 State Manipulation . . . . . . . . . . . . . . . . . . 44
5.2.2 Message Forwarding . . . . . . . . . . . . . . . . . . 44
5.2.3 Standard Message Processing Rules . . . . . . . . . . 44
5.3 Object Formats Processing . . . . . . . . . . . . . . . . . . . . 45
5.3.1 Reservation Sequence Number . . . . 32
5.1.3 RESERVE Messages . . . . . . . . . 45
5.3.2 Response Request . . . . . . . . . . . . . . . 34
5.1.4 QUERY Messages . . . . 45
5.3.3 Bound Session ID . . . . . . . . . . . . . . . . . . . 37
5.1.5 RESPONSE 46
5.3.4 Refresh Period . . . . . . . . . . . . . . . . . . . . 46
5.3.5 Scoping . . . . . . . . . . . . . . . . . . . . . . . 47
5.3.6 Error Spec . . . . . . . . . . . . . . . . . . . . . . 48
5.3.7 Policy Data . . . . . . . . . . . . . . . . . . . . . 48
5.3.8 QSpec . . . . . . . . . . . . . . . . . . . . . . . . 48
5.4 Message Processing Rules . . . . . . . . . . . . . . . . . 48
5.4.1 RESERVE Messages . . . . . . . . . . . . . . . . . . . 48
5.4.2 QUERY Messages . . . 38
5.1.6 NOTIFY . . . . . . . . . . . . . . . . . 49
5.4.3 RESPONSE Messages . . . . . . . . . . . . . . . . . . 51
5.4.4 NOTIFY Messages . . . . . . 40 . . . . . . . . . . . . . 52
6. IANA considerations . . . . . . . . . . . . . . . . . . . . 40 . 52
7. Requirements for the NSIS Transport Layer Protocol (NTLP) (GIMPS) . 42 . 54
7.1 Session identification . . . . . . . . . . . . . . . . . . . 42 54
7.2 Support for bypassing intermediate nodes . . . . . . . . . . 42 54
7.3 Support for peer change identification . . . . . . . . . . . 42 54
7.4 Support for stateless operation . . . . . . . . . . . . . . 43 55
7.5 Last node detection . . . . . . . . . . . . . . . . . . . . 43 55
7.6 Re-routing detection . . . . . . . . . . . . . . . . . . . . 44 56
7.7 Priority of signalling messages . . . . . . . . . . . . . . 44 56
7.8 Knowledge of intermediate QoS NSLP unaware nodes . . . . . . 44 56
7.9 NSLP Data Size . . . . . . . . . . . . . . . . . . . . . . . 44 56
7.10 NAT Traversal Notification of NTLP 'D' flag value . . . . . . . . . . . 56
7.11 NAT Traversal . . . . . . . . . . . . 45
8. Open issues . . . . . . . . . . 57
8. Assumptions on the QoS model . . . . . . . . . . . . . . 45 . . . 57
8.1 Aggregation error handling Resource sharing . . . . . . . . . . . . . . . . . 45 . . . . 57
8.2 Region scoping Reserve/commit support . . . . . . . . . . . . . . . . . . 57
9. Open issues . . . . . 45
8.3 Priority of reservations . . . . . . . . . . . . . . . . . . 45
9. Security Considerations . . 57
9.1 Region scoping . . . . . . . . . . . . . . . . . . 46
9.1 Introduction and Threat Overview . . . . 57
9.2 Priority of reservations . . . . . . . . . . 46
9.2 Trust Model . . . . . . . 58
9.3 Peering agreements on interdomain links . . . . . . . . . 58
9.4 GIMPS Modifications for Refresh Overhead Reduction . . . . 59
9.5 Path state maintenance implementation at NSLP . . . . . . 59
9.6 GIMPS Path State Maintenance . . . . . . . 47
9.3 QoS Authorization . . . . . . . . 59
9.7 Protocol Operating Environment Assumptions . . . . . . . . 60
10. Security Considerations . . . . . . 49
9.3.1 Authorization for the two party approach . . . . . . . . . . 49
9.3.2 Token based three party approach . . 60
10.1 Introduction and Threat Overview . . . . . . . . . . . . . 50
9.3.3 Generic three party approach 61
10.2 Trust Model . . . . . . . . . . . . . . . . 52
9.3.4 . . . . . . . 62
10.3 Computing the authorization decision . . . . . . . . . . . . 54
10. 64
11. Change History . . . . . . . . . . . . . . . . . . . . . . . 55
11. 64
12. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 55
12. 65
13. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 55 65
14. References . . . . . . . . . . . . . . . . . . . . . . . . . 65
14.1 Normative References . . . . . . . . . . . . . . . . . . . . 55 65
14.2 Informative References . . . . . . . . . . . . . . . . . . . 55 66
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . 57 . 68
A. Object Definitions . . . . . . . . . . . . . . . . . . . . . 58 . 68
A.1 RESPONSE_REQUEST Class . . . . . . . . . . . . . . . . . . . 58 68
A.2 RSN Class . . . . . . . . . . . . . . . . . . . . . . . . . 59 69
A.3 REFRESH_PERIOD Class . . . . . . . . . . . . . . . . . . . . 59 69
A.4 SESSION_ID Class . . . . . . . . . . . . . . . . . . . . . . 60 70
A.5 SCOPING Class . . . . . . . . . . . . . . . . . . . . . . . 60 71
A.6 ERROR_SPEC Class . . . . . . . . . . . . . . . . . . . . . . 61 71
A.7 POLICY_DATA Class . . . . . . . . . . . . . . . . . . . . . 62 72
A.7.1 Base Format . . . . . . . . . . . . . . . . . . . . . . . . 62 73
A.7.2 Options . . . . . . . . . . . . . . . . . . . . . . . . . . 63 73
A.7.3 Policy Elements . . . . . . . . . . . . . . . . . . . . . . 64 74
A.8 QSPEC Class . . . . . . . . . . . . . . . . . . . . . . . . 66 76
Intellectual Property and Copyright Statements . . . . . . . 67 . 77

1. Introduction

1.1 Scope and background

This document defines a Quality of Service (QoS) NSIS Signaling Layer
Protocol (NSLP), henceforth referred to as the "QoS-NSLP". This
protocol establishes and maintains state at nodes along the path of a
data flow for the purpose of providing some forwarding resources for
that flow. It is intended to satisfy the QoS-related requirements of
[15].
[14]. This QoS-NSLP is part of a larger suite of signaling
protocols, whose structure is outlined in [3]; [15]; this defines a common
NSIS Transport Layer Protocol (NTLP) which QoS-NSLP uses to carry out
many aspects of signaling message delivery. The specification of the
NTLP is done in another document [4]. [3].

The design of QoS-NSLP is conceptually similar to RSVP [6], [5], and uses
soft-state peer-to-peer refresh messages as the primary state
management mechanism (i.e. state installation/refresh is performed
between pairs of adjacent NSLP nodes, rather than in an end-to-end
fashion along the complete signalling path. path). Although there is no
backwards compatibility at the level of protocol messages,
interworking with RSVP at a signaling application gateway would be
possible in some circumstances. QoS-NSLP extends the set of
reservation mechanisms to meet the requirements of [15], [14], in
particular support of sender or receiver-initiated reservations, as
well as a type of bi-directional reservation and support of
reservations between arbitrary nodes, e.g. edge-to-edge,
end-to-access, etc. On the other hand, there is no support for IP
multicast.

QoS-NSLP does not mandate any specific 'QoS Model', i.e. a
particular QoS provisioning method or QoS architecture; this is
similar to (but stronger than) the decoupling between RSVP and the
IntServ architecture [5]. [4]. It should be able to carry information for
various QoS models; the specification of Integrated Services for use
with RSVP given in [7] [6] could form the basis of one QoS model.

1.2 Model of operation

This section presents a logical model for the operation of the QoS-
NSLP and associated provisioning mechanisms within a single node. It
is adapted from the discussion in section 1 of [6]. [5]. The model is
shown in Figure 1.

+---------------+
| Local |
|Applications or|
|Management (e.g|
|for aggregates)|
+---------------+
^
^
V
V
+----------+ +----------+ +---------+
| QoS-NSLP | | Resource | | Policy |
|Processing|<<<<<<>>>>>>>|Management|<<<>>>| Control |
+----------+ +----------+ +---------+
. ^ | * ^
| V . * ^
+----------+ * ^
| NTLP GIMPS | * ^
|Processing| * V
+----------+ * V
| | * V
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
. . * V
| | * .............................
. . * . Traffic Control .
| | * . +---------+.
. . * . |Admission|.
| | * . | Control |.
+----------+ +------------+ . +---------+.
<-.-| Input | | Outgoing |-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.->
| Packet | | Interface | .+----------+ +---------+.
===>|Processing|====| Selection |===.| Packet |====| Packet |.==>
| | |(Forwarding)| .|Classifier| Scheduler|.
+----------+ +------------+ .+----------+ +---------+.
.............................
<.-.-> = signaling flow
=====> = data flow (sender --> receiver)
<<<>>> = control and configuration operations
****** = routing table manipulation

Figure 1: QoS-NSLP in a Node

This diagram shows an example implementation scenario where QoS
conditioning is performed on the output interface. However, this
does not limit the possible implementations. For example, in some
cases traffic conditioning may be performed on the incoming
interface, or it may be split over the input and output interfaces.

From the perspective of a single node, the request for QoS may result
from a local application request, or from processing an incoming QoS-
NSLP message.
o The 'local application case' includes not only user applications
(e.g. multimedia applications) but also network management (e.g.
initiating a tunnel to handle an aggregate, or interworking with
some other reservation protocol - such as RSVP). RSVP) and the policy
control module (e.g. for explicit teardown triggered by AAA). In
this sense, the model does not distinguish between hosts and
routers.
o The 'incoming message' case requires NSIS messages to be captured
during input packet processing and handled by the NTLP. GIMPS. Only
messages related to QoS are passed to the QoS-NSLP. The NTLP GIMPS may
also generate triggers to the QoS-NSLP (e.g. indications that a
route change has occurred).

The QoS request is handled by a local 'resource management' function,
which coordinates the activities required to grant and configure the
resource.
o The grant processing involves two local decision modules, 'policy
control' and 'admission control'. Policy control determines
whether the user has administrative permission to make the
reservation. Admission control determines whether the node has
sufficient available resources to supply the requested QoS.
o If both checks succeed, parameters are set in the packet
classifier and in the link layer interface (e.g., in the packet
scheduler) to obtain the desired QoS. Error notifications are
passed back to the request originator. The resource management
function may also manipulate the forwarding tables at this stage,
to select (or at least pin) a route; this must be done before
interface-dependent actions are carried out (including forwarding
outgoing messages over any new route), and is in any case
invisible to the operation of the protocol.

Policy control is expected to make use of a AAA service external to
the node itself. Some discussion can be found in [16] and [17].
More generally, the processing of policy and resource management
functions may be outsourced to an external node leaving only 'stubs'
co-located with the NSLP; this is not visible to the protocol
operation, although it may have some influence on the detailed design
of protocol messages to allow the stub to be minimally complex. A
more detailed discussion on authentication and authorization can be
found in Section 4.8. The definition of the POLICY_DATA class is
given in Appendix A.7.

The group of user plane functions, which implement QoS for a flow
(admission control, packet classification, and scheduling) is
sometimes known as 'traffic control'.

Admission control, packet scheduling, and any part of policy control
beyond simple authentication have to be implemented using specific
definitions for types and levels of QoS; we refer to this as a QoS
model. Our assumption is that the QoS-NSLP is independent of the QoS
model, that is, QoS parameters (e.g. IntServ service elements) are
interpreted only by the resource management and associated functions,
and are opaque to the QoS-NSLP itself. QoS Models are discussed
further in Section 3.1.

The final stage of processing for a resource request is to indicate
to the QoS-NSLP protocol processing that the required resources have
been configured. The QoS-NSLP may generate an acknowledgement
message in one direction, and may propagate the resource request
forwards in the other. Message routing is (by default) carried out
by the NTLP GIMPS module. Note that while Figure 1 shows a unidirectional
data flow, the signaling messages can pass in both directions through
the node, depending on the particular message and orientation of the
reservation.

2. Terminology

The terminology defined in [3] applies to this draft. In addition,
the following terms are used:
QNE: an NSIS Entity (NE), which supports the QoS-NSLP.
QNI: the first node in the sequence of QNEs that issues a reservation
request.
request for a session.
QNR: the last node in the sequence of QNEs that receives a
reservation request. request for a session.
Source or message source: The one of two adjacent NSLP peers that is
sending a signalling message (maybe the upstream or the downstream
peer). NB: this is not necessarily the QNI.

QoS NSLP nodes
IP address (QoS unware NSIS nodes are IP address
= Flow not shown) = Flow
Source | | | Destination
Address | | | Address
V V V
+--------+ Data +------+ +------+ +------+ +--------+
| Flow |-------|------|------|------|-------|------|---->| Flow |
| Sender | Flow | | | | | | |Receiver|
+--------+ | QNI | | QNE | | QNR | +--------+
| | | | | |
+------+ +------+ +------+
=====================>
<=====================
Signaling
Flow

3. Protocol Overview

The QoS NSLP uses four message types: RESERVE, QUERY, RESPONSE and
NOTIFY. These contain three types of objects: Control Information
(CI), QSpecs, and Policy objects. The set of objects permissible
depends on the message type.

Messages are passed to

An interface exists between the NTLP to be delivered to neighbouring NSIS
nodes. Similarly, QoS NSLP data from NTLP messages is passed processing and GIMPS processing
for sending/receiving NSIS messages. In addition to the
QoS NSLP component for processing. Additional message
data itself, other meta-data (e.g. session identifier, NSLP identifier) flow routing
information) can also be sent in both directions.

The QoS NSLP separates the actual description of resources from the
QoS signalling protocol used to transport them. It uses
interchangeable QoS Models that allow the resource specification to be performed in various ways, and to provide different processing
models (including reserve/commit models, measurement based models,
etc). transferred across this interface.

Control information objects carry general information for the QoS
NSLP processing, such as sequence numbers or whether a response is
required.

QSpec objects describe the actual resources that are required and are
specific to the QoS Model being used. Besides any resource
description they may also contain QoS Model specific control
information used by the QoS Model's processing.

The Policy objects contain data used to authorise the reservation of
resources.

3.1 QoS Models

A QoS model is a defined mechanism for achieving QoS as a whole. The
specification of a QoS model includes a description of its QoS
parameter information, as well as how that information should be
treated or interpreted in the network. In that sense, the QoS model
goes beyond the QoS-NSLP protocol level in that it could also
describe underlying assumptions, conditions and/or specific
provisioning mechanisms appropriate for it.

A QoS model provides a specific set of parameters to be carried in
the QSpec, or restricts the values these parameters can take.
Integrated Services [5], [4], Differentiated Services [9] [8] and RMD [22] are
all examples that could provide the basis of an NSIS QoS model.
There is no restriction on the number of QoS models. QoS models may
be local (private to one network), implementation/vendor specific, or
global (implementable by different networks and vendors). The
authors are currently aware of three efforts related to QoS model
specification: [18], [19] and [20]. This specification will not
attempt to select between the moppling number of possible QoS models.

The QSpec contains a set of parameters and values describing the
requested resources. It is opaque to the QoS-NSLP and similar in
purpose to the TSpec, RSpec and AdSpec specified in [6][7]. [5][6]. At each
QNE, its content is interpreted by the resource management function
for the purposes of policy control and traffic control (including
admission control and configuration of the packet classifier and
scheduler).

3.2 NTLP GIMPS Interactions

The QoS NSLP uses the NTLP GIMPS for delivery of all its messages. Messages
are normally passed from the NSLP to the NTLP GIMPS via an API, which also
specifies additional information, including an identifier for the
signaling application (e.g. 'QoS-NSLP'), the flow/session
identifier, and an indication of the intended direction - towards
data sender or receiver. On reception, the NTLP GIMPS provides the same
information to the QoS-NSLP.

The QoS NSLP does not provide any method of interacting with
firewalls or Network Address Translators (NATs). It assumes that a
basic NAT traversal service is provided by the NTLP (as described in
the requirement given in Section 7.10). GIMPS.

3.3 Authentication and authorization

The QoS signaling protocol needs to exchange information which is
subsequently used as input to the AAA infrastructure. The response
from the AAA infrastructure must also returned and processed by the
respective entities.

+-------------+
| Entity |
| authorizing |
| resource |
| request |
+-----+-------+
|
|
/-\----+-----/\
//// \\\\
|| ||
| AAA Cloud |
|| ||
\\\\ ////
\-------+-----/
|
+-------------+ QoS signaling +---+--+
| Entity |<=================>| |<=========>
| requesting | Data Flow | QNE |
| resource |-------------------|------|---------->
+-------------+ +------+

3.4 Aggregation

In some cases it is desirable to create reservations for an
aggregate, rather than on a per-flow basis, in order to reduce the
amount of reservation state needed as well as the processing load for
signalling messages.

The QoS NSLP, therefore, provides facilities to provide similar
aggregation facilities to [11]. [10]. However, the aggregation scenarios
supported are wider than that proposed there. Aggregate reservations
are further described in Section 4.3.3.

