A Common API for Transparent Hybrid
Multicastlink-lab & FU BerlinHoenower Str. 35Berlin10318Germanymw@link-lab.nethttp://www.inf.fu-berlin.de/~waehlHAW HamburgBerliner Tor 7Hamburg20099Germanyschmidt@informatik.haw-hamburg.dehttp://inet.cpt.haw-hamburg.de/members/schmidtcisco SystemsTasman DriveSan JoseCA95134USAstig@cisco.comSAM Research GroupGroup communication services are most efficiently implemented on the
lowest layer available. However, as the deployment status of multicast
technologies largely varies throughout the Internet, globally
operational group solutions are frequently forced to using a stable,
upper layer protocol controlled by the application itself. This document
describes a common multicast API that is suitable for transparent
underlay and overlay communication. It proposes abstract naming and
addressing by multicast URIs and discusses mapping mechanisms between
different namespaces and distribution technologies. Additionally, it
describes the application of this API for building gateways that
interconnect current multicast domains throughout the Internet.Currently, group application programmers need to make a choice of the
distribution technology required at runtime. There is no common
communication interface that abstracts multicast subscriptions from the
underlying deployment. The standard multicast socket options , are bound to an
IP version and do not distinguish between naming and addressing of
multicast identifiers. Group communication, however, is commonly
implemented in different flavors and on different layers (e.g., IP vs.
application layer multicast), and may be based on different technologies
on the same tier (e.g., IPv4 vs. IPv6). Multicast application development should be decoupled of
technological deployment throughout the infrastructure. It requires a
common multicast API that offers calls to transmit and receive multicast
data independent of the supporting layer and the underlying
technological details. For inter-technology transmissions, a consistent
view on multicast states is needed, as well. This document describes an
abstract group communication API and core functions necessary for
transparent operations. Specific implementation guidelines with respect
to operating systems or programming languages are out-of-scope of this
document.In contrast to the standard multicast socket interface, the API
introduced in this document abstracts naming and addressing. Using a
multicast address in the current socket API predefines the corresponding
routing layer. In this memo, the multicast address used for joining a
group denotes an application layer data stream that is identified by a
multicast URI and without an association to the underlying distribution
technology. Group name can be mapped to variable routing
identifiers.The aim of this common API is twofold: Enable any application programmer to implement group-oriented
data communication independent of the underlying delivery
mechanisms. In particular, allow for a late binding of group
applications to multicast technologies that makes applications
efficient, but robust with respect to deployment aspects.Allow for a flexible namespace support in group addressing, and
thereby separate naming and addressing/routing schemes from the
application design. This abstraction not only reduces the dependency
on specific apects of underlying protocols, but may open application
design to extend to specifically flavored group services.Multicast technologies may be various P2P-based, IPv4 or IPv6 network
layer multicast, or implemented by some other application service.
Corresponding namespaces may be IP addresses, overlay hashes, other
application layer group identifiers, e.g., <sip:*@peanuts.org>, or
names defined by the applications.This document also proposes and discusses mapping mechanisms between
different namespaces and forwarding technologies. Additionally, the
multicast API provides internal interfaces to access current multicast
states at the host. Multiple multicast protocols may run in parallel on
a single host. These protocols may interact to provide a gateway
function that bridges data between different domains. The application of
this API at gateways operating between current multicast instances
throughout the Internet is described, as well.This document uses the terminology as defined for the multicast
protocols ,,,,. In addition,
the following terms will be used.A Group Address is a routing
identifier. It represents a technological identifier and thus
reflects the distribution technology in use. Multicast packet
forwarding is based on this ID.A Group Name is an application identifier
that is used by applications to manage (e.g., join/leave and
send/receive) a multicast group. The Group Name does not imply any
distribution technologies but represents a logical identifier.A Multicast Namespace is a
collection of designators (i.e., names or addresses) for groups that
share a common syntax. Typical instances of namespaces are IPv4 or
IPv6 multicast addresses, overlay group ids, group names defined on
the application layer (e.g., SIP or Email), or some human readable
strings.A Multicast Domain accommodates
nodes and routers of a common, single multicast forwarding
technology and is bound to a single namespace.An Interface is a forwarding instance of a
