Core Protocols in the Internet Protocol
SuiteCisco SystemsSanta Barbara93117CaliforniaUSAfred@cisco.com
General
This note attempts to identify the core of the Internet Protocol
Suite. The target audience is NIST, in the Smart Grid discussion, as
they have requested guidance on how to profile the Internet Protocol
Suite. In general, that would mean selecting what they need from the
picture presented here.In the discussion of the Smart Grid, a question has arisen as to what
the "core" protocols of the Internet Protocol Suite are. In this note, I
will attempt to identify the structure of the Internet Protocol Suite
and the key protocols that should be considered as critical in
integrating Smart Grid devices into an IP-based infrastructure. In many
cases, the protocols are options - one might choose, for example, TCP,
SCTP, DCCP, or some other transport, or use UDP as a label and build the
transport into the application itself. In the Transport layer,
therefore, one is not limited to exactly one of those, nor is one
required to implement them all. One should, however, pick the right one
for the purpose one intends. This kind of discussion will be had at
every layer.The set of protocols defined in this document focus on the use of the
IP Protocol Suite in end systems, also known as hosts. In the Smart
Grid, these end systems will be various devices such as power meters,
sensors and actuators. These end systems can leverage infrastructure
built on networking components using the IP Protocol Suite, which have
well-proven implementations and deployments in the Internet.IETF participants in the Smart Grid discussion have been wary of the
desire to write a "profile", repeatedly expressed. The IETF is all about
interoperability, and in our experience attempts to "profile" protocols
and architectures has resulted in a failure to interoperate. Examples of
such failures abound. In IETF experience, writing a conforming and
interoperable implementation of the right set of protocols works.
Selecting some options and deselecting others within a defined protocol,
however, is a dangerous course of action. So while this document is
clearly a step in the direction of writing a "Smart Grid Profile", such
a profile should in our opinion be a selected set of protocols, not of
protocol subsets.For its own purposes, the IETF has written several documents that
describe its expectations regarding implementations of the Internet
Protocol Suite. These include: Requirements for Internet Hosts -
Communication Layers,Requirements for Internet Hosts -
Application and Support,Requirements for IP Version 4
Routers, andIPv6 Node Requirements,At this writing, RFC 4294 is in the process of being updated, in
.This document will read like an annotated list of RFCs. That is
because that is what it is.Before listing a list of protocols, it would be well to lay out how
they relate to each other. In this section, we will discuss the layered
architecture of the Internet Protocol Suite and the jobs of the various
layers and their protocols.The Internet Architecture uses the definitions and language of the
ISO Open System Interconnect Reference Model, as shown in . It actually predates that model, and as a
result uses some different words - an "end system" is generally called
a "host", and an "intermediate system" is more generally called an
"internet gateway" or "router". But the fundamental concepts are
essentially the same.In implementation, the Application, Presentation, and Session
layers are generally compressed into a single entity, which the IETF
calls "the application". The SNMP protocol, for example, describes
an application (a management application or a client that it
communicates with) that encodes its data in a profile of ASN.1 (a
presentation layer) and engages in a session to manage a network
element. In the Internet, therefore, the distinction between these
layers exists but is not generally highlighted.The term "transport" is perhaps among the most confusing words in
the communication architecture, because people with various
backgrounds use it to refer to "the layer below that which I am
interested in, which gets my data to my peer". In these contexts,
optical fiber and other physical layers, the Internet Protocol or
other networked protocols, and in some cases application layer
protocols like HTTP are referred to as "the transport".In the Internet context, the "transport" is the lowest layer that
travels end-to-end unmodified, and is responsible for end-to-end
data delivery services. At minimum these include the ability to
multiplex several applications on one IP address, and may also
include the delivery of data (either as a stream of messages or a
stream of bytes) in a stated sequence with stated expectations
regarding delivery rate and loss. TCP, for example, will reduce rate
to avoid loss, while DCCP accepts some level of loss if necessary to
maintain timeliness.The network layer is nominally that which identifies a remote