3.5 Examples of QoS NSLP Operation

The QoS NSLP can be used in a number ways. The examples given here
give an indication of some of the basic processing. However, they
are not exhaustive and do not attempt to cover the details of the
protocol processing.

3.5.1 Simple Resource Reservation

NI NF NF NR
| | | |
| RESERVE | | |
+--------->| | |
| | RESERVE | |
| +--------->| |
| | | RESERVE |
| | +--------->|
| | | |
| | | RESPONSE |
| | |<---------+
| | RESPONSE | |
| |<---------+ |
| RESPONSE | | |
|<---------+ | |
| | | |
| | | |

Figure 4: Basic Sender Initiated Reservation

To make a new reservation, the QNI constructs a RESERVE message
containing a QSpec object, from its chosen QoS model, which describes
the required QoS parameters.

The RESERVE message is passed to the NTLP GIMPS which transports it to the
next QoS NSLP node. There it is delivered to the QoS NSLP processing
which examines the message. Policy control and admission control
decisions are made. The exact processing also takes into account the
QoS Model being used. The node performs appropriate actions (e.g.
installing reservation) based on the QSpec object in the message.

The QoS NSLP then generates a new RESERVE message (usually based on
the one received). This is passed to the NTLP, GIMPS, which forwards it to
the next QNE.

The same processing is performed at further QNEs along the path, up
to the QNR. The determination that a node is the QNR may be made
directly (e.g. that node is the destination for the data flow), or
using some NTLP GIMPS functionality to determine that there are no more
QNEs between this node and the data flow destination.

Any node may include a request for a RESPONSE in its RESERVE
messages. One such use is to confirm the installation of state,
which allows the use of summary refreshes that later refer to that
state. The RESPONSE is forwarded peer-to-peer along the reverse of
the path that the RESERVE message took (using NTLP GIMPS path state), and
so is seen by all the QNEs on the reverse-path. It is only forwarded
as far as the node which requested the RESPONSE. A RESPONSE message
can also indicate an error when, for example, a reservation has
failed to be installed.

The reservation can subsquently be refreshed by sending further
RESERVE messages containing the complete reservation information, as
for the initial reservation. The reservation can also be modified in
the same way, by changing the QoS model specific data to indicate a
different set of resources to reserve.

The overhead required to perform refreshes can be reduced, in a
similar way to that proposed for RSVP in [10]. [9]. Once a RESPONSE
message has been received indicating the successful installation of a
reservation, subsequent refreshing RESERVE messages can simply refer
to the existing reservation, rather than including the complete
reservation specification.

3.5.2 Sending a Query

QUERY messages can be used to gather information from QNEs along to
path. For example, it can be used to find out what resources are
available before a reservation is made.

In order to perform a query along a path, the QNE constructs a QUERY
message. This message includes QoS model specific objects containing
the actual query to be performed at QoS NSLP nodes along the path.
It also contains an object used to match the response back to the
query, and an indicator of the query scope (next node, whole path).

The QUERY message is passed to the NTLP GIMPS to forward it along the path.
The NTLP may use datagram mode or connection mode for forwarding the
QUERY message.

A QNE (including the QNR) receiving a QUERY message should inspect it
and create a new message, based on that received with the query
objects modified as required. For example, the query may request
information on whether a flow can be admitted, and so a node
processing the query might record the available bandwidth. The new
message is then passed to the NTLP GIMPS for further forwarding (unless it
knows it is the QNR, or is the limit for the scope in the QUERY).

At the QNR, a RESPONSE message is generated. must be generated if the QUERY
includes a RESPONSE_REQUEST object. Into this is copied various
objects from the received QUERY message. It is then passed to
the NTLP GIMPS
to be forwarded peer-to-peer back along the path. This makes
use of the neighbour state retained by the NTLP, and may use datagram
or connection mode.

Each QNE receiving the RESPONSE message should inspect the
ResponseRequest
RESPONSE_REQUEST object to see if it 'belongs' to it (i.e. it was
the one that originally created it). If it does not then it simply
passes the message back to the NTLP GIMPS to be forwarded back down the path.

3.5.3 Basic Receiver Initiated Reservation

As described in [15] in some signaling applications, a node at one
end of the data flow takes responsibility for requesting special
treatment - such as a resource reservation - from the network. Both
ends then agree whether sender or receiver initiated reservation is
to be done. In case of a receiver initiated reservation, both ends
agree whether a "One Pass With Advertising" (OPWA) [23] model is
being used. This negotiation can be accomplished using mechanisms
that are outside the scope of NSIS, see Section 9.7.

To make a receiver initiated reservation, see Figure 5, the QNI
constructs a QUERY message containing a QSpec object, from its chosen
QoS model, which describes, among others, the required QoS
parameters. This QUERY message does not need to trigger a RESPONSE
message and therefore, the QNI must not include the RESPONSE_REQUEST
object, see Section 5.4.2, into the QUERY message. The QUERY message
may be used to to gather information along the path, which is carried
by the QSPEC object. An example of such information is the "One Pass
With Advertising" (OPWA) [23]. This QUERY message is instructing the
QoS-NSLP process running on each QNE's located on the path followed
by this QUERY message to install GIMPS reverse-path state.

NI NF NF NR
| | | |
| QUERY | | |
+--------->| | |
| | QUERY | |
| +--------->| |
| | | QUERY |
| | +--------->|
| | | |
| | | RESERVE |
| | |<---------+
| | RESERVE | |
| |<---------+ |
| RESERVE | | |
|<---------+ | |
| | | |
| RESPONSE | | |
+--------->| | |
| | RESPONSE | |
| +--------->| |
| | | RESPONSE |
| | +--------->|
| | | |

Figure 5: Basic Receiver Initiated Reservation

The QUERY message is transported by the NTLP to the next downstream
QoS NSLP node. There it is delivered to the QoS NSLP processing
which examines the message. The exact processing also takes into
account the QoS Model being used. The node performs appropriate
actions, such as installing GIMPS reverse path state. Another action
that may be performed by the node is to gather advertising
information that may be used to predict the end-to-end QoS.

The QoS NSLP then generates a new QUERY message (usually based on the
one received). This is passed to the NTLP, which forwards it to the
next QNE. The same processing is performed at further QNEs along the
path, up to the receiver, which in this situation is the QNR. The
QNR detects that this QUERY message does not carry a RESPONSE_REQUEST
object and by using the information contained in the received QUERY
message, such as the QSPEC, constructs a RESERVE message, which can
be used as a reservation request.

The RESERVE message must follow the same path that has been followed
by the QUERY message. Therefore, the NTLP is informed, over the
NSLP/NTLP API, to pass the message upstream, i.e., by setting the
GIMPS "D" flag, see [3].

The RESERVE is forwarded peer-to-peer along the reverse of the path
that the QUERY message took (using the NTLP reverse path state). It
is only forwarded as far as the QNI. The RESERVE message received by
a QNE is delivered to the QoS-NSLP processing, which examines the
message. The NTLP includes the value of the 'D' flag in the
information it passes with the message over the NTLP/NSLP API
(Section 7.10). The exact processing of this message is accomplished
in the same way as for the processing of a RESERVE message received
by a QNE in a sender initiated approach. The main difference is that
the RESERVE message has to be forwarded peer-to-peer along the
reverse of the path that the QUERY message took. Therefore, the
QoS-NSLP functionality of each QNE requires the NTLP, over the NSLP/
NTLP API, to pass the message upstream, i.e.,by setting the GIMPS "D"
flag.

Similar to the sender initiated approach, any node may include a
request for a RESPONSE in its RESERVE messages. The RESPONSE is
forwarded peer-to-peer along the path that the initial QUERY message
took (using NTLP path state). It is only forwarded as far as the
node which requested the RESPONSE. Note that the use of RESPONSE is
optional.

The reservation can subsquently be refreshed in the same was as for
the refresh procedure of a reservation in a sender initiated
approach, by sending further RESERVE messages containing the complete
reservation information, as for the initial reservation. This
RESERVE message may be also used to refresh the NTLP reverse path
state. Additionally, refreshing the NTLP reverse path state could be
performed by sending periodic QUERY messages. However, it might
potentially also be done by the NTLP, without needing any downstream
NSLP messages. This is an open issue (see Section 9.6), and it will
be worked out in a future version of this draft.

3.5.4 Bidirectional Reservations

A bi-directional reservation combines the reservations for a pair of
coupled flows going in opposite directions. The main difficulty here
is that the two flows, although between the same end points, may
traverse different paths across the Internet.

We distinguish two types of bi-directional reservations:
o sender+sender: where a sender-initiated reservation is done in one
direction, e.g., from QNE A towards QNE B, and then another
sender-initiated reservation that is done back, the opposite
direction, i.e., from QNE B towards QNE A,
o sender+receiver: where a sender-initiated reservation is done in
one direction, e.g., from QNE A towards QNE B, and a
receiver-initiated reservation that is done for the opposite
direction, i.e., a reservation from QNE A towards QNE B for the
data flow from QNE B to QNE A.

Both ends have to agree on which bi-directional reservation type they
need to use. This negotiation/agreement can be accomplished using
mechanisms that are outside the scope of NSIS, see Section 9.7.

In the sender+sender bi-directional reservation scenario (see Figure
6) the QNI, i.e., QNE A, generates a sender initiated reservation by
creating and sending a RESERVE message towards the QNR, i.e., QNE B.

When performing this type of bi-directional reservation, there are
situations where the QNR, i.e., QNE B, is not able to retrieve the
QoS parameter set required to perform a reservation in the downstream
direction from QNE B to QNE A, while the QNI, i.e., QNE A, is able to
retrieve and maintain the Qos parameter sets required to perform the
reservation in both downstream directions. In these situations there
is a need to carry both QoS parameter sets in QoS-NSLP messages from
the QNI, i.e., QNE A, to the QNR, i.e., QNE B.

The RESERVE is passed to the NTLP, which forwards it to the next QNE.
The same processing is performed at further QNEs along the path, up
to the receiver, which in this situation is the QNR, i.e., QNE B.
Considering that the QNE B, knows that it has to start a
sender-initiated reservation, the QNE B creates and sends a RESERVE
message downstream towards the QNE A. This RESERVE message carries
among others the QoS parameter set required to perform the sender
initiated reservation in the downstream direction from QNE B to QNE
A. Furthermore, the QNI, (i.e., QNE A), and QNR, (i.e., QNE B),
might be willing to bind the two sessions. In this situation the
QoS-NSLP messages that are used in the direction from QNE B to QNE A,
may carry information that identifies the two bound sessions. This
can be accomplished by using the BOUND_SESSION_ID object.

Note that any node may generate RESPONSE messages as answer to the
received RESERVE messages, but this is optional.

A NF1 NF2 NF3 NF4 B
| RESERVE | | | | |
+--------->|RESERVE | | | |
| +-------------------->| RESERVE | |
| | | +-------------------->|
| | | | | RESERVE |
| | | RESERVE | |<---------|
| RESERVE | |<--------------------| |
|<--------------------| | | |
| | | | | |

Figure 6: Bi-directional reservation for sender+sender scenario

In the sender+receiver bi-directional reservation scenario (see
Figure 7) the QNI, i.e., QNE A, generates both a sender initiated and
a receiver-initiated reservation. Note that before that the QNI,
i.e., QNE A, is starting this procedure it must receive a QUERY
message from the QNR, i.e., QNE B, that requires from the QNI a
receiver initiated reservation, see Section 3.5.3. Note that the
QUERY message is carrying the QSPEC that can be used by the receiver
initiated reservation.

The QNI, i.e., QNE A, creates two RESERVE messages. One RESERVE
message that is used for the sender initiated reservation (session
A->B), see Section [(reference to the sender initiated section)] and
another RESERVE message that is used for the receiver initiated
reservation (session B->A), see Section 3.5.3. Furthermore, the
QNI, (i.e., QNE A), and QNR, (i.e., QNE B), might be willing to bind
the two sessions. In this situation the QoS-NSLP messages that are
used in the bi-directional scenario may carry information that
identifies the two bound sessions. This can be accomplished by using
the BOUND_SESSION_ID object. When these two RESERVE messages follow
exactly the same path, then it should be possible that these two
RESERVE messages are bundled at the NTLP level, by using the GIMPS
message bundling feature.

Note that any node may generate RESPONSE messages as answer to the
received RESERVE messages, but this is optional.

A NF1 NF2 B
| | | QUERY |
| | QUERY |<--------------------|
| QUERY |<--------------------| |
|<---------+ | |
| RESERVE | | |
+--------->|RESERVE | |
| +-------------------->| RESERVE |
| | +-------------------->|
| RESERVE | | |
+--------->|RESERVE | |
| +-------------------->| RESERVE |
| | +-------------------->|
| | | |

Figure 7: Bi-directional reservation for sender+receiver scenario

3.5.5 Use of Local QoS Models

In some cases it may be required to use a different QoS Model along a
particular segment of the signalling. In this case a node at the
edge of this region needs to map between the two resource
descriptions (and any auxiliary data).

+--------+ +----+----+ +--------+ +----+----+ +--------+

| QM1 | |QM1 | QM2| | QM2 | |QM2
| QM1| RESERVE | QM1 |
+--------+ +----+----+ +--------+ +----+----+ +--------+
|QoS-NSLP| |QoS-NSLP | |QoS-NSLP| |QoS-NSLP | |QoS-NSLP|
+--------+ +---------+ +--------+ +---------+ +--------+
| NTLP |===| NTLP |===| NTLP |===| NTLP |===| NTLP {QSpec1} | | | |
+-------------->| | | |
| | RESERVE | | |
| |{QSpec2,QSpec1}| | |
| +-------------->| | |
| | | RESERVE | |
| | |{QSpec2,QSpec1}| |
| | +-------------->| |
| | | | RESERVE |
| | | | {QSpec1} |
| | | +-------------->|
| | | | |
+--------+ +---------+ +--------+ +---------+ +--------+

<-------> <--------------------> <------->
RESV{QSpec1} RESV{QSpec1,QSpec2} RESV{QSpec1}

Figure 5: 8: Reservation with local QoS Models

This initially proceeds as for the basic example, with peer-to-peer
installation of reservations. However, within a region of the
network a different QoS Model needs to be used. At the edge of this
region the QNEs support both the end-to-end and local QoS models.

When the RESERVE message reaches the QNE at the ingress, the initial
processing of the RESERVE proceeds as normal. However, the QNE also
determines the appropriate description using the second QoS model.
The RESERVE message to be sent out is constructed mostly as usual but
with a second QSpec object added, which becomes the 'current' one.

When this RESERVE message is received at the next node the QoS NSLP
only uses the QSpec at the top of the stack (i.e. the 'current'
one), rather than the end-to-end QSpec. Otherwise, processing
proceeds as usual. The RESERVE message that it generates should
include the complete stack of QSpecs from the message it received.

At the QNE at the egress of the region the local QSpec is removed
from the message so that subsequent QNEs receive only the end-to-end
QSpec.

QSpecs can be stacked in this way to an arbitrary depth.

3.5.4

3.5.6 Aggregate Reservations

In order to reduce signalling and per-flow state in the network, the
reservations for a number of flows may be aggregated together.

NI NF NF/NI' NF' NR'/NF NR
aggregator deaggregator
| | | | | |
| RESERVE | | | | |
+--------->| | | | |
| | RESERVE | | | |
| +--------->| | | |
| | | RESERVE | | |
| | +-------------------->| |
| | | RESERVE' | | |
| | +=========>| RESERVE' | |
| | | +=========>| RESERVE |
| | | | +--------->|
| | | | RESPONSE'| |
| | | RESPONSE'|<=========+ |
| | |<=========+ | |
| | | | | RESPONSE |
| | | | RESPONSE |<---------+
| | |<--------------------+ |
| | RESPONSE | | | |
| |<---------+ | | |
| RESPONSE | | | | |
|<---------+ | | | |
| | | | | |
| | | | | |

Figure 6: 9: Sender Initiated Reservation with Aggregation

An end-to-end per-flow reservation is initiated as normal (with
messages shown in Figure 6 9 as "RESERVE").

At the aggregator a reservation for the aggregated flow is initiated
(shown in Figure 6 9 as "RESERVE'"). This may use the same QoS model
as the end-to-end reservation but has a flow identifier for the
aggregated flow (e.g. tunnel) instead of for the individual flows.

Markings are used so that intermediate routers do not need to inspect
the individual flow reservations. This might be done by creating an
NTLP connection mode association between the aggregator and The deaggregator then becomes the
next hop QoS NSLP node for the end-to-end per-flow reservation.

Aggregator Deaggregator

+---+ +---+ +---+ +---+
|QNI|-----|QNE|-----|QNE|-----|QNR| aggregate
+---+ +---+ +---+ +---+ reservation

+---+ +---+ ..... ..... +---+ +---+
|QNI|-----|QNE|-----. .-----. .-----|QNE|-----|QNR| end-to-end
+---+ +---+ ..... ..... +---+ +---+ reservation

The deaggregator acts as the QNR for the aggregate reservation.