distribution technology on a given node. For example, the IP
interface 192.168.1.1 at an IPv4 host.An Inter-domain
Multicast Gateway (IMG) is an entity that interconnects different
multicast domains. Its objective is to forward data between these
domains, e.g., between IP layer and overlay multicast.The default use case addressed in this memo targets at applications
that participate in a group by using a common identifier taken from
some common namespace. Programmers should be able to transparently use
this identifier in their program without the need to consider a
deployment status in target domains. Aided by gateways and, where
available, by a node-specific multicast middleware, applications shall
be enabled to establish group communication, even if resident in
domains that are not connected by a common multicast service
technology.This document covers the following two general scenarios:Multicast domains running the same multicast technology but
remaining isolated, possibly only connected by network layer
unicast.Multicast domains running different multicast technologies, but
hosting nodes that are members of the same multicast group.It is assumed throughout the document that the domain composition,
as well as the node attachement to a specific technology remain
unchanged during a multicast session.Multicast applications may use a group communication stack to
deliver and receive multicast data. This group communication stack
exhibits two tasks:It provides an extended API that supports a common multicast
interface with namespace support.It bridges data between different multicast technologies.The group communication API consists of three parts: (1)
Send/Receive Calls, (2) Socket Options, and (3) Service Calls. (1)
provides a minimal API to initiate a multicast socket, send and
receive multicast data in a technology-transparent fashion. (2) allows
for the configuration of the multicast socket, i.e., setting path
length and associate interfaces explicitly. (3) returns internal
multicast states per interface such as the multicast groups under
subscription.The general procedure to initiate multicast communication is the
following:An application opens a multicast socket.An application subscribes/leaves/sends to a logical group
identifier.A function maps the logical group ID (Group Name) to a
technical group ID (Group Address).The technical group ID is allocated or revised if already in
use.The multicast socket describes a group communication channel
composed of one or multiple interfaces. A socket may be created
without explicit interface association by the application, which
leaves the choice of the underlying forwarding technology to the group
communication stack. However, an application may also bind the socket
to one or multiple dedicated interfaces, which predefines the
forwarding technology and the namespace(s) of the Group
Address(es).Applications are not required to maintain states for Group
Addresses. The group communication stack accounts for the mapping of
the Group Name to the Group Address(es) and vice versa. Multicast data
passed to the application will be augmented by the corresponding Group
Name. Multiple multicast subscriptions thus can be conducted on a
single multicast socket without the need for Group Name encoding at
the application side.Hosts may support several multicast protocols. The group
communication stack discovers available multicast-enabled
communication interfaces. It provides a minimal hybrid function that
bridges data between different interfaces and multicast domains.
Details of service discovery are out-of-scope of this document.The extended multicast functions can be implemented by a
middleware, for example.A mapping is required between a Group Name and the Group Address
space, as well as between Group Addresses in different namespaces.Two (or more) identifiers in different namespaces may belong tothe same multicast channel (i.e., same technical ID).different multicast channels (i.e., different technical
IDs).This decision can be solved based on invertible mappings. However,
the application of such functions depends on the cardinality of the
namespaces and thus does not hold in general. A large identifier space
(e.g., IPv6) cannot obviously be mapped to a smaller set (e.g.,
IPv4).A mapping can be realized by embedding smaller in larger namespaces
or selecting an arbitrary, unused ID in the target space. The relation
between logical and technical ID is stored based on a mapping service
(e.g., DHT). The middleware thus queries the mapping service first,
and creates an new technical group ID only if there is no identifier
available for the namespace in use. The Group Name is associated with
one or more Group Addresses, which belong to different namespaces.
Depending on the scope of the mapping service, it ensures a consistent
use of the technical ID in a local or global domain.All group members subscribe to the same Group Name within the same
namespace.The Group Name is used by applications to identify groups. It hides
the deployed technology employed to distribute data. In contrast to
this, multicast forwarding operates on Group Addresses. Although both
identifiers may be identical in symbols, they carry different meaning.