Information is carried in the reservations to enable the deaggregator
to associate the end-to-end and aggregate reservations with one
another. For example, this is necessary so that the size of the
aggregate reservation can be reduced when the end-to-end reservation
is removed.

The key difference between this example, and previous ones is that
the flow identifier for the aggregate is expected to be different to
that for the end-to-end reservation. The aggregate reservation can
be updated independently of the per-flow end-to-end reservations.

3.5.5

3.5.7 Reduced State or stateless Stateless Interior Nodes

This example uses a different QoS model within a domain, in
conjunction with NTLP GIMPS and NSLP functionality which allows the
interior nodes to avoid storing NTLP GIMPS and QoS NSLP state. As a
result the interior nodes only store the QoS model specific
reservation state, or even no state at all. This allows the QoS
model to use a form of "reduced-state" operation, where reservation
states with a coarser granularity (e.g. per-class) are used, or a
"stateless" operation where no reservation state is needed (or
created).

The key difference between this example and the use of different QoS
Models in Section 3.5.3 3.5.5 is that the transport characteristics for the
'local' reservation can be different from that of the end-to-end
reservation, i.e. the NTLP GIMPS can be used in a different way for the
edge-to-edge and hop-by-hop sessions. The reduced state reservation
can be updated independently of the per-flow end-to-end reservations.

NF NF NF NF
ingress interior interior egress
NTLP
GIMPS stateful NTLP GIMPS stateless NTLP GIMPS stateless NTLP GIMPS stateful
| | | |
RESERVE | | | |
-------->| RESERVE | | |
+--------------------------------------------->|
| RESERVE' | | |
+-------------->| | |
| | RESERVE' | |
| +-------------->| |
| | | RESERVE' |
| | +------------->|
| | | | RESERVE
| | | +-------->
| | | | RESPONSE
| | | |<--------
| | | RESPONSE |
|<---------------------------------------------+
RESPONSE| | | |
<--------| | | |

Figure 8: 11: Reservation with Reduced State Interior Nodes

The QNI performs the same processing as before to generate the
initial RESERVE message, and it is forwarded by the NTLP GIMPS as usual. At
the QNEs at the edges of the stateless or reduced-state region the
processing is different and the nodes support two QoS models.

At the ingress the original RESERVE message is forwarded but using
facilities provided ignored
by the NTLP to bypass the stateless or reduced-state nodes. The egress node is the next
QoS NSLP hop for that session. After the initial discovery phase
using datagram mode, connection mode between the ingress and egress
can be used. At the egress node the RESERVE message is then
forwarded normally.

At the ingress a second RESERVE' message is also built. This makes
use of a QoS model suitable for a reduced state or stateless form of
operation (such as the RMD per hop reservation). When processed by
interior (stateless) nodes the QoS NSLP processing excercises its
options to not keep state wherever possible, so that no per flow QoS
NSLP state is stored. Some state, e.g. per class, for the QoS model
related data may be held at these interior nodes. The QoS NSLP also
requests that the NTLP GIMPS use different transport characteristics (i.e.
sending of messages in datagram mode, and not retaining optional
reverse path state).

Nodes, such as those in the interior of the stateless or
reduced-state domain, that do not retain reservation state cannot
send back RESPONSE messages (and so cannot use summary refreshes) cannot send back RESPONSE messages. refreshes).

At the egress node the RESERVE' message is interpreted in conjunction
with the reservation state from the end-to-end RESERVE message (using
information carried in the message to correlate the signalling
flows). The RESERVE message is only forwarded further if the
processing of the RESERVE' message was successful at all nodes in the
local domain, otherwise the end-to-end reservation is regarded as
having failed to be installed.

Since NTLP GIMPS neighbour relations are not maintained in the
reduced-state region, only sender initiated signalling can be
supported. If a bi-directional reservation is required then the
interior
end-to-end QoS model must provide an object that requests the egress last
node to generate a sender initiated session in the reverse direction.

3.6 Authorization Model Examples

Various authorization models can be used in conjunction with the QoS
NSLP.

3.6.1 Authorization for the two party approach

The two party approach is conceptually the simplest authorization
model.

+-------------+ QoS request +--------------+
| Entity |----------------->| Entity |
| requesting | | authorizing |
| resource |granted / rejected| resource |
| |<-----------------| request |
+-------------+ +--------------+
^ ^
+...........................+
financial establishment

Figure 12: Two party approach

In this example the the authorization decision only involves the two
entities, or makes use of previous authorisation using an out-of-band
mechanism to avoid the need for active participation of an external
entity during the NSIS protocol execution.

This type of model may be applicable, for example, between two
neighboring networks (inter-domain signaling) where a long-term
contract (or other out-of-band mechanisms) exist to manage charging
and provide sufficient information to authorize individual requests.

3.6.2 Token based three party approach

An alternative approach makes use of authorization tokens, such as
those described in [11] and [12] or used as part of the Open
Settlement protocol [27]. The former ('authorization tokens') are
used to associate two different signaling protocols (i.e. SIP and
NSIS) and their authorization with each other whereas the latter is a
form of digital money. As an example, with the authorization token
mechanism, some form of authorization is provided by the SIP proxy,
which acts as the resource authorizing entity in Figure 13. If the
request is authorized, then the SIP signaling returns an
authorization token which can be included in the QoS signaling
protocol messages to refer to the previous authorization decision.
The tokens themselves may take a number of different forms, some of
which may require the entity performing the QoS reservation to query
external state.

Authorization
Token Request +--------------+
+-------------->| Entity C | financial settlement
| | authorizing | <..................+
| | resource | .
| +------+ request | .
| | +--------------+ .
| | .
| |Authorization .
| |Token .
| | .
| | .
| | .
| | QoS request .
+-------------+ + Authz. Token +--------------+ .
| Entity |----------------->| Entity B | .
| requesting | | performing | .
| resource |granted / rejected| QoS | <..+
| A |<-----------------| reservation |
+-------------+ +--------------+

Figure 13: Token based three party approach

For the digital money type of systems (e.g. OSP tokens), the token
represents a limited amount of credit. So, new tokens must be sent
with later refresh messages once the credit is exhausted.

3.6.3 Generic three party approach

Another method is for the node performing the QoS reservation to
delegate the authorization decision to a third party, as illustrated
in Figure 14.

+--------------+
| Entity C |
| authorizing |
| resource |
| request |
+-----------+--+
^ |
| |
QoS | | QoS
authz| |authz
req.| | res.
| |
QoS | v
+-------------+ request +--+-----------+
| Entity |----------------->| Entity B |
| requesting | | performing |
| resource |granted / rejected| QoS |
| A |<-----------------| reservation |
+-------------+ +--------------+

Figure 14: Three party approach

Authorization may be performed on a per-request basis, periodically,
or on a per-session basis. The authorization request might make use
of EAP authentication between entities A and C, and a subsequent
protocol exchange between A and B to create a secure channel for
further communications. Such a technique gives flexibility in terms
of the authentication and key exchange protocols used.

A further extension to this model is to allow Entity C to reference a
AAA server in the user's home network when making the authorization
decision.

4. Design decisions

4.1 Message types

The QoS-NSLP specifies four message types: RESERVE, QUERY, RESPONSE
and NOTIFY.

The fundamental properties of each message determine how it is
analyzed from the perspective of re-ordering, loss, end-to-end
reliability and so on. It is important for applications to know
whether NSLP messages can be repeated, discarded or merged and so on
(e.g. for edge mobility support, rerouting, etc). Indeed, the
ordering of messages that do not manipulate state at QNEs matters
little, does not
matter, whereas the way that messages that manipulate state are
interleaved matters very much. Therefore NSLP is designed such that
the message type identifies whether a message is manipulating state
(in which case it is idempotent) or not (it is impotent).

4.1.1 RESERVE

The RESERVE message is the only message that manipulates QoS
reservation state. It is used to create, refresh, modify and remove
such state. The common message header contains a TEAR flag that
indicates complete QoS NSLP state removal (as opposed to a
reservation of zero resources). This QoS NSLP state comprises
reservation state and QoS NSLP operation state. The TEAR flag QoS NSLP
indicates to the NTLP GIMPS that it is no longer interested in the
corresponding NTLP (reverse) state is not required. GIMPS state. The NTLP
the GIMPS then autonomously decides
whether or not to keep such state or not. this state.

The RESERVE message opaquely transports one or more QSPEC objects,
describing the desired service level and a POLICY_DATA object,
authorizing the requestor of the service. It carries the lifetime of
the reservation in the Common Control Information.

RESERVE messages are sent peer-to-peer. This means that a QNE
considers its adjacent upstream or downstream peer to be the source
of the RESERVE message.

The RESERVE message is idempotent; the resultant effect is the same
whether a message is received once or many times. In addition, the
ordering of RESERVE messages matters - an old RESERVE message should
not replace a newer one. Both of these features are required for
protocol robustness - messages may be re-ordered on route (e.g.
because of mobility, or at intermediate NTLP GIMPS nodes) or spuriously
retransmitted. Message Handling of message re-ordering is supported by the
inclusion of the Reservation Sequence Number (RSN) in the RESERVE
message.

The sender of a RESERVE message may want to receive confirmation of
successful state installation from a downstream node. Therefore, a
RESERVE message optionally contains a RESPONSE_REQUEST object
(Section 4.2.2).

4.1.2 QUERY

A QUERY message is used to request information about the data path
without making a reservation. This functionality can be used to
'probe' the network for path characteristics or for support of
certain QoS models. The information obtained from a QUERY may be
used in the admission control process of a QNE (e.g. in case of
measurement-based admission control). Note that a QUERY does not
change existing reservation state, nor does it cause state to be
installed in nodes other than the one that generated the QUERY.

A QUERY message contains one or more QSPEC objects and a POLICY_DATA
object. The QSPEC object describes what is being queried for and may
contain objects that gather information along the data path. The
POLICY_DATA object authorizes the requestor of the QUERY message.

A QUERY message may be scoped if a RESPONSE message from some other
node than the QNR is desired.

A QUERY message must may contain a RESPONSE_REQUEST object (Section
4.2.2), the contents of which allow matching back RESPONSE messages
to the QUERY request. The RESPONSE_REQUEST object is transported
unchanged along the data path and may be used to scope the RESPONSE
to a QUERY message (Section 4.2.3).

4.1.3 RESPONSE

The REPONSE message is used to provide information about the result
of a previous QoS-NSLP message. This includes explicit confirmation
of the state manipulation signaled in the RESERVE message, the
response to a QUERY message or an error code if the QNE or QNR is
unable to provide the requested information or if the response is
negative. For this purpose, the RESPONSE message carries one or more
QSPEC objects.

The RESPONSE message is impotent, it does not cause any reservation
state to be installed or modified.

The forwarding of the RESPONSE message along the path does not
necessarily imply the existence of NTLP reverse-path state at every
node. For example, the NTLP may have a mechanism to pass a message
directly from the egress to the ingress of a region of QNEs that do
not store per-flow reverse-path state.

4.1.4 NOTIFY

NOTIFY messages are used to convey information to a QNE. NOTIFY
messages are impotent (they do not cause a change in state directly).
They may carry one or more QSPEC objects. An example use of NOTIFY
would be to indicate when a reservation has been pre-empted.

NOTIFY messages differ from RESPONSE messages in that they need not
refer to any particular state or previously received message. They
are sent asynchronously. The NOTIFY message itself does not trigger
or mandate any action in the receiving QNE.

The information conveyed by a NOTIFY message is typically related to
error conditions. One example would be notification to an upstream
peer about state being torn down.

4.2 Control information

Control information conveys information on how specific messages
should be handled by a QNE, e.g. sequencing of messages. This may
include some mechanisms that are useful for many QoS models (Common
Control Information) and some that are for a particular QoS model
only (QoS-model specific Control Information). QoS-model specific
Control Information is specified together with a QoS model. This
specification only defines Common Control Information. Currently,
Common Control Information is defined for session identification,
message sequencing, response request, message scoping and session
lifetime.

4.2.1 Message sequencing

RESERVE messages affect the installed reservation state. Unlike
NOTIFY, QUERY and RESPONSE messages, the order in which RESERVE
messages are received influences the eventual reservation state that
will be stored at a QNE. Therefore, a QNE may need to detect
re-ordered or duplicated RESERVE messages.

Detection of RESERVE message re-ordering or duplication is supported
by the Reservation Sequence Number (RSN). The RSN is a counter,
locally chosen to be unique (on a hop-by-hop basis) within a session.
The RSN has local significance only, i.e. between QNEs. Attempting
to make an identifier that was unique in the context of a SESSION_ID
but the same along the complete path would be very hard. Since
RESERVE messages can be sent by any node on the path that maintains
reservation state (e.g. for path repair) we would have the difficult
task of attempting to keep the identifier synchronized along the
whole path. Since message ordering only ever matters between a pair
of peer QNEs, this means that we can make the Reservation Sequence
Number unique just between a pair of neighboring stateful QNEs. By
managing the sequence numbers in this manner, the source of the
RESERVE does not need to determine how the next NSLP node will
process the message.

The RSN refers to a particular instance of the RESERVE state. This
allows explicit acknowledgment of state installation actions (by
including the RSN in a RESPONSE). It also allows an abbreviated form
of refreshing RESERVE message ("summary refresh"). In this case, the
refreshing RESERVE references the reservation using the RSN (and the
SESSION_ID), rather than including the full reservation specification
(including QSPEC, ...). Note that summary refreshes require an
explicit acknowledgment of state installation to ensure that the RSN
reference will be understood. It is up to a QNE that receives a
RESPONSE_REQUEST to decide whether it wants to accept summary
refreshes and provide this explicit acknowledgment.

4.2.2 Requesting responses

Some QNEs may require explicit responses to messages they send. A
QNE which sends a QUERY message (Section 4.1), for instance, will
require a response with the requested information to be sent to it.
A QNE which sends a RESERVE message may want explicit confirmation
that the requested reservation state was installed.

A QNE that desires an explicit response includes a RESPONSE_REQUEST
object in its message. RESPONSE_REQUEST objects are used in RESERVE
and QUERY messages. The RESPONSE_REQUEST object may be used in
combination with message scoping (Section 4.2.3) to influence which
QNE will respond.

A message contains at most one RESPONSE_REQUEST object. The
RESPONSE_REQUEST object contains Request Identification Information
(RII) that is unique within a session and different for each message,
in order to allow responses to be matched back to requests (without
incorrectly matching at other nodes). Downstream nodes that desire
responses may keep track of this RII to identify the RESPONSE when it
passes back through them.

A message containing a RESPONSE_REQUEST object causes a RESPONSE
message to be sent back. The RESPONSE message contains the original
RESPONSE_REQUEST object and may be scoped, e.g. using the RII
(Section 4.2.3), to influence which (upstream) QNEs will receive the
RESPONSE.

4.2.3 Message scoping

For some messages, QNEs may want to restrict message propagation.
For a RESERVE message, this may be the case when state installation
is required on part of the path towards the destination only. For a
QUERY message, it allows limiting the nodes that can respond to the
QUERY. For a RESPONSE message, it allows limiting the nodes that
receive the RESPONSE.

Message scoping is supported by a SCOPING object. Different scopes
are supported. By default, no SCOPING object is present which
indicates that the scope is either "whole path" or limited by
configuration (and therefore not signalled). Other supported scopes
are "single hop" and "back to me". The latter is supported by
copying the RII from the RESPONSE_REQUEST object into the SCOPING
object that is put in the RESPONSE message, so that its forwarding
can be terminated by the node that requested the RESPONSE.

It is currently

This specification does not support an open issue whether explicit notion of a "region" should region
scope or "to the CRN". If needed, this can be supported easily proposed as an
extension later on.

4.2.4 State handling

The default behaviour of a separate scope QNE that receives a RESERVE with a
SESSION_ID for which it already has state installed but with a
different flow ID is to replace the existing reservation (and tear
down the reservation on the old branch if the RESERVE is received
with a different SII).

In some cases, this may not be the desired behaviour. In that case,
the QNI or whether its application a QNE may set the NO_REPLACE flag in the common header to
indicate that the new session does not replace the existing one. A
QNE that receives a RESERVE with the NO_REPLACE flag set but with the
same SII will update the flow ID and indicate NO_REPLACE to the RMF
(where it will be used for the resource handling). If the SII is sufficiently
supported
different, this means that the QNE is a merge point. In that case,
the NO_REPLACE also indicates that a tearing RESERVE SHOULD NOT be
sent on the old branch.

At a QNE, resource handling is performed by configuration and/or aggregation.

4.2.4 the RMF. For sessions
with the NO_REPLACE flag set, we assume that the QoS-model specific
control information includes directions to deal with resource
sharing. This may include, adding the reservations, or taking the
maximum of the two or more complex mathematical operations.

This resource handling mechanism in the QoS model is also applicable
to sessions with different SESSION_ID but related through the
BOUND_SESSION_ID object. Session replacement is not an issue here,
but the QoS model may specify whether to let the sessions that are
bound together share resources on common links or not.

Finally, it is possible that a RESERVE is received with no QSPEC at
all. This is the case of a summary refresh. In this case, rather
than sending a refreshing RESERVE with the full QSPEC, only the
SESSION_ID and the SII are sent to refresh the reservation. Note
that this mechanism just reduces the message size (and probably eases
processing). One RESERVE per session is still needed. The
combination of different SESSION_IDs (and SIIs) in the same
refreshing RESERVE message is currently an open issue.

4.2.5 State timers

The NSIS protocol suite takes a soft-state approach to state
management. This means that reservation state in QNEs must be
periodically refreshed. The frequency with which state installation
is refreshed is expressed in the REFRESH_PERIOD object. This object
contains a value in seconds indicating how long the state that is
signalled for remains valid. Maintaining the reservation beyond this
lifetime can be done by sending a ("refreshing") RESERVE message.

The REFRESH_PERIOD has local significance only. At the expiration of
a "refresh timeout" period, each QNE independently examines its state
and sends a refreshing RESERVE message to the next QNE peer where it
is absorbed. This peer-to-peer refreshing (as opposed to the QNI
initiating a refresh which travels all the way to the QNR) allows
QNEs to choose refresh intervals as appropriate for their
environment. For example, it is conceivable that refreshing
intervals in the backbone, where reservations are relatively stable,
are much larger than in an access network. The "refresh timeout" is
calculated within the QNE and is not part of the protocol; however,
it must be chosen to be compatible with the reservation lifetime as
expressed by the REFRESH_PERIOD, and an assessment of the reliability
of message delivery.

The details of timer management and timer changes (slew handling and
so on) are given in Section 5.

4.2.5

4.2.6 Session binding

Some QNEs may need to have knowledge of session binding. With session

Session binding we mean that is defined as the enforcement of a relation exists between signalled
different QoS NSLP sessions (i.e. signalling flows with potentially different SESSION_IDs and/or flow IDs. The
SESSION_ID is (SID) as defined in [4] This situation [3]).

A relation between two sessions is indicated by including the
BOUND_SESSION_ID object in the messages. A session with SID_A (the
binding session) can occur express its relation to another session with
SID_B (the bound session) by including a BOUND_SESSION_ID object
containing SID_B in case of
layering its messages. Note that the session with SID_B
may or aggregation may not carry a BOUND_SESSION_ID object containing SID_A.

Three examples where multiple reservations session binding can be used for aggregate
reservation, bi-directional reservation and fate sharing are aggregated
together (and the
described below.

Aggregated sessions may have a different flow ID changes) or when some local properties
(e.g. connection mode) for from the session change.

Layering or aggregation may cause loss of information. end-to-end
message. If the edge QNEs of the aggregation domain want to maintain
some end-to-end properties, they may establish a peering relation by
sending the end-to-end message transparantly over the domain.
Updating the end-to-end properties in this message may require some
knowledge of the aggregated session (e.g. for updating delay
values). For this purpose, a the end-to-end session (e.g., end to end session), may contain contains a
BOUND_SESSION_ID (the carrying the SESSION_ID of another session (e.g., the aggregate one) session.

Including the BOUND_SESSION_ID object in addition to its own SESSION_ID a session indicates a
dependency relation. By default, a session that is bound to indicate another
session
binding. This with the BOUND_SESSION_ID object shares fate with it. This
means that if a bound session is torn down, every binding session
should be torn down as well. The reverse is not true. If a binding
session is called torn down, the bound session and other binding object. sessions
remain. A QNE that wants to prevent fate sharing sets a flag in the
common header (NO_FATE_SHARING).

Bi-directional reservations are a special case of fate sharing. In
this case, a reservation in one direction only makes sense if the
reservation in the reverse direction is also up. In some cases
bi-directional reservations also allow local optimizations.
Therefore, when a reverse reservation is set up, it should carry a
BOUND_SESSION_ID containing the SESSION_ID of the forward direction.
This SESSION_ID is copied from the sender-initiated (forward) RESERVE
or from the QUERY message triggering the sender-initiated (reverse)
RESERVE message. Alternatively, it can be inserted by the QNE that
sets up both the sender-initiated RESERVE and the receiver-initiated
RESERVE.

4.3 Layering

The QoS NSLP supports layered reservations. Layered reservations may
occur when certain parts of the network (domains) implement one or
more local QoS models, or when they locally apply specific control
plane characteristics (e.g. datagram mode instead of connection
mode). They may also occur when several per-flow reservations are
locally combined into an aggregate reservation.

4.3.1 Local QoS models

Parameters of the QoS model that is being signalled for are carried
in the QSPEC object. A domain may have local policies regarding QoS
model implementation, i.e. it may map incoming traffic to its own
locally defined QoS models. The QoS NSLP supports this by allowing
QSPEC objects to be stacked.

When a domain wants to apply a certain QoS model to an incoming
per-flow reservation request, each edge of the domain is configured
to map the incoming QSPEC object to a local QSPEC object and push
that object onto the stack of QSPEC objects (typically immediately
following the Common Control Information, i.e. the first QSPEC that
is found in the message). QNEs inside the domain look at the top of
the QSPEC object stack to determine which QoS model to apply for the
reservation.

The position of the local QSPEC object in the stack implies a
tradeoff between the speed with which incoming messages can be
processed and the time it takes to construct the outgoing message (if
any). By mandating the locally valid object to be on top of the
stack we value ease of processing over ease of message construction.

A QNE that knows it is the last QNE to understand a local QSPEC
object (e.g. by configuration of the egress QNEs of a domain) SHOULD
remove the topmost QSPEC object from the stack. It SHOULD update the
underlying QoS model parameters if needed.

A QNE that receives a message with a QSPEC object stack of which the
topmost object is not understood SHOULD MUST NOT forward the message and
MUST send an error indication to its upstream neighbour. It is currently an open issue whether this
QNE MAY search MUST NOT
attempt local recovery by inspecting the stack for a QSPEC object it understands to recover
from this situation. It is also an open issue if such a message can
be forwarded and if and how the QSPEC object stack should be updated.

4.3.2 Local control plane properties

The way signalling messages are handled
understands.

4.3.2 Local control plane properties

The way signalling messages are handled is mainly determined by the
parameters that are sent over the NTLP-NSLP GIMPS-NSLP API and by the Common
Control Information. A domain may have a policy to implement local
control plane behaviour. It may, for instance, elect to use datagram
mode an
unreliable transport locally in the domain while still keeping e2e
end-to-end reliability intact.

The QoS NSLP supports this situation by allowing two sessions to be
set up for the same reservation. The local session has the desired
local control plane properties and is interpreted in internal QNEs.
This solution poses two requirements: the end-to-end session must be
able to bypass intermediate nodes and the egress QNE needs to bind
both sessions together.

The local session and the end-to-end session are bound at the egress
QNE by means of the BOUND_SESSION_ID object. One approach could be
that the end-to-end session carries the SESSION_ID of the local
session in its session binding object. Another approach could be
that the local session carries the SESSION_ID of the end-to-end
session in its BOUND_SESSION_ID object. This allows the QNE that
performs session binding to maintain end-to-end connection mode.

4.3.3 Aggregate reservations

For scalability reasons, a domain MAY may want to combine two or more
end-to-end reservations into a single local aggregate reservation.
The domain over which the aggregation is done is limited by
configuration.

The essential difference with the layering approaches described in
Section 4.3.1 and Section 4.3.2 is that the aggregate reservation
needs a FlowID that describes all traffic carried in the aggregate
(e.g. a DSCP in case of IntServ over DiffServ).

The need for a different FlowID mandates the use of two different
sessions, similar to Section 4.3.2 and to the RSVP aggregation
solution (reference to 3175). [10]. In addition to the different FlowID, the aggregate
session may specify a local QoS model and local control plane
parameters as explained above.

The aggregate reservation may or may not change source and
destination IP addresses, i.e. either the end-to-end adresses may be
used (if possible) or the IP address of ingress and egress of the
domain may be used as source and destination IP address. In some
cases, the latter option may cause data plane divergence between both
sessions. RSVP solves this by using tunnelling between the edges of
the domain.

In any case, session binding and a solution for intermediate node
bypass (as explained before) are required in this case as well.

4.4 Extensibility

The QoS NSLP specification foresees future specification of new error
codes and new Common Control Information objects. Specification of
new messages is not foreseen but not explicitly precluded.

Specification of new error codes and Common Control Information
objects is subject to IANA approval and assignment of ClassNum and
CType. ClassNum and CType of currently existing objects and error
codes are described in Section 6. New Common Control Information
objects need to specify whether they are mandatory or optional to
implement. Mandatory CCI that is not understood by a QNE needs to
generate an error. Optional CCI that is not understood by a QNE
needs to be passed transparantly.

The QoS NSLP specification allows future QoS model specific
extensions, including the definition of new QoS models, the
specification of new objects for existing QoS models, the
specification of new processing rules for new or existing objects and
the specification of new QoS model specific error codes.

Different types of QoS models are foreseen: standardized QoS models,
well-known QoS models and QoS models for private use. We assume the
IANA registry of QoS models to distinguish between those. Apart from
the QoS model ID, all QoS model specific extensions are opaque to the
QoS NSLP (and have no impact on its IANA considerations section).

4.5 Priority

This specification acknowledges the fact that in some situations,
some messages or some reservations may be more important than others
and therefore foresees mechanisms to give these messages or
reservations priority.

Priority of certain signalling messages over others may be required
in mobile scenarios when a message loss during call set-up is less
harmful then during handover. This situation only occurs when the GIMPS
or QoS NSLP processing is the congested part or scarce resource.
This specification requests the NTLP GIMPS design to foresee a mechanism to
support a number of levels of message priority that can be requested
over the NSLP-NTLP NSLP-GIMPS API.

Priority of certain reservations over others may be required when QoS
resources are oversubscribed. In that case, existing reservations
may be preempted in other to make room for new higher-priority
reservations. A typical approach to deal with priority and
preemption is through the specification of a setup priority and
holding priority for each reservation. The resource management
function at each QNE then keeps track of the resource consumption at
each priority level. Reservations are established when resource at
their setup priority level are still available. They may cause
preemption of reservations with a lower holding priority than their
setup priority.

Support of reservation priority is a QoS model specific issue and
therefore outside the scope of this specification. However, the
concepts of setup and holding priority are widely accepted and we
expect the specification of a Priority object in the QSPEC template
to be useful for a wide range of QoS models.

4.6 Rerouting

The QoS NSLP needs to adapt to route changes in the data path. This
assumes the capability to detect rerouting events, perform QoS
reservation on the new path and optionally tear down reservations on
the old path.

Rerouting detection can be performed at three levels. First, routing
modules may detect route changes through their interaction with
routing protocols. Certain QNEs or NTLP GIMPS implementations may
interact with local routing module to receive quick notification of
route changes. This is largely implementation-specific and outside
of the scope of NSIS. Second, route changes may be detected at the NTLP GIMPS
layer. This specification requests the NTLP GIMPS design to foresee
notification of this information over the API. This is outside the
scope of the QoS NSLP specification. Third, rerouting may be
detected at the NSLP layer. A QoS NSLP node is able to detect
changes in its QoS NSLP peers by keeping track of a Source
Identification Information (SII) object that is similar in nature to
the RSVP_HOP object described in [6]. [5]. When a RESERVE message with an
existing SESSION_ID and a different SII is received, the QNE knows
its upstream peer has changed.

Reservation on the new path automatically happens when a refreshing
RESERVE message arrives at the QNE where the old and the new path
diverge. Rapid recovery at the NSLP layer therefore requires short
refresh periods. Detection before the next RESERVE message arrives
is only possible at the IP layer or through monitoring of the NTLP GIMPS
peering relations (e.g. by TTL counting the number of NTLP GIMPS hops
between NSLP peers or the observing changes in the outgoing interface
towards the NTLP GIMPS peer). These mechanisms are outside the scope of this
specification.

When the QoS NSLP is aware of the route change, it needs to set up
the reservation on the new path. This is done by sending a RESERVE
message with RSN+2. RSN+1. On links that are common to the old and the new
path, this RESERVE message is interpreted as a refreshing RESERVE.
On new links, it creates the reservation.

After the reservation on the new path is set up, the branching node
or the merging node may want to tear down the reservation on the old
path (faster than what would result from normal soft-state time-out).
This functionality is supported by keeping track of the old SII.
This specification requests the NTLP GIMPS design to provide support for an
SII that is interpreted as a random identifier at the QoS NSLP but
that allows, when passed over the API, to forward QoS NSLP messages
to the QNE identified by that SII. Then,

A QNE that has detected the route change via the SII change sends a
RESERVE message with the TEAR flag set (tearing RESERVE) set. A QNE that is notified of the route
change in another way and RSN+1 can be sent over want to tear down the old branch
of needs to
send the path. Setting RESERVE on the new path with a RESPONSE_REQUEST. When it
receives the RESPONSE message back, it can check whether its peer has
effectively changed and send a RESERVE with the RSN+1 ensures TEAR flag set if it
has. Otherwise, teardown is not needed. A QNE that is unable to
perform RESPONSE_REQUEST or does not receive a RESPONSE needs to rely
on sof-state timeout on the old branch.

A QNI or a branch node may wish to keep the reservation will not on the old
branch. This could for instance be torn down if the neighbouring QNE case when a mobile node has not,
experienced a mobility event and wishes to keep reservation to its
old attachment point in fact, changed. case it moves back there. In that case, it
sets the NO_REPLACE flag in the common header.

4.7 State storage

For each flow, the QoS NSLP a QNE stores QoS reservation state. This state
includes QoS (QoS model specific specific) reservation state
which is different for each QoS model and QoS NSLP operation state
which includes non-persistent state (e.g. the API parameters while a
QNE is processing a message) and persistent state which is kept as
long as the session is active.

The persistent QoS NSLP state is conceptually organised in a table
with the following structure. The primary key (index) for the table
is the Session ID:
SESSION_ID

A large identifier provided by the NTLP. GIMPS or set locally.

The state information for a given key includes:
Flow ID

Copied from the NTLP. GIMPS. Several entries are possible in case of
mobility events.

QoS model ID

8 bit identification of the QoS model.

SII for each upstream and downstream peer

The SII is a 128 bit large identifier (minimum 128 bits) generated by the NTLP
GIMPS and passed over the API.

RSN from each upstream peer

The RSN is a 32 bit counter.

Current own RSN

A 32 bit random number.

List of RII for outstanding responses with processing information

the RII is a 32 bit number.

State lifetime

The state lifetime indicates how long the state that is being
signalled for remains valid.

BOUND_SESSION_ID

The BOUND_SESSION_ID is a 128 bit random number.

Adding the state requirements of all these items gives an upper bound
on the state to be kept by a QNE. The need to keep state depends on
the desired functionality at the NSLP layer.

4.8 Authentication and authorization

QoS NSLP requests allow particular user(s) to obtain preferential
access to network resources. To prevent abuse, some form of an
access control (or also (also known as policy based admission control) will
generally be required on users who make reservations. Typically,
such authorization is expected to make use of an AAA service external
to the node itself. In any case, cryptographic user identification
and selective admission will generally be needed when a reservation
is requested.

The QoS NSLP request is handled by a local 'resource management'
function, which coordinates the activities required to grant and
configure the resource. The grant processing involves two local
decision modules, 'policy control' and 'admission control'. Policy
control determines whether the user is sufficiently authorized to
make the reservation. Admission control determines whether the node
has sufficient available resources to offer the requested QoS.

4.8.1 Policy Ignorant Nodes

It is generally assumed that policy enforcement is likely to
concentrate on border nodes between autonomous systems. administrative domains. Figure 9
15 below illustrates a simple autonomous administrative domain with:
o two boundary nodes (A, C), which represent QNEs authorized by AAA
entities.
o A core node (B) represents an Policy Ignorant QN QNE (PIN) with
capabilities limited to default admission control handling.

Authorizing Entity 1 Authorizing Entity 2
| |
| |
+---+ +---+ +---+
| A +---------+ B +---------+ C |
+---+ +---+ +---+
QN1
QNE1 PIN QN2 QNE2

Figure 9: Autonomous 15: Administrative Domain scenario

Here, policy objects transmitted across the domain traverse an
intermediate PIN node (B) that is allowed to process QoS NSLP message
but considered non-trusted for handling policy information.

4.8.2 Policy Data

The input to policy control is referred to as "Policy data", which
QoS NSLP carries in the Policy object. Policy data may include
credentials identifying entities and traits depending on the
authorization model in use (2-party, 3-party, token-based 3-party).
There are no requirements for all nodes to process this object.
Policy data itself is opaque to NSIS, the QoS NSLP, which simply passes it
to policy control when required. The policy data is independent from
the QoS model in use.

Policy control depends on successful user authentication and
authorization of a QoS NSLP reservation request. The authorization
decision might be valid for a certain amount of time or even for the
entire lifetime of the session. It is a decision of the involved
party to trigger a re-authorization procedure. This feature is
supported by the Policy Refresh Timer (PRT) option of the Policy
object.

Policy objects are carried by QoS NSLP messages and contain policy
information. All policy-capable nodes (at any location in the
network) can generate, modify, or remove policy objects, even when
senders or receivers do not provide, and may not even be aware of
policy data objects.