They may also belong to different namespaces. The namespace of the
Group Address reflects the routing technology, and the namespace of
the Group Name represents the context in which the application
operates.A multicast socket (IPv4/v6 interface) can be used by multiple
logical multicast IDs from different namespaces (IPv4-group address,
IPv6-group address). In practice, a library that implements the
defined API would provide high-level data types to the application
similar to the current socket API (e.g., InetAddress in Java). Using
this data type would implicitly determine the namespace.To reflect namespace specific treatment for applications,
identifiers in API calls are represented by URIs. An implementation of
the API may provide convenience functions that detect the namespace of
a Group Name (e.g., InetAddress instead of Inet6Address and
Inet4Address). Details of automatic identifcation is out-of-scope of
this document.is any kind of Group Address or Group Name that
follows the syntax defined in . For example,
ipv4://224.1.2.3:5000 and sip://news@cnn.com.denotes the interface and thus the layer
and instance on which the corresponding call will be effective.
This may be unspecified to leave the decicision to the group
communication stack.references on an instance of a
multicast socket.This
call initiates a multicast socket and provides the application
programmer with a corresponding handle. If no interfaces will be
assigned based on the call, the default interface will be selected
and associated with the socket. The call may return an error code
in the case of failures, e.g., due to a non-operational
middleware.This
operation initiates a group subscription. Depending on the
interfaces that are associated with the socket , this may result
in an IGMP/MLD report or overlay subscriptions.This
operation results in an unsubscription for the given address.This
call passes multicast data for a Multicast Name g from the
application to the multicast socket.This
call passes multicast data and the corresponding Group Name g to
the application.This call
returns a list of all available multicast communication interfaces
at the current host.This
call adds a distribution channel to the socket. This may be an
overlay or underlay interface, e.g., IPv6 or DHT. Multiple
interfaces of the same technology may be associated with the
socket.This
call removes an interface from the socket.This function configures
the maximum hop count for the socket h a multicast message is
allowed to traverse.This
operation returns all registered multicast groups. The information
can be provided by group management or routing protocols. The
return values distinguish between sender and listener states.This
function can be invoked to get the set of multicast routing
neighbors.This function
returns true, if the host has the role of a designated forwarder
or querier. Such an information is provided by almost all
multicast protocols to handle packet duplication, if multiple
multicast instances serve on the same subnet.This
upcall is invoked to inform a group service about a change of
listener states for a group. This is the result of receiver new
subscriptions or leaves. The group service may call groupSet to
get updated information.In this section, we describe the functional details of the API and
the middleware.TODOGroup Name to Group Address, SSM/ASM TODOMulticast Names and Multicast Addresses are described based on a
URI scheme. The scheme defines a subset of the URI specified in and follows the guidelines in .The multicast URI is defined as follows:scheme "://" group "@" instantiation ":" port "/"
sec-credentialsThe parts of the URI are defined as follows:referes to the specification of the assigned
identifier .identifies the group.identifies the entitiy that generates
the instance of the group (e.g., a SIP domain or a source in
SSM).identifies a specific application at an
instance of a group.used to implement security
credentials (e.g., to authorize a multicast group access).TODOThis document makes no request of IANA.This draft does neither introduce additional messages nor novel
protocol operations. TODOWe would like to thank the HAMcast-team (Dominik Charousset, Gabriel
Hege, Fabian Holler, Alexander Knauf, Sebastian Meiling, and Sebastian
Woelke) at the HAW Hamburg for fruitful discussions.This work is partially supported by the German Federal Ministry of
Education and Research within the HAMcast project, which is part of
G-Lab.This section describes the application of the defined API to
implement an IMG.The following procedure describes a transparent mapping of a
DVMRP-based any source multicast service to another many-to-many
multicast technology.An arbitrary DVMRP router will not
be informed about new receivers, but will learn about new sources
immediately. The concept of DVMRP does not provide any central
multicast instance. Thus, the IMG can be placed anywhere inside the
multicast region, but requires a DVMRP neighbor connectivity. The
group communication stack used by the IMG is enhanced by a DVMRP
implementation. New sources in the underlay will be advertised based
on the DVMRP flooding mechanism and received by the IMG. Based on this
the updateSender() call is triggered. The relay agent initiates a
corresponding join in the native network and forwards the received
source data towards the overlay routing protocol. Depending on the
group states, the data will be distributed to overlay peers.DVMRP establishes source specific multicast trees. Therefore, a
graft message is only visible for DVMRP routers on the path from the
new receiver subnet to the source, but in general not for an IMG. To
overcome this problem, data of multicast senders will be flooded in
the overlay as well as in the underlay. Hence, an IMG has to initiate
an all-group join to the overlay using the namespace extension of the
API. Each IMG is initially required to forward the received overlay
data to the underlay, independent of native multicast receivers.