The exchange of Policy objects between policy-capable QNEs along the
data path, supports the generation of consistent end-to-end policies.

Furthermore, such policies can be successfully deployed across
multiple administrative domains when border nodes manipulate and
translate Policy objects according to established sets of bilateral
agreements.

5. QoS-NSLP Functional specification

5.1 QoS-NSLP Message and Object Formats

A QoS-NSLP message consists of a common header, followed by a body
consisting of a variable number of variable-length, typed "objects".
The following subsections define the formats of the common header,
the standard object header, and each of the QoS-NSLP message types.

For each QoS-NSLP message type, there is a set of rules for the
permissible choice of object types. These rules are specified using
the Augmented Backus-Naur Form (BNF) augmented (see [2]). with square brackets
surrounding optional sub-sequences. specified in [2]. The BNF
implies an order for the objects in a message. However, in many (but
not all) cases, object order makes no logical difference. An
implementation should create messages with the objects in the order
shown here, but accept the objects in any permissible order.

5.1.1 Common header

0 1
+---------------------------+---------------------------+ 2 3
+-------------+-------------+-------------+-------------+
| Msg Message Type | Flags |
+---------------------------+---------------------------+
+-------------+-------------+-------------+-------------+

The fields in the common header are as follows:

Msg Type: 8 16 bits
1 = RESERVE
2 = QUERY
3 = RESPONSE
4 = NOTIFY
Flags: 8 16 bits

1 = TEAR flag

2 = BIDIRECTIONAL flag
Other
The set of appropriate flags have to depend on the particular message
being processed. Any bit not defined as a flag for a
particular message MUST be defined. set to zero on sending and MUST
ignored on receiving.

5.1.2 Object Formats

Every object consists of one or more 32-bit words with a one-word
header, with the following format:

0 1 2 3
+-------------+-------------+-------------+-------------+
| Length (bytes) | Class-Num | C-Type |
+-------------+-------------+-------------+-------------+
| |
// (Object contents) //
| |
+-------------+-------------+-------------+-------------+

An object header has the following fields:

Length:
A 16-bit field containing the total object length in bytes.
Must always be a multiple of 4, and at least 4.
Class-Num:
Identifies the object class; values of this field are defined
in Appendix A. Each object class has a name, which is always
capitalized in this document. An QoS-NSLP implementation must
recognize the following classes:
RESPONSE_REQUEST:
Contains the request for the generation of a response
message and the Request Identification Information (RII).
RSN:
The Reservation Sequence Number (RSN) contains an
incrementing sequence number that indicates the order in
which state modifying actions are performed by a QNE.
The RSN has local significance only, i.e. between a pair
of neighbouring stateful QNEs. RSN is a common control
information object.

REFRESH_PERIOD
REFRESH_PERIOD:
Contains the value for the refresh period R used by the
creator of the message. Required in every RESERVE
message. REFRESH_PERIOD is a common control information
object.

SESSION_ID
BOUND_SESSION_ID:
It represents the SESSION_ID Session ID as specified in [3] [15] of the
session that must be bound to the session associated to
the message carrying this object.

SCOPING
SCOPING:
contains information that limits the scope of the message
carrying this object. When no SCOPING object is
available in a message it means that its scoping is
either the whole path or it is defined by configuration.
SCOPING is a common control information object.

ERROR_SPEC
ERROR_SPEC:
Contains an error code and can be carried by a Response
or a NOTIFY message. ERROR_SPEC is a common control
information object.

POLICY_DATA

POLICY_DATA:
Carries authentication, authorization and accounting
information.

QSPEC
QSPEC:
Carries the information that is QoS model specific. This
information consists of the QoS model specific control
information and the QoS specification parameters.
C-Type:
Object type, unique within Class-Num. Values are defined in
Appendix A.

The maximum object content length is 65528 bytes. The Class- Num Class-Num and
C-Type fields may be used together as a 16-bit number to define a
unique type for each object.

The high-order two bits of the Class-Num are used to determine what
action a node should take if it does not recognize the Class-Num of
an object;

5.1.3 RESERVE Messages object. The RESERVE first two bits of the Class-Num take one of the
following three values:
00 - Abort processing and send error back
01 - Ignore object, and do not forward it in onward message is used to manipulate QoS reservation state
10 - Ignore object, but forward it in
QNEs. A RESERVE onward message may create, refresh, modify or remove such
state.

5.2 General Processing Rules

5.2.1 State Manipulation

The format processing of a RESERVE message and its component objects involves
manipulating the QoS NSLP and reservation state of a QNE.

The state used by the QoS NSLP is as follows:

RESERVE = COMMON_HEADER

RSN [ SCOPING ] [ RESPONSE_REQUEST ]

REFRESH_PERIOD [ BOUND_SESSION_ID ]

POLICY_DATA QSPEC [ *QSpec ] listed in Section 4.7.

5.2.2 Message Forwarding

QoS NSLP messages are sent peer-to-peer along the path. The QSPEC object(s) must occur at QoS NSLP
does not have the end concept of a message being sent along the message. There entire
path. Instead, messages are
no other requirements received by a QNE, which may then send
another message (which may be identical to the received message, or
contain some subset of objects from it) to continue in the same
direction (i.e. towards NI or NR) as the message received.

The decision on transmission order, although whether to generate a message to forward may be
affected by the above order presence of a SCOPING object.

5.2.3 Standard Message Processing Rules

If a mandatory object is recommended.

The SESSION_ID missing from a message then the receiving
QNE MUST NOT propagate the message any further. It MUST construct an
RESPONSE message indicating the error condition and send it back to
the peer QNE that sent the message.

If a message contains an object must be included of an unrecognised type, then the
behaviour depends on the first two bits in the type value.

5.3 Object Processing

5.3.1 Reservation Sequence Number

A QNE's own RSN is a sequence number which applies to a particular
NSIS signalling session (i.e. with a particular GIMPS Session ID).
It MUST be incremented for each new RESERVE message only if where the
reservation for the session associated to this message changes. Once the RSN has reached its
maximum value, the next value it takes is zero.

When receiving a RESERVE message a QNE uses the RSN given in the
message to be bound determine whether the state being requested is different
to another
session. The content of that already stored. If the SESSION_ID object represents RSN is the
SESSION_ID of same as for the session that must current
reservation the current state MUST be bound to refreshed. If the session
associated RSN is
greater than the current stored value, the current reservation MUST
be modified appropriately (provided that admission control and policy
control succeed), and the stored RSN value updated to that for the
new reservation. If the RSN is less than the current value, then it
indicates an out-of-order message and the RESERVE message carrying MUST be
discarded.

If the QNE does not store per-session state (and so not keep any
previous RSN values) then it MAY ignore the value of the RSN. It
MUST also copy the same RSN into the RESERVE message (if any) it
sends as a consequence of receiving this one.

5.3.2 Response Request

A QNE sending some types of message may require a response to be
sent. It does so by including a RESPONSE_REQUEST object. The binding

When creating a RESPONSE_REQUEST object the sender MUST select the
value for the RII such that it is probabilistically unique within the
given session.

A number of these two sessions choices are available when implementing this.
Possibilities might include using a totally random value, or a node
identifier together with a counter. If the value is only possible selected by
another QNE then RESPONSE messages may be incorrectly terminated, and
not passed back to the node that requested them.

When sending a RESPONSE_REQUEST the sending node MUST remember the
value used in stateful QNEs.

The RESERVE message opaquely must transport the RII to match back any RESPONSE received. It SHOULD
use a QSPEC object,
describing timer to identify situations where it has taken too long to
receive the desired service level and expected RESPONSE. If the timer expires without
receiving a POLICY_DATA object,
authorizing RESPONSE it MAY perform a retransmission.

When receiving a message containing a RESPONSE_REQUEST object the requestor
node MUST send a RESPONSE if either
o The message contained a SCOPING object with a value of 'next hop',
or
o This QNE is the service. Based last one on configured local
policy, a node may ignore the content of path for the POLICY_DATA object. given session.
and the QNE keeps per-session state for the given session.

5.3.3 Bound Session ID

As shown in the examples in Section 3.5, the QoS NSLP can relate
multiple sessions together. It does this by including the Session ID
from one session in a BOUND_SESSION_ID object in messages in another
session.

5.3.4 Refresh Period

Refresh timer management values are carried by the TIMER_VALUES REFRESH_PERIOD
object. The details of timer management and timer changes (slew
handling and so on) are identical to the ones specified in Section
3.7 of [6]. [5]. There are two time parameters relevant to each QoS-NSLP
state in a node: the refresh period R between generation of
successive refreshes for the state by the neighbor node, and the
local state's lifetime L. Each RESERVE message may contain a
REFRESH_PERIOD object specifying the R value that was used to
generate this (refresh) message. This R value is then used to
determine the value for L when the state is received and stored. The
values for R and L may vary from peer to peer. This peer-to-peer
refreshing (as opposed to the QNI initiating a refresh which travels
all the way to the QNR) allows QNEs to choose refresh intervals as
appropriate for their environment. For example, it is conceivable
that refreshing intervals in the backbone, where reservations are
relatively stable, are much larger than in an access network.

In more detail:
1. Floyd and Jacobson [25] [26] have shown that periodic messages
generated by independent network nodes can become synchronized.
This can lead to disruption in network services as the periodic
messages contend with other network traffic for link and
forwarding resources. Since QoS-NSLP sends periodic refresh
messages, it must avoid message synchronization and ensure that
any synchronization that may occur is not stable. For this
reason, it is recommended that the the refresh timer should be
randomly set to a value in the range [0.5R, 1.5R].

2. To avoid premature loss of state, L must satisfy L >= (K +
0.5)*1.5*R, where K is a small integer. Then in the worst case,
K-1 successive messages may be lost without state being deleted.
To compute a lifetime L for a collection of state with different R
values R0, R1, ..., replace R by max(Ri).
Currently K = 3 is suggested as the default. However, it may be
necessary to set a larger K value for hops with high loss rate. K
may be set either by manual configuration per interface, or by
some adaptive technique that has not yet been specified.
3. Each RESERVE message carries a REFRESH_PERIOD object
containing the refresh time R used to generate refreshes. The
recipient node uses this R to determine the lifetime L of the
stored state created or refreshed by the message.
4. The refresh time R is chosen locally by each node. If the
node does not implement local repair of reservations disrupted by
route changes, a smaller R speeds up adaptation to routing
changes, while increasing the QOS-NSLP overhead. With local
repair, a router can be more relaxed about R since the periodic
refresh becomes only a backstop robustness mechanism. A node may
therefore adjust the effective R dynamically to control the amount
of overhead due to refresh messages.
The current suggested default for R is 30 seconds. However, the
default value Rdef should be configurable per interface.
5. When R is changed dynamically, there is a limit on how fast it
may increase. Specifically, the ratio of two successive values
R2/R1 must not exceed 1 + Slew.Max.
Currently, Slew.Max is 0.30. With K = 3, one packet may be lost
without state timeout while R is increasing 30 percent per refresh
cycle.
6. To improve robustness, a node may temporarily send refreshes
more often than R after a state change (including initial state
establishment).
7. The values of Rdef, K, and Slew.Max used in an implementation
should be easily modifiable per interface, as experience may lead
to different values. The possibility of dynamically adapting K
and/or Slew.Max in response to measured loss rates is for future
study.

Each

5.3.5 Scoping

When sending some types of message, the node may insert wish to limit the
distance that the message should travel along the path (the default
being the whole path). To do so a local QSPEC node MUST include a SCOPING
object.

When receiving a message, before sending the message further, the QNE
MUST inspect any scoping object provided to determine if it has a way of
scoping this information (e.g. at reached the boundary
end of a domain or by
using the SCOPING object).

In some cases, scoped region. If so, it MUST NOT pass it further, and
MUST generate a QNE needs to be able to distinguish between newly
created, modified state or refreshed state based on RESPONSE if a RESPONSE_REQUEST object is present.

5.3.6 Error Spec

The ERROR_SPEC object contains a value which indicates the RESERVE
message alone (not in combination with state information obtained
from previous messages). Therefore, one or more additional flags condition
that
provide this differentiation may be needed. The specifictaion of
these flags occurred. Error values are to be specified for various
conditions, such as:
o OK
o Message type not supported
o Object type not supported
o Insufficient resources
o Authentication failure
o Authorisation denied
o QoS model specific. Therefore, specific condition occurred
o ...

5.3.7 Policy Data

POLICY_DATA objects may contain various items to authenticate the contents
user and
encoding rules for this object allow the reservation to be authorised. Some possible
contents are given in those Appendix A.7, and some issues are also
discussed in Section 3.3.

5.3.8 QSpec

The contents of the QSpec depends on the QoS model
specifications.

In order to clearly distinguish between a RESERVE message being used. It
may be that sets parts of the reserved resources to zero and a QSpec are standardised across multiple QoS
models. This topic is currently the topic of further study.

5.4 Message Processing Rules

5.4.1 RESERVE Messages

The RESERVE message that tears down
QoS-NSLP state completely, a TEAR flag is foreseen that is carried used to manipulate QoS reservation state in
the common header. Note that the potential initiation
QNEs. A RESERVE message may create, refresh, modify or remove such
state.

The format of (reverse
path) state removal at the NTLP is a separate issue. This will be
signaled over the API between NTLP and QoS-NSLP. RESERVE messages message is as follows:

RESERVE = COMMON_HEADER
RSN [ SCOPING ] [ RESPONSE_REQUEST ]
[ REFRESH_PERIOD ] [ BOUND_SESSION_ID ]
[ POLICY_DATA ] [ *QSPEC ]

If any QSPEC objects are sent peer-to-peer. This means that a QNE
considers its adjacent upstream or downstream peer to be present, they MUST occur at the source end of the RESERVE
message. Note that two nodes that There are adjacent at no other requirements on transmission order,
although the
QoS-NSLP layer may above order is recommended.

Three flags are defined for use in fact be separated by several NTLP hops. A QoS- the common header with the RESERVE
message. These are:
TEAR - when set, indicates that reservation state and QoS NSLP node may want to
operation state should be able to detect changes in its QoS-NSLP
peers, or send a message torn down. This is indicated to an explicitly identified node, e.g. for
tearing down the
RMF.
NO_REPLACE - when set, indicates that a reservation on RESERVE with different
Flow Routing Information (FRI) does not replace an existing one,
so the old path. This functionality can one should not be
provided by keeping track of a Source Identification Information
(SII) object torn down immediately
NO_FATE_SHARING - when set, indicates that is similar sessions in nature to the RSVP_HOP object
described in [6]. We assume such an SII (section 7.2) to bundle
should not share fate with one another

An RSN MUST be available
as a service from the NTLP.

The present.

RESERVE message is idempotent; messages MUST only be sent towards the resultant effect is NR.

If the same
whether QNE sending a RESERVE message is received once or many times. In addition, wishes to use the reduced
overhead refresh mechanism described in Section 4.2.1, then it SHOULD
include a RESPONSE_REQUEST in the
ordering of RESERVE messages matters - an old RESERVE message. It MUST NOT send
a reduced overhead refresh message does
not replace (i.e. a newer one. Both of these features are required RESERVE with a
non-incremented RSN and no QSPEC) unless it has received a RESPONSE
message for
protocol robustness - messages may be re-ordered on route (e.g.
because that RESERVE message.

If the session of mobility, or at intermediate NTLP nodes) or spuriously
retransmitted.

In order this message is bound to tackle these issues, another session, then the
RESERVE message contains a
Reservation Sequence Number (RSN) object. An RSN is an incrementing
sequence number that indicates MUST include the order SESSION_ID of that other session in which state modifying
actions are performed by
a QNE. The RSN has BOUND_SESSION_ID object.

Before admitting a reservation a QNE MUST determine whether the
request is authorized. It SHOULD exercise its local significance only,
i.e. between QNEs. Attempting to make an identifier that was unique policy in
conjunction with the context of POLICY_DATA object to do this.

When a QNE receives a session identifier but the same along the
complete path would be very hard. Since RESERVE messages can be sent
by any node on message, its processing may involve
sending out a RESERVE message. When doing so, the path that maintains reservation state (e.g. for
path repair) we would have QNE may insert or
remove 'local' QSPEC objects from the difficult task top of attempting to keep the identifier synchronized along stack. If there are
one or more QSPECs in the whole path. Since message
ordering only ever matters between a pair of peer QNEs, this means
that we can make received RESERVE message, the Reservation Sequence Number unique just between
a pair of neighboring stateful QNEs. Note that an alternative might last QSPEC
MUST NOT be for the NTLP to guarantee in-order delivery between removed when sending on the NSLP
peers.

A Flow identifier groups together state items for RESERVE message.

When a single flow. The
RSN is one of these state items, and reservation is used to identify reordering
of messages and to allow no longer required the use of partial refresh messages. The
state items for QNI SHOULD send a number of flows can be linked together and
identified as part of
RESERVE message with the TEAR bit set in the header. On receiving
such a single RESERVE message the QNE MUST remove any reservation using a Session
Identifier. The identifiers play complementary roles in state for
that session. It SHOULD remove the
management of QoS NSLP state. The flow identifier is carried by It MAY signal to
the NTLP and that it is augmented by additional flow identifying information
in the QSPEC, which is QoS model specific.

The sender of a RESERVE message may want to receive some confirmation
from a downstream node. In this case the RESERVE message must contain
a RESPONSE_REQUEST object. The RESPONSE_REQUEST object contains the
Request Identification Information (RII) value used to match back a
RESPONSE to a request no longer interested in a RESERVE message.

5.1.4 NTLP state for that
session.

5.4.2 QUERY Messages

The format of a QUERY message is as follows:

QUERY = COMMON_HEADER
[ SCOPING ] RESPONSE_REQUEST [ REFRESH_PERIOD RESPONSE_REQUEST ]
[ BOUND_SESSION_ID ]
[ POLICY_DATA QSPEC ] [ *QSPEC ]

The

If any QSPEC object(s) must objects are present, they MUST occur at the end of the
message. There are no other requirements on transmission order,
although the above order is recommended.

No flags are defined are defined for use with the QUERY message.

A QUERY message may MAY be scoped using the SCOPING object.

A QUERY message must MUST contain a RESPONSE_REQUEST object, that carries
the Request Identification Information (RII) that allows matching
back RESPONSE to unless the
QUERY request. It is transported unchanged along
the data path and should be being used in combination with the SCOPING
object to scope initiate reverse-path state for a
receiver-initiated reservation.

If the RESPONSE session of this message is bound to a QUERY message.

The QUERY another session, then the
RESERVE message can gather information along MUST include the data path SESSION_ID of that other session in
a
number of objects. Some of these objects may BOUND_SESSION_ID object.

On receiving a QUERY message, the QoS model specific QUERY processing
MAY cause the QSpec to be part of modified, to provide the answer to the
query. Future QoS
model. Others NSLP objects may be generic added to the QoS-NSLP protocol.

The protocol with
similar properties in this respect.

If this is last node to process this QUERY message opaquely must transport a QSPEC object, describing (either because it
is the desired service level and a POLICY_DATA object, authorizing last node on the
requestor path, or because of the service. Based on configured local policy, scoping), and a node
may ignore
RESPONSE_REQUEST object is present, then a RESPONSE message MUST be
generated and passed back along the content reverse of the POLICY_DATA object.

The QUERY message may carry path used by the REFRESH_PERIOD object. It
QUERY.

If this is
RECOMMENDED that in case of a receiver initiated reservation, the
QUERY message carries the REFRESH_PERIOD object.

The SESSION_ID last node and a RESPONSE_REQUEST object must be included in is not present
then the QUERY message only if
the session associated is a trigger to initiate a reservation in the reverse
direction. If this message has node was not expecting to perform a
receiver-initiated reservation then an error MUST be bound sent back along
the path.

When generating a QUERY to another
session. The content of send out to pass the SESSION_ID object represents query further along
the
SESSION_ID of path, the session that must be bound to QNE MUST copy the session
associated to RESPONSE_REQUEST (if present) into
the new QUERY message carrying this object. The binding of
these two sessions is only possible in stateful QNEs.

5.1.5 unchanged.

5.4.3 RESPONSE Messages

The RESPONSE message is used to provide information about the result
of a previous QoS-NSLP message, e.g. confirmation, error confirmation of a reservation
or information resulting from a query. The RESPONSE message is
impotent, it does not cause any state to be installed or modified.

The format of a RESPONSE message is as follows:

RESPONSE = COMMON_HEADER
[ RSN ] [ SCOPING / RSN ] [ ERROR_SPEC ]

QSPEC
[ *QSPEC]

The *QSPEC ]

If any QSPEC object(s) must objects are present, they MUST occur at the end of the
message. There are no other requirements on transmission order,
although the above order is recommended.

No flags are defined are defined for use with the RESPONSE message.

A QNE may want to receive a RESPONSE message to inform it that MUST be sent where the
reservation has been successfully installed. A QNE is the last node to
process a RESERVE or a QUERY message may contain a RESPONSE_REQUEST object for this purpose. Such containing a RESPONSE_REQUEST
object can be used to request an explicit
confirmation (based on scoping of the state manipulation signaled in the RESERVE
message.

The forwarding of the RESPONSE message along or QUERY, or because this is
the path does not
necessarily imply last node on the existence of NTLP reverse-path state at every
node. For example, path). In this case, the NTLP may have a mechanism to pass RESPONSE MUST contain
a message
directly from the egress to the ingress SCOPING object of a region type RII. The contents of QNEs that do
not store per-flow reverse-path state.

A RESPONSE the RESPONSE_REQUEST
object in the received RESERVE or QUERY message may MUST be scoped using the copied into
this SCOPING object. A QUERY
always causes object in the RESPONSE.

In addition, a RESPONSE to be sent. Therefore, a QUERY message will
always contain a RESPONSE_REQUEST object. A RESERVE may cause a
RESPONSE to MUST be sent if this is explicitly requested, by using a
RESPONSE_REQUEST object or when an error occurs. The RESPONSE
Identification Information (RII) included in occurs
while processing a received message message. If the received message
contains a RESPONSE_REQUEST object, then an RII SCOPING object should MUST
be included put in the SCOPING object RESPONSE, as described above. If the RESPONSE is sent
as a result of the receipt of a RESERVE message without a
RESPONSE_REQUEST object, then the RSN of the received RESERVE message
MUST be copied into the RESPONSE message.

A RESPONSE message may carry MUST contain an RSN object. The content of this ERROR_SPEC object must be identical to which indicates
the content success of a reservation installation or an error condition.
Depending on the RSN object contained
in value of the RESERVE ERROR_SPEC, the RESPONSE MAY also
contain a QSPEC object.

On receipt of a RESPONSE message containing an RII SCOPING object,
the QNE MUST attempt to match it to the outstanding response requests
for that generated this signalling session. If the match succeeds, then the
RESPONSE message. MUST NOT be forwarded further along the path. If a QNE or the QNR is unable to provide match
fails, then the requested information or
if QNE MUST attempt to forward the response is negative, RESPONSE to the next
peer QNE.

On receipt of a RESPONSE message must carry containing an
ERROR_SPEC object. RSN object, the QNE
MUST compare the RSN to that of the appropriate signalling session.
If the match succeeds then the error MUST be processed. The RESPONSE
message opaquely must transport a QSPEC object(s),
describing MUST NOT be forwarded further along the desired service level.

5.1.6 path whether or not
the match succeeds.

5.4.4 NOTIFY Messages

NOTIFY messages are used to convey information to a QNE. QNE
asynchronously. NOTIFY messages are impotent (they do not cause a
change in state directly).

NOTIFY messages differ from RESPONSE messagess in that they need not
refer to any particular state or previously received message. They
are sent asynchronously. The NOTIFY message itself does not trigger
or mandate any action in the receiving QNE.

The format of a NOTIFY message is as follows:

NOTIFY = COMMON_HEADER

[
ERROR_SPEC ] QSPEC

The [ QSPEC ]

No flags are defined are defined for use with the NOTIFY message.

A NOTIFY message MUST contain an ERROR_SPEC object must occur at indicating the end of
reason for the message. There are no
other requirements notification. Depending on transmission order, although the above order is
recommended.

The information conveyed by ERROR_SPEC value, it
MAY contain a QSpec providing additional information.

NOTIFY message may be related messages are sent asynchronously, rather than in response to error
conditions. In this case the ERROR_SPEC object must
other messages. They may be carried by the
NOTIFY message.

The NOTIFY message opaquely must transport a QSPEC object, describing
the desired service level. sent in either direction (upstream or
downstream).

6. IANA considerations

This section provides guidance to the Internet Assigned Numbers
Authority (IANA) regarding registration of values related to the
QoS-NSLP, in accordance with BCP 26 [8]. [7].

The QoS NSLP requires IANA to create two registries. One for QoS
NSLP message types, the other for QoS NSLP objects.

This specification defines four message types: RESERVE=1, QUERY=2,
RESPONSE=3 and NOTIFY=4. Values are taken from the Message type name
space (8 bits). New Message types may be defined and assigned values
by IANA. For this, standards action is required.

Common Control Information has a Class and C-type assigned by IANA.
This specification defines the following Common Control Information
objects
RESPONSE_REQUEST: Class=1
C-type=1: empty
C-type=2: Request Identification Information

RSN: Class=2

C-type=1: RSN

REFRESH_PERIOD: Class=3

C-type=1: REFRESH_PERIOD

SESSION_ID: Class=4

C-type=1: SESSION_ID

SCOPING: Class=5

C-type=1: single hop
C-type=2: Region scoping
C-type=3: RII scoping

ERROR_SPEC: Class=6

C-type=1: empty

IANA will assign new ClassNum values and/or C-type for Common Control
Information upon specification. The required specification needs to
indicate what the correct behaviour is in case the new ClassNum or
C-type is not understood.

This specification defines a QSPEC object with assigned class = 8.
The C-type identifies the QoS model, which can be standardized,
well-known or private.
Standardized

Standardized QoS models have a C-type value in the range of 1-64.
C-type values for standardized QoS models are assigned by IANA and
require standards action.

Well-known

Well-known QoS models have a C-type value in the range of 65-128.
They are assigned by IANA and require IETF consensus.

Private

C-type values from the range 129-256 are for private use.

7. Requirements for the NSIS Transport Layer Protocol (NTLP) (GIMPS)

For the moment this section will merely describe what we assume and/
or request to be available from the NTLP. GIMPS. This section will later be
updated to describe the eventual interface when NTLP GIMPS work gets
finalized.

7.1 Session identification

The QoS NSLP keeps message and reservation state per session. A
session is identified by a Session Identifier (SESSION_ID). The
SESSION_ID is the primary index for stored NSLP state and needs to be
constant and unique (with a sufficiently high probability) along a
path through the network. We rely on the NTLP to QoS NSLP will pick a value for the
Session ID
SESSION_ID and pass it over the API.

7.2 Support for bypassing intermediate nodes

The QoS NSLP may want to restrict the handling of its messages to
specific nodes. This functionality is needed to support layering
(explained in Section 4.3), when only the edge QNEs of a domain
process the message. This requires a mechanism at the NTLP GIMPS level (which
can be invoked by the QoS NSLP) to bypass intermediates nodes between
the edges of the domain.

As a suggestion, we identified two ways for bypassing intermediate
nodes. One solution is for the end-to-end session to carry a
different protocol ID (QoS-NSLP-E2E-IGNORE protocol ID, similar to
the RSVP-E2E-IGNORE that is used for RSVP aggregation ([11]). ([10]).
Another solution is based on the use of multiple levels of the router
alert option. In that case, internal routers are configured to
handle only certain levels of router alerts. The choice between both
approaches or another approach that fulfills the requirement is left
to the NTLP GIMPS design.

7.3 Support for peer change identification

There are several circumstances where it is necessary for a QNE to
identify the adjacent QNE peer, which is the source of a signaling
application message; for example, it may be to apply the policy that
"state can only be modified by messages from the node that created
it" or it might be that keeping track of peer identity is used as a
(fallback) mechanism for rerouting detection at the NSLP layer.

We rely on the NTLP GIMPS to provide this functionality and suggest it be
implemented as an opaque identifier (Source Identification
Information (SII)) which, by default, all outgoing QoS-NSLP messages
are tagged with at the NTLP GIMPS layer. This identifier is propagated to the
next QNE, where it can be used to identify the state associated with
the message; The SII is logically similar to the RSVP_HOP object of [6];
[5]; however, any IP (and possibly higher level) addressing
information is not interpreted in the QoS-NSLP. Indeed, the
intermediate NTLP GIMPS nodes could enforce topology hiding by masking the
content of the SII (provided this is done in a stable way).

Keeping track of the SII of a certain reservation also provides a
means for the QoS-NSLP to detect route changes. When a QNE receives
a RESERVE referring to existing state but with a different SII, it
knows that its upstream peer has changed. It can then use the old
SII to send initiate a teardown along the old section of the path.
This functionality would require the NTLP GIMPS to be able to route based on
the SII. We would like this functionality to be available as a
service from the NTLP. GIMPS.

7.4 Support for stateless operation

Stateless or reduced state QoS-NSLP operation makes the most sense
when some nodes are able to operate in a stateless way at the NTLP GIMPS level
as well. Such nodes should not worry about keeping reverse state,
message fragmentation and reassembly (at the NTLP), GIMPS), congestion control
or security associations. A stateless or reduced state QNE will be
able to inform the underlying NTLP GIMPS of this situation. We rely on the NTLP
GIMPS design to allow for a mode of operation that can take advantage
of this information.

7.5 Last node detection

There are situations in which a QNE needs to determine whether it is
the last QNE on the data path (QNR), e.g. to construct and send a
RESPONSE message.

A number of conditions may result in a QNE determining that it is the
QNR:
o the QNE may be the flow destination
o the QNE have some other prior knowledge that it should act as the
QNR
o the QNE may be the last NSIS-capable node on the path
o the QNE may be the last NSIS-capable node on the path supporting
the QoS NSLP

Of these four conditions, the last two can only be detected by the
NTLP. GIMPS.
We rely on the NTLP GIMPS to inform the QoS-NSLP about these cases by
providing a trigger to the QoS-NSLP when it determines that it is the
last NE on the path, which supports the QoS-NSLP. It requires the
NTLP GIMPS
to have an error message indicating that no more NSLPs of a
particular type are available on the path.

7.6 Re-routing detection

Route changes may be detected at the NTLP GIMPS layer or the information
may be obtained by the NTLP GIMPS through local interaction with or
notification from routing protocols or modules. This specification
requests the NTLP GIMPS design to foresee notification of this information
over a route change
(over the API. API) to the QNEs upstream of the NE where the route change
is detected.

7.7 Priority of signalling messages

The QoS-NSLP will generate messages with a range of performance
requirements for the NTLP. GIMPS. These requirements may result from a
prioritization at the QoS-NSLP (Section 4.3) or from the
responsiveness expected by certain applications supported by the
QoS-NSLP.

The NTLP

GIMPS design should be able to ensure that performance for one class
of messages was not degraded by aggregation with other classes of
messages. It is currently an open issue how many priority levels are
required.

7.8 Knowledge of intermediate QoS NSLP unaware nodes

In some cases it is useful to know that a reservation has not been
installed at every router along the path. It is not possible to
determine this using only NSLP functionality.

The NTLP

GIMPS should be able to provide information to the NSLP about whether
the message has passed through nodes that did not provide support for
this NSLP.

This might be realised by the NTLP GIMPS by a mixture of NTLP GIMPS node counting,
and examination of the IP TTL or Hop Limit. The QoS NSLP, however,
does not need to know the number of intermediate nodes, only that one
or more exists.

7.9 NSLP Data Size

When the NTLP GIMPS passes the QoS NSLP data to the NSLP for processing, it
must also indicate the size of that data. (It is assumed that the
NTLP GIMPS
message structure will indicate how long this part of the NTLP GIMPS message
is.)

7.10 Notification of NTLP 'D' flag value

When the NTLP passes the QoS NSLP data to the NSLP for processing, it
must also indicate the value of the 'D' (Direction) flag for that
message.

7.11 NAT Traversal

The QoS NSLP relies on the NTLP GIMPS for NAT traversal.

8. Open issues

8.1 Aggregation error handling

QSPEC objects may be stacked to allow aggregation and layering. In
error-free conditions, the top of the QSPEC stack has Assumptions on the QSPEC
object QoS model

8.1 Resource sharing

This specification assumes that resource sharing is locally valid.

A QNE may receive a QoS NSLP message possible between
flows with the same SESSION_ID that originate from the same QNI or
between flows with a QSPEC stack of which different SESSION_ID that are related through
the
top object BOUND_SESSION_ID object. For flows with the same SESSION_ID,
resource sharing is not recognised. This can occur under error conditions,
e.g. only applicable when a domain boundary is misconfigured, or it me be the result
from a policy to detect domain boundaries existing reservation is
not just replaced (which is indicated by encountering
unrecognised QSPEC objects.

In some situations, a QNE the NO_REPLACE flag in the
common header.

The Resource Management Function (RMF) reserves resources for each
flow. We assume that the QoS model supports resource sharing between
flows. A QoS model may be able elect to recover from the error
condition by inspecting implement a larger portion more general behaviour
of supporting relative operations on existing reservations, such as
ADDING or SUBTRACTING a certain amount of resources from the stack.It is currently
an open question whether
o current
reservation. A QNE should be allowed QoS model may also elect to do allow resource sharing
more generally, e.g. between all flows with the same DSCP.

8.2 Reserve/commit support

Reserve/commit behaviour means that instead of or in addition to
sending an error.
o How far the stack can time at which the reservation
is made may be inspected.
o If and how different from the QNE should update time when the stack reserved resources
are actually set aside for the requesting session. This
specification acknowledges the usefulness of such a mechanism but
assumes that its implementation is opaque to QoS NSLP and is fully
handled by the QoS model. A COMMIT flag in case it finds the QoS-model specific
control information is suggested as a
QSPEC it recognises.

8.2 way to support this
functionality.