Subsequent prunes may limit unwanted data distribution thereafter.The following procedure describes a transparent mapping of a
PIM-SM-based any source multicast service to another many-to-many
multicast technology.The Protocol Independent Multicast Sparse Mode (PIM-SM) establishes rendezvous points (RP). These
entities receive listener and source subscriptions of a domain. To be
continuously updated, an IMG has to be co-located with a RP. Whenever
PIM register messages are received, the IMG must signal internally a
new multicast source using updateSender(). Subsequently, the IMG joins
the group and a shared tree between the RP and the sources will be
established, which may change to a source specific tree after a
sufficient number of data has been delivered. Source traffic will be
forwarded to the RP based on the IMG join, even if there are no
further receivers in the native multicast domain. Designated routers
of a PIM-domain send receiver subscriptions towards the PIM-SM RP. The
reception of such messages invokes the updateListener() call at the
IMG, which initiates a join towards the overlay routing protocol.
Overlay multicast data arriving at the IMG will then transparently be
forwarded in the underlay network and distributed through the RP
instance.The following procedure describes a transparent mapping of a
PIM-SSM-based source specific multicast service to another one-to-many
multicast technology.PIM Source Specific Multicast (PIM-SSM) is defined as part of
PIM-SM and admits source specific joins (S,G) according to the source
specific host group model . A multicast
distribution tree can be established without the assistance of a
rendezvous point.Sources are not advertised within a PIM-SSM domain. Consequently,
an IMG cannot anticipate the local join inside a sender domain and
deliver a priori the multicast data to the overlay instance. If an IMG
of a receiver domain initiates a group subscription via the overlay
routing protocol, relaying multicast data fails, as data are not
available at the overlay instance. The IMG instance of the receiver
domain, thus, has to locate the IMG instance of the source domain to
trigger the corresponding join. In the sense of PIM-SSM, the signaling
should not be flooded in underlay and overlay.One solution could be to intercept the subscription at both, source
and receiver sites: To monitor multicast receiver subscriptions
(updateListener()) in the underlay, the IMG is placed on path towards
the source, e.g., at a domain border router. This router intercepts
join messages and extracts the unicast source address S, initializing
an IMG specific join to S via regular unicast. Multicast data arriving
at the IMG of the sender domain can be distributed via the overlay.
Discovering the IMG of a multicast sender domain may be implemented
analogously to AMT by anycast.
Consequently, the source address S of the group (S,G) should be built
based on an anycast prefix. The corresponding IMG anycast address for
a source domain is then derived from the prefix of S.The following procedure describes a transparent mapping of a
BIDIR-PIM-based any source multicast service to another many-to-many
multicast technology.Bidirectional PIM is a variant of
PIM-SM. In contrast to PIM-SM, the protocol pre-establishes
bidirectional shared trees per group, connecting multicast sources and
receivers. The rendezvous points are virtualized in BIDIR-PIM as an
address to identify on-tree directions (up and down). However, routers
with the best link towards the (virtualized) rendezvous point address
are selected as designated forwarders for a link-local domain and
represent the actual distribution tree. The IMG is to be placed at the
RP-link, where the rendezvous point address is located. As source data
in either cases will be transmitted to the rendezvous point address,
the BIDIR-PIM instance of the IMG receives the data and can internally
signal new senders towards the stack via updateSender(). The first
receiver subscription for a new group within a BIDIR-PIM domain needs
to be transmitted to the RP to establish the first branching point.
Using the updateListener() invocation, an IMG will thereby be informed
about group requests from its domain, which are then delegated to the
overlay.Changes since draft-waehlisch-sam-common-api-01TODO