9. Open issues

9.1 Region scoping

This specification allows QNEs to scope their messages, i.e. to
restrict the extent to which messages may travel along and be
interpreted on the path. For this, the scopes of whole path, single
hop and back to me (RII) are defined. Also, a region can be
configured administratively or it can be derived from some other
means (e.g. RAO levels) in case of aggregation.

It is currently an open question whether this

This specification should currently does not define and support a more
generic notion of region (e.g. to implement region policies
independent from aggregation regions,...).

8.3 It is proposed to use the
concept of Localized RSVP for regions.

9.2 Priority of reservations

Support of reservation priority is a QoS model specific issue and
therefore outside the scope of this specification. However, the
concepts of setup and holding priority are widely accept and we
expect the specification of a Priority object in the QSPEC template
to be useful for a wide range of QoS models.

It is an open question to the NSIS community whether the concepts of
setup and holding priority are useful enough to define a priority
object in this specification. Alternatively, this could be left as
QoS model specific.

9.3 Peering agreements on interdomain links

This specification proposes ways to carry AAA information that may be
used at the edges of a domain to check whether the requestor is
allowed to use the requested resources. It is less likely that the
AAA information will be used inside a domain. In practice, there may
be peering relations between domains that allow for a certain amount
of traffic to be sent on an interdomain link without the need to
check the authorization of each individual session (effectively
making the peering domain the requestor of the resources). The
per-session authorization check may be avoided by setting up an
aggregate reservation on the inter-domain link for a specified amount
of resources and relating the end-to-end sessions to it using the
BOUND_SESSION_ID. In this way, the aggregate session is authorized
once (and infrequently updated). An alternative is for the edge node
of a domain to insert a token that authorizes the flow for the next
domain.

9.4 GIMPS Modifications for Refresh Overhead Reduction

As described in Section 4.2.4 a mechanism is available to reduce the
overhead of refresh messages. As currently specified, GIMPS may opt
to bundle several NSLP messages together. However, this still has
some additional overhead, particularly due to the requirement to
include Flow Routing Information (FRI) in every message. This may
not be strictly necessary for a message going only to the next GIMPS
hop.

It is an open issue to examine what additional optimisations are
possible and appropriate for refresh overhead reduction.

9.5 Path state maintenance implementation at NSLP

As currently described in Section 3.5.3, a QoS NSLP message is
required to create the necessary GIMPS reverse path state so that the
receiver-initiated RESERVE message can be sent upstream. (This must
be an NSLP message so that it visits the correct subset of GIMPS
nodes on the path.) This message may also be used to refresh the
GIMPS path state (see Section 9.6).

In this version, we have implemented this functionality by making the
RESPONSE_REQUEST object in QUERY optional. Basically, this means
that we send an empty QUERY message along the path. It is not clear
whether this path state maintenance functionality is sufficiently
different from QUERY to warrant the definition of a separate (NULL)
message.

It is also worth investigating whether other NSLPs have a similar
need and whether in that case it would be better to define a separate
NULL message across all NSLPs.

9.6 GIMPS Path State Maintenance

As currently described in Section 3.5.3, when performing
receiver-initiated reservations the QoS NSLP sends periodic
downstream QUERY messages to ensure that GIMPS reverse path state is
refreshed.

It is unclear how often such messages need to be sent, since the
lifetime of the path state is a matter for GIMPS. It, therefore, may
be necessary for GIMPS to inform the NSLP about this state lifetime.

An alternative solution to the refreshing of GIMPS path state by NSLP
messages is for GIMPS to refresh (and, where necessary, detect
changes in) its path state automatically, by sending periodic probe/
refresh messages of its own. The transmission of upstream messages
(such as the RESERVE in a receiver-initiated QoS NSLP reservation)
can be used to indicate that the reverse path state is still needed.
Whether such a technique is appropriate is for further consideration.

9.7 Protocol Operating Environment Assumptions

For receiver initiated and bidirectional reservations the question
arises of what assumptions to make about what end-to-end information
should be determined outside the NSIS protocols and what should be
carried end-to-end in NSLP messages in order to initiate signalling.
The NSIS protocol is not used alone. It is used in conjunction with
a variety of applications. The question arises of what the
interactions between NSIS (and the NSLP in particular) and these
applications is.

For a receiver initiated reservation, the we have the questions: How
do the sender and receiver determine that a receiver initiated
reservation is to be performed? And, how does information needed by
the receiver to perform the reservation, but only available at the
sender, be made transferred to the receiver so that the RESERVE
message can be sent?

In the bi-directional reservation case, we can either perform this as
a pair of two sender-initiated reservations or as a combination of
sender-initiated and receiver-initiated reservations. The latter
case has the same issues as for the general receiver initiated
reservation problem. The former raises similar questions: How does
the remote end know that a reservation is needed? And, how does it
know what resources to request?

Is it reasonable to assume that the decision that an end should
initiate a reservation is made totally outside the QoS NSLP itself
(e.g. through prior configuration, or application end-to-end
signalling such as SIP) or, should the QoS NSLP messages include some
method to trigger the other end to perform a reservation (whether
that be a receiver initiated reservation, or a sender initiated
reservation for the first bidirectional reservation case)?

In addition, should the QoS NSLP messages be able to carry extra data
(e.g. a QSpec object for the reverse direction) end-to-end that is
needed by the remote end to perform its reservation? (And, should
this be in the QoS NSLP, or through individual QoS models?) The
alternative to providing support in the QoS NSLP for this is to leave
it to application signalling to transfer any required information.

10. Security Considerations

9.1
10.1 Introduction and Threat Overview

The security requirement for the QoS NSLP is to protect the signaling
exchange for establishing QoS reservations against identified
security threats. For the signaling problem as a whole, these
threats have been outlined in [21]; the NSIS framework [3] [15] assigns a
subset of the responsibility to the NTLP GIMPS and the remaining threats need
to be addressed by NSLPs. The main issues to be handled can be
summarised as:
Authorization:

The QoS NSLP must assure that the network is protected against
theft-of-service by offering mechanisms to authorize the QoS
reservation requestor. A user requesting a QoS reservation might
want proper resource accounting and protection against spoofing
and other security vulnerabilities which lead to denial of service
and financial loss. In many cases authorization is based on the
authenticated identity. The authorization model must provide
guarantees that replay attacks are either not possible or limited
to a certain extent. Authorization can also be based on traits
which enables the user to remain anonymous. Support for user
identity confidentiality can be accomplished.

Message Protection:

Signaling message content should be protected against
modification, replay, injection and eavesdropping while in
transit. Authorization information, such as authorization tokens,
need protection. This type of protection at the NSLP layer is
neccessary to protect messages between NSLP nodes which includes
end-to-middle, middle-to-middle and even end-to-end protection.

In addition to the above-raised issues we see the following
functionality provided at the NSLP layer:
Prevention of Denial of Service Attacks:

GIMPS and QoS NSLP nodes have finite resources (state storage,
processing power, bandwidth). The protocol mechanisms suggested
in this document should try to minimise exhaustion attacks against
these resources when performing authentication and authorization
for QoS resources.

To some extent the QoS NSLP relies on the security mechanisms
provided by GIMPS which by itself relies on existing authentication
and key exchange protocols. Some signaling messages cannot be
protected by GIMPS and hence should be used with care by the QoS
NSLP. An API must ensure that the QoS NSLP implementation is aware
of the underlying security mechanisms and must be able to indicate
which degree of security is provided between two GIMPS peers. If a
level of security protection for QoS NSLP messages is required which
goes beyond the security offered by GIMPS or underlying security
mechanisms, additional security mechanisms described in this document
must be used. The different usage environments and the different
scenarios where NSIS is used make it very difficult to make general
statements without reducing its flexibility.

9.2

10.2 Trust Model

For this version of the document we will rely on a model which
requires trust between neighboring NSLP nodes to establish a
chain-of-trust along the QoS signaling path. This model is simple to
deploy, was used in previous QoS authorization environments (such as
RSVP) and seems to provide sufficiently strong security properties.
We refer to this model as the 'New Jersey Turnpike' model.

On the New Jersey Turnpike, motorists pick up a ticket at a toll
booth when entering the highway. At the highway exit the ticket is
presented and payment is made at the toll booth for the distance
driven. For QoS signaling in the Internet this procedure is roughly
similar. In most cases the data sender is charged for transmitted
data traffic whereby where charging is provided only between neighboring
entities.

+------------------+ +------------------+ +------------------+
| Network | | Network | | Network |
| X | | Y | | Z |
| | | | | |
| -----------> -----------> |
| | | | | |
| | | | | |
+--------^---------+ +------------------+ +-------+----------+
| .
| .
| v
+--+---+ Data Data +--+---+
| Node | ==============================> | Node |
| A | Sender Receiver | B |
+------+ +------+

Legend:

----> Peering relationship which allows neighboring
networks/entities to charge each other for the
QoS reservation and data traffic

====> Data flow

..... Communication to the end host

Figure 16: 22: New Jersey Turnpike Model

The model shown in Figure 16 22 uses peer-to-peer relationships between
different administrative domains as a basis for accounting and
charging. As mentioned above, based on the peering relationship a
chain-of-trust is established. There are several issues which come
to mind when considering this type of model:
o This model allows authorization on a request basis or on a
per-session basis. Authorization mechanisms will be elaborated in
Section 9.3. 3.6. The duration for which the QoS authorization is
valid needs to be controlled. Combining the interval with the
soft-state interval is possible. Notifications from the networks
also seem to be viable approach.
o The price for a QoS reservation needs to be determined somehow and
communicated to the charged entity and to the network where the
charged entity is attached. Price distribution protocols are not
covered in this version of the document. This model assumes, per
default, that the data sender is authorizing the QoS reservation.
Please note that this is only a simplification and further
extensions are possible and left for a future version of this
document.

o This architecture seems to be simple enough to allow a scalable
solution (ignoring reverse charging, multicast issues and price
distribution).

Charging the data sender as performed in this model simplifies
security handling by demanding only peer-to-peer security protection.
Node A would perform authentication and key establishment. The
established security association (together with the session key)
would allow the user to protect QoS signaling messages. The identity
used during the authentication and key establishment phase would be
used by Network X (see Figure 16) 22) to perform the so-called
policy-based admission control procedure. In our context this user
identifier would be used to establish the necessary infrastructure to
provide authorization and charging. Signaling messages later
exchanged between the different networks are then also subject to
authentication and authorization. The authenticated entity thereby
is, however, the neighboring network and not the end host.

The New Jersey Turnpike model is attractive because of its
simplicity. S. Schenker et. al. [23] [24] discuss various accounting
implications and introduced the edge pricing model. The edge pricing
model shows similarity to the model described in this section with
the exception that mobility and the security implications itself are
not addressed.

9.3 QoS Authorization

Authorization is a necessary function in order to prevent
theft-of-service and to enable charging. With regard to authorization
a few issues still need to be resolved to specify the protocol
interaction for a QoS NSLP with regard to authorization of resource
requests.

This section provides a description of the different approaches for
providing authorization for QoS resource requests. Three different
approaches are shown, whereby one is a two-party and two others
describe a three party approach.

9.3.1 Authorization for the two party approach

This section starts with the conceptually simpler two party approach.

Figure 17: Two party approach

Figure 17 describes the simple and basic approach where
the authorization decision is purely executed between the two
entities only or
where previous (out-of-band) mechanisms separated the signaling
protocol from executing other entities during NSIS protocol
execution.

The entity authorizing the resource request and the entity actually
performing the QoS reservation are in the same administrative domain.
Hence they are treated as a single logical entity.

Examples for this type of model can be found between two neighboring
networks (inter-domain signaling) where a long-term contract (or
other out-of-band mechanisms) exists and allows both networks to know
how to charge the other entity (i.e. how the authorizing entity
finally gets paid for the consumed resources) and
how to authorize the resource requesting entity (i.e. associating
the identifier of the protected signaling message to the identity
used in the authentication and key exchange protocol run and
finally this identity to the user identity of the contract for the
purpose of charging).

No additional message signaling for authorization is required. In
this scenario the identity used during the authentication and key
exchange process is used for authorizing the same entity. The QoS
NSLP needs to have access to this authenticated identity via an API.

9.3.2 Token based three party approach

This section describes an approach which uses authorization tokens
such as those introduced with [12] and [13] or with the Open
Settlement protocol [26]. The former only associates two different
signaling protocols and their authorization with each other whereas
the latter is a form of digital money. In this text we refer to the
former as the 'authorization tokens' and in the latter case as 'OSP
tokens'. In case of authorization tokens the entity which requests
authorization wants to run, for example, SIP with an entity in the
local network and wants to experience quality of service for the
media traffic. Some form of authorization will be provided at the SIP
proxy, which acts as the resource authorizing entity in Figure 18. In
case of a successful verification of the request SIP signaling
returns an authorization token which is subsequently included in the
QoS signaling protocol to refer to the previous authorization
decision. The authorization decision can be passed by value or by
reference. The advantage of the latter is that the token is smaller
(i.e., effectively only a pointer to installed state in the network)
with the disadvantage that the entity performing the QoS reservation
has to query the state, possibly from a central entity.

The token based approach assumes that the entity which authorizes the
QoS request (and which also creates the token) is trusted by the
entity which performs the QoS reservation. These two entities do not
necessarily need to be in the same administrative domain. Security
mechanisms must ensure that
the token cannot be modified
the token questing entity is authenticated and authorized at the
token granting entity
the token cannot be stolen and reused by an adversary

Hence, to prevent an adversary from eavesdropping and stealing the
authorization token it is necessary to establish at least a
unilateral authenticated secure channel between entity A and B. As a
side-effect it is possible to provide anonymous authorization since
the authorization decision based on the received token by entity B
does not need to be based on the identity of A. This assumes that
entity C does not provide entity B with the identity.

Figure 18: Token based three party approach

The token is only an attribute in the QoS NSLP message. The token
acts as a form of voucher and is therefore a one-shot message. For
the OSP token (or digital money) alike approach, as soon as the
credits are consumed a new token needs to be requested in order.
Refresh messages can therefore be used to trigger the transmission of
new tokens. A trigger message from the network is necessary to
request a new token. Tokens provide a good mechanism for the client
to restrict the amount of spend resources and to quickly learn about
the cost of a QoS reservation if tokens represent only a small value
(such as those used in hash-chain based approaches). The refresh
interval is therefore, in some sense, bound to the "charging"
interval.

Please note that OSP tokens only serve as an example here. The
content of the OSP token is tailored towards its usage in the
telephony environment. Therefore, we see OSP tokens as a prominent
representative of authorization token usage.

Since authorization tokens or OSP tokens can be fairly large
fragmentation is possible or even likely.

9.3.3 Generic three party approach

This section covers a generic three party approach. Figure 19 shows
the intra-domain variant of the exchange.

Figure 19: Three party approach (intra-domain)

The main difference between the scenario in Figure 19 and Figure 20
is the trust relationship between the participating entities. In
Figure 20 the home AAA server is responsible for authoring the QoS
request. This might be on a per-request basis, periodically, or on a
per-session basis. In both cases the EAP authentication runs between
the EAP Peer (entity A in Figure 19) and between the EAP Server
(entity C in Figure 19). For the EAP method protocol run the
Authenticator (entity B in Figure 19) is not actively involved. To
fulfill the requirements of the EAP keying framework it is necessary
to execute a protocol exchange between entity A and entity B
subsequently to successful EAP authentication. This exchange should
lead to a secure channel between these two entities.

The main advantage of this exchange is that
a number of authentication and key exchange protocols can be used
in a very flexible fashion; these protocols can be tailed exactly
to the needs of the architecture and the environment
a secure channel can be established
the protocol exchange is effectively a three party protocol
authorization can be incorporated in a very flexible way which
allows the home network (or some other entity) to give tight
control over the sessions

The disadvantage of this approach is that there is no out-of-the-box
solution available. Further investigation is required here.

+-----------------------------+ +-----------------+
| Local Network | | |
| |QoS | |
| +------------+ authz. req. +---------+ |
| | Local |-----+----+--->| Home | |
| | AAA | |QoS | | AAA | |
| | Server |<----+----+----| Server | |
| +---------+--+ authz. res. +---------+ |
| ^ | | | |
| | | <...financial...> |
| QoS | QoS | settlement |
| authz| authz| | | |
| req.| res.| | | |
| | | | | |
| | v | | |
+----------+ QoS +----+---------+ | | Users |
|Entity | request | Entity | | | Home Network |
|requesting|--+------->| performing | | +-----------------+
|resource |<-+--------| QoS | |
+----------+ granted/ | reservation | |
rejected +--------------+ |
| |
+-----------------------------+

Figure 20: Three party approach (inter-domain)

9.3.4

10.3 Computing the authorization decision

Whenever an authorization decision has to be made then there is the
question which information serves as an input to the authorizing
entity. The following information items have been mentioned in the
past for computing the authorization decision (in addition to the
authenticated identity):
Price
QoS objects
Policy rules

Policy rules include attributes like time of day, subscription to
certain services, membership, etc. into consideration when computing
an authorization decision.

A detailed description of the authorization handling will be left for
a future version of this document. The authors assume that the QoS
NSLP needs to provide a number of attributes to support the large
range of scenarios.

10.

11. Change History
Changes from -00
* Additional explanation of RSN versus Session ID differences.
(Session IDs still need to be present and aren't replaced by
RSNs. Explain how QoS-NSLP could react once it notes that it
maintains stale state.)
* Additional explanation of message types - why we don't just
have RESERVE and RESPONSE.
* Clarified that figure 1 is not an implementation restriction.
Changes from -01
* Significant restructuring.
* Added more concrete details of message formats and processing.
* Added description of layering/aggregation concepts.
* Added details of authentication/authorisation aspects.

11.
Changes from -02
* Addressed comments from early review.
* Added text on receiver-initiated and bi-directional
reservations.
* Extended description of session binding. Added support for
fate sharing.
* Restructured message formats and processing section.
* Clarified refresh reduction mechanism.
* Added assumptions on QoS model.
* Added assumptions on operating environment.

12. Acknowledgements

The authors would like to thank Eleanor Hepworth for her useful
comments.

12.

13. Contributors

This draft combines work from three individual drafts. The following
authors from these drafts also contributed to this document: Robert
Hancock (Siemens/Roke Manor Research), Hannes Tschofenig and Cornelia
Kappler (Siemens AG), Lars Westberg and Attila Bader (Ericsson) and
Maarten Buechli (Dante) and Eric Waegeman (Alcatel).

Yacine El Mghazli (Alcatel) contributed text on AAA.

14. References

14.1 Normative References

[1] Bradner, S., "Key words for use in RFCs to Indicate Requirement
Levels", BCP 14, RFC 2119, March 1997.

[2] Crocker, D. and P. Overell, "Augmented BNF for Syntax
Specifications: ABNF", RFC 2234, November 1997.

[3] Hancock, R., "Next Steps in Signaling: Framework",
draft-ietf-nsis-fw-05 (work in progress), October 2003.

[4] Schulzrinne, H., "GIMPS: General Internet Messaging Protocol for
Signaling", draft-ietf-nsis-ntlp-00 draft-ietf-nsis-ntlp-01 (work in progress), October
2003. February
2004.

14.2 Informative References

[5]

[4] Braden, B., Clark, D. and S. Shenker, "Integrated Services in
the Internet Architecture: an Overview", RFC 1633, June 1994.

[6]

[5] Braden, B., Zhang, L., Berson, S., Herzog, S. and S. Jamin,
"Resource ReSerVation Protocol (RSVP) -- Version 1 Functional
Specification", RFC 2205, September 1997.

[7]

[6] Wroclawski, J., "The Use of RSVP with IETF Integrated
Services", RFC 2210, September 1997.

[8]

[7] Narten, T. and H. Alvestrand, "Guidelines for Writing an IANA
Considerations Section in RFCs", BCP 26, RFC 2434, October
1998.

[9]

[8] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z. and W.
Weiss, "An Architecture for Differentiated Services", RFC 2475,
December 1998.

[10]

[9] Berger, L., Gan, D., Swallow, G., Pan, P., Tommasi, F. and S.
Molendini, "RSVP Refresh Overhead Reduction Extensions", RFC
2961, April 2001.

[11]

[10] Baker, F., Iturralde, C., Le Faucheur, F. and B. Davie,
"Aggregation of RSVP for IPv4 and IPv6 Reservations", RFC 3175,
September 2001.

[12]

[11] Hamer, L-N., Gage, B., Kosinski, B. and H. Shieh, "Session
Authorization Policy Element", RFC 3520, April 2003.

[13]

[12] Hamer, L-N., Gage, B. and H. Shieh, "Framework for Session
Set-up with Media Authorization", RFC 3521, April 2003.

[14]

[13] Chaskar, H., "Requirements of a Quality of Service (QoS)
Solution for Mobile IP", RFC 3583, September 2003.

[15]

[14] Brunner, M., "Requirements for Signaling Protocols",
draft-ietf-nsis-req-09 RFC 3726,
April 2004.

[15] Hancock, R., "Next Steps in Signaling: Framework",
draft-ietf-nsis-fw-05 (work in progress), August October 2003.

[16] Tschofenig, H., "NSIS Authentication, Authorization and
Accounting Issues", draft-tschofenig-nsis-aaa-issues-01 (work
in progress), March 2003.

[17] Tschofenig, H., "QoS NSLP Authorization Issues",
draft-tschofenig-nsis-qos-authz-issues-00 (work in progress),
June 2003.

[18] Ash, J., "NSIS Network Service Layer Protocol QoS Signaling
Proof-of-Concept",
draft-ash-nsis-nslp-qos-sig-proof-of-concept-01 (work in
progress), February 2004.

[19] Kappler, C., "A QoS Model for Signaling IntServ Controlled-Load
Service with NSIS",
draft-kappler-nsis-qosmodel-controlledload-00 (work in
progress), February 2004.

[20] Bader, A., "RMD (Resource Management in Diffserv) QoS-NSLP
model", draft-bader-rmd-qos-model-00 (work in progress),
February 2004.

[21] Tschofenig, H. and D. Kroeselberg, "Security Threats for NSIS",
draft-ietf-nsis-threats-03
draft-ietf-nsis-threats-04 (work in progress), October 2003. February 2004.

[22] Westberg, L., "Resource Management in Diffserv (RMD)
Framework", draft-westberg-rmd-framework-04.txt, work in
progress, September 2003.

[23] Breslau, L., "Two Issues in Reservation Establishment", Proc.
ACM SIGCOMM '95 , Cambridge , MA , August 1995.

[24] Shenker, S., Clark, D., Estrin, D. and S. Herzog, "Pricing in
computer networks: Reshaping the research agenda", Proc. of
TPRC 1995, 1995.

[24]

[25] Metro Ethernet Forum, "Ethernet Services Model", letter ballot
document , August 2003.

[25]

[26] Jacobson, V., "Synchronization of Periodic Routing Messages",
IEEE/ACM Transactions on Networking , Vol. 2 , No. 2 , April
1994.

[26]

[27] ETSI, "Telecommunications and internet protocol harmonization
over networks (tiphon); open settlement protocol (osp) for
inter- domain pricing, authorization, and usage exchange",
Technical Specification 101 321, version 2.1.0.

Authors' Addresses

Sven Van den Bosch
Alcatel
Francis Wellesplein 1
Antwerpen B-2018
Belgium

EMail: sven.van_den_bosch@alcatel.be

Georgios Karagiannis
University of Twente/Ericsson
P.O. Box 217
Enschede 7500 AE
The Netherlands

EMail: karagian@cs.utwente.nl

Andrew McDonald
Siemens/Roke Manor Research
Roke Manor Research Ltd.
Romsey, Hants SO51 0ZN
UK

EMail: andrew.mcdonald@roke.co.uk

Appendix A. Object Definitions

The currentlly specified C-Types definitions are contained in this
Appendix. To accommodate other address families, additional C-Types
could easily be defined.

All unused fields should be sent as zero and ignored on receipt.

A.1 RESPONSE_REQUEST Class

RESPONSE_REQUEST Class = 1.

RESPONSE_REQUEST object: Class = 1, C-Type = 1

The object content is empty

RESPONSE_REQUEST object: Class = 1, C-Type = 2

+-------------+-------------+-------------+-------------+
| Request Identification Information (RII)(4 bytes) |
+-------------+-------------+-------------+-------------+

Request Identification Information (RII) (4 bytes)

An identifier which must be (probabilistically) unique
within the context of a SESSION_ID, and SHOULD be different
for each response request. Used by a node to match back a
RESPONSE to a request in a RESERVE or QUERY message.

A.2 RSN Class

RSN class = 2.

RSN object: Class = 2, C-Type = 1

+-------------+-------------+-------------+-------------+
| Reservation Sequence Number (RSN) (4 bytes) |
+-------------+-------------+-------------+-------------+

Reservation Sequence Number (RSN) (4 bytes)

An incrementing sequence number that indicates the order in
which state modifying actions are performed by a QNE. It
has local significance only, i.e. between a pair of
neighbouring stateful QNEs.

A.3 REFRESH_PERIOD Class

REFRESH_PERIOD class = 3.

REFRESH_PERIOD Object: Class = 3, C-Type = 1

+-------------+-------------+-------------+-------------+
| Refresh Period R (4 bytes) |
+-------------+-------------+-------------+-------------+

Refresh Period R (4 bytes)

The refresh timeout period R used to generate this message;
in milliseconds.

A.4 SESSION_ID Class

SESSION_ID class = 4.

SESSION_ID Object: Class = 4, C-Type = 1

+-------------+-------------+-------------+-------------+
| |
+ +
| |
+ SESSION_ID (16 bytes) +
| |
+ +
| |
+-------------+-------------+-------------+-------------+

SESSION_ID (16 bytes)

It represents the SESSION_ID as specified in [3] [15] of the
session that must be bound to the session associated to the
message carrying this object.

A.5 SCOPING Class

SCOPING class = 5.

SCOPING Object: Class = 5, C-Type = 1

No content value. Selection of a single hop message scoping.

SCOPING Object: Class = 5, C-Type = 2

+-------------+-------------+-------------+-------------+
| Region scoping (4 bytes) |
+-------------+-------------+-------------+-------------+

Region scoping (4 bytes)

Ordered number, forwarded by routers belonging to region
with same or higher number;

SCOPING Object: Class = 5, C-Type = 3

+-------------+-------------+-------------+-------------+
| RII scoping (4 bytes) |
+-------------+-------------+-------------+-------------+

RII (back to me) scoping (4 bytes)

A.6 ERROR_SPEC Class
ERROR_SPEC class = 6.

ERROR_SPEC object: Class = 6, C-Type = 1

+-------------+-------------+-------------+-------------+
| Error (4 bytes) |
+-------------+-------------+-------------+-------------+
| Flags | Error Code | Error Value |
+-------------+-------------+-------------+-------------+

Error (4 bytes)

To be done

Flags (1 byte)

To be done

Error Code (1 byte)

A one-octet error description.

Error Value (2 bytes)

A two-octet field containing additional information about
the error. Its contents depend upon the Error Type.

The values for Error Code and Error Value are defined in
Appendix B (to be done).

A.7 POLICY_DATA Class

This section presents a set of specifications for supporting generic
authorization in QoS NSLP. These specs include the standard format
of POLICY_DATA objects, and a description of QoS NSLP handling of
authorization events. This section does not advocate a particular
authorization approach (2-party, 3-party, token-based 3-party).

The traffic control block is responsible for controlling and
enforcing access and usage policies.

A.7.1 Base Format

POLICY_DATA object: Class=7, C-Type=1

+-------------------------------------------------------+
| |
// Option List //
| |
+-------------------------------------------------------+
| |
// Policy Element List //
| |
+-------------------------------------------------------+

Option List: Variable length. See more details in Appendix A.7.2.
Policy Element List: Variable length. See more details in Appendix
A.7.3.

A.7.2 Options

This section describes a set of options that may appear in
POLICY_DATA objects. Some policy options appear as QoS NSLP objects
but their semantic is modified when used as policy data options.

Policy Refresh TIME_VALUES (PRT) object:

The Policy Refresh TIME_VALUES (PRT) option is used to slow policy
refresh frequency for policies that have looser timing constraints
relative to QoS NSLP. If the PRT option is present, policy
refreshes can be withheld as long as at least one refresh is sent
before the policy refresh timer expires. A minimal value for PRT
is the NSLP session refresh period R; lower values are assumed to
be R (neither error nor warning should be triggered). This option
is especially useful to combine strong (high overhead) and weak
(low overhead) authentication certificates as policy data. In
such schemes the weak certificate can support admitting a
reservation only for a limited time, after which the strong
certificate is required. This approach may reduce the overhead of
POLICY_DATA processing. Strong certificates could be transmitted
less frequently, while weak certificates are included in every QoS
NSLP refresh.

Policy Source Identification Information (PSII) object:

The Policy SII object identifies the neighbor/peer policy-capable
QN that constructed the policy object. When policy is enforced at
border QNEs, peer policy nodes may be several NSLP hops away from
each other and the SII is the basis for the mechanism that allows
them to recognize each other and communicate safely and directly.
As stated above, we assume such an (P)SII to be available from a
service from the NTLP. GIMPS. If no PSII object is present, the policy data
is implicitly assumed to have been constructed by the QoS NSLP HOP
indicated in the SII (i.e., the neighboring QoS NSLP node is
policy-capable).

Integrity object:

The INTEGRITY object option inside POLICY_DATA object creates
direct secure communications between non-neighboring policy aware
nodes without involving PIN nodes.

A.7.3 Policy Elements

There are no requirements for all nodes to process this container.
Policy data is opaque to NSLP, which simply passes it to policy
control when required.

The content of policy elements is opaque to the QoS NSLP layer. Only
policy peers understand their internal format and NSLP layer simply
passes it to policy control when required.

Policy Elements have the following format:

+-------------+-------------+-------------+-------------+
| Length | P-Type |
+---------------------------+---------------------------+
| |
// Policy information (Opaque to QoS NSLP) //
| |
+-------------------------------------------------------+

A.7.3.1 Authorization token Policy Element

The AUTHZ_TOKEN policy element contains a list of fields, which
describe the session, along with other attributes.

+-------------+-------------+-------------+-------------+
| Length | P-Type = AUTHZ_TOKEN |
+-------------+-------------+-------------+-------------+
// Session Authorization Attribute List //
+-------------------------------------------------------+

Session Authorization Attribute List: variable length. The
session authorization attribute list is a collection of objects
which describes the session and provides other information
necessary to verify the resource reservation request. See [12] [11]
for a details.
Session Authorization Attributes. A session authorization
attribute may contain a variety of information and has both an
attribute type and subtype. The attribute itself MUST be a
multiple of 4 octets in length, and any attributes that are not a
multiple of 4 octets long MUST be padded to a 4-octet boundary.
All padding bytes MUST have a value of zero.

+--------+--------+--------+--------+
| Length | X-Type |SubType |
+--------+--------+--------+--------+
| Value ... |
+--------+--------+--------+--------+

Length: 16 bits

The length field is two octets and indicates the actual length of
the attribute (including Length, X-Type and SubType fields) in
number of octets. The length does NOT include any bytes padding
to the value field to make the attribute a multiple of 4 octets
long.

X-Type: 8 bits

Session authorization attribute type (X-Type) field is one octet.
IANA acts as a registry for X-Types as described in Section 6.
Initially, the registry contains the following X-Types:
1 AUTH_ENT_ID: The unique identifier of the entity which
authorized the session.
2 SESSION_ID: Unique identifier for this session.
3 SOURCE_ADDR: Address specification for the session originator.

4 DEST_ADDR: Address specification for the session end-point.
5 START_TIME: The starting time for the session.
6 END_TIME: The end time for the session.
7 RESOURCES: The resources which the user is authorized to
request.
8 AUTHENTICATION_DATA: Authentication data of the session
authorization policy element.
SubType: 8 bits

Session authorization attribute sub-type is one octet in length.
The value of the SubType depends on the X-Type.

Value: variable length

The attribute specific information is defined in [12]. [11].

A.7.3.2 OSP Token Policy Element

To be completed.

A.7.3.3 User Identity Policy element

To be completed.

A.8 QSPEC Class

QSPEC class = 8.

QSPEC object: Class = 8, C-Type = (QoS model ID)

This object contains the QSPEC (QoS specification) information.
Its content has a variable length and it is QoS model specific.
Such a QoS model can be a standardized one, a private one, or a
well-known one. The C-Type contains the QoS model ID that
identifies the used QSPEC.

The contents and encoding rules for this object are specified
in other documents, prepared by QoS model designers.

Intellectual Property Statement

The IETF takes no position regarding the validity or scope of any
intellectual property
Intellectual Property Rights or other rights that might be claimed to
pertain to the implementation or use of the technology described in
this document or the extent to which any license under such rights
might or might not be available; neither nor does it represent that it has
made any independent effort to identify any such rights. Information
on the
IETF's procedures with respect to rights in standards-track and
standards-related documentation RFC documents can be
found in BCP-11. BCP 78 and BCP 79.

Copies of
claims of rights IPR disclosures made available for publication to the IETF Secretariat and any
assurances of licenses to be made available, or the result of an
attempt made to obtain a general license or permission for the use of
such proprietary rights by implementors implementers or users of this
specification can be obtained from the IETF Secretariat. on-line IPR repository at
http://www.ietf.org/ipr.

The IETF invites any interested party to bring to its attention any
copyrights, patents or patent applications, or other proprietary
rights which that may cover technology that may be required to practice implement
this standard. Please address the information to the IETF Executive
Director.

Full Copyright Statement

This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose at
ietf-ipr@ietf.org.

Disclaimer of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.

The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assignees. Validity

This document and the information contained herein is are provided on an
"AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
ENGINEERING TASK FORCE DISCLAIMS DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Copyright (C) The Internet Society (2004). This document is subject
to the rights, licenses and restrictions contained in BCP 78, and
except as set forth therein, the authors retain all their rights.

Acknowledgment

Funding for the RFC Editor function is currently provided by the
Internet Society.