destination and gets data to it. In connection-oriented networking,
such as MPLS or ATM, a path (one of many "little tubes") is set up
once, and data is delivered through it. In connectionless
("datagram") networks, which include Ethernet and IP among others,
each datagram contains the addresses of both the source and
destination devices, and the network is responsible to deliver
it.IPv4 and IPv6, each of which is called the Internet Protocol,
are connectionless ("datagram") architectures. They are designed
as common elements that interconnect network elements across a
network of lower layer networks. In addition to the basic service
of identifying a datagram's source and destination, they offer
services to fragment and reassemble datagrams when necessary,
assist in diagnosis of network failures, and carry additional
information necessary in special cases.The Internet Layer provides a uniform network abstraction or
virtual network that hides the differences between different
network technologies. This is the layer that allows diverse
networks such as Ethernet, 802.15.4, etc. to be combined into a
uniform IP network. New network technologies can be introduced
into the IP Protocol Suite by defining how IP is carried over
those technologies, leaving the other layers of the IP Protocol
Suite and applications that use those protocol unchanged.The network layer is recursively subdivided as needed. For
various reasons, IP may be carried in virtual private networks
across more public networks using tunneling technologies like
IP-in-IP or GRE, traffic engineered in circuit networks such as
MPLS, GMPLS, or ATM, and distributed across local wireless (IEEE
802.11, 802.15.4, or 802.16) and switched Ethernet (IEEE
802.3).At the lowest layer of the architecture, we encode digital data
in messages onto appropriate physical media. While the IETF
specifies algorithms for carrying IPv4 and IPv6 on such media, it
rarely actually defines the media - it happily uses specifications
from IEEE, ITU, and other sources.It is popular to complain about the security of the Internet; that
said, solutions exist but are often left unused. As with automobile
seat belts, they are of more value when actively used. Security
designs attempt to mitigate a set of known threats at a specified
cost; addressing security issues requires first a threat analysis and
assessment and a set of mitigations appropriate to the threats. Since
we have threats at every layer, we should expect to find mitigations
at every layer.At the physical and data link layers, threats involve physical
attacks on the network - the effects of backhoes, deterioration of
physical media, and various kinds of environmental noise.
Radio-based networks are subject to signal fade due to distance,
interference, and environmental factors; it is widely noted that
IEEE 802.15.4 networks frequently place a metal ground plate between
the meter and the device that manages it. Fiber was at one time
deployed because it was believed to be untappable; we have since
learned to tap it by bending the fiber and collecting incidental
light, and we have learned about backhoes. So now some installations
encase fiber optic cable in a pressurized sheath, both to quickly
identify the location of a cut and to make it more difficult to
tap.While there are protocol behaviors that can detect certain
classes of physical faults, such as keep-alive exchanges, physical
security is generally not a protocol problem.At the transport and application layers, and in lower layer
networks where dynamic connectivity like ATM SVCs or "dial"
connectivity is in use, there tend to be several different classes
of authentication/authorization requirements. One must Verify that the peers one exchanges data with are appropriate
partners; this generally means knowing "who" they are and that
they have a "need to know" or are trusted sources.Verify that information that appears to be from a trusted
peer is in fact from that peer.Validate the content of the data exchanged; it must conform
to the rules of the exchange.One must also defend the channel against denial of service
attacks.In other words, there is a need to secure the channel that
carries a message, and there is a need to secure the exchanges, both
by knowing the source of the information and to have proof of its
validity. Three examples suffice to illustrate the challenges.One common attack is to bombard a transport session (an
application's channel) with reset messages. If the attacker is
lucky, he might cause the session to fail. Including information in
the transport header or a related protocol like IPsec or TLS that
identifies the right messages and facilitates speedy discard of the
rest can mitigate this.Another common attack involves unauthorized communication with a
router or a service. For example, an unauthorized party might try to
join the routing system. One wants the ISP's router, before
accepting routing information from a new peer, todemand identification from the new peer,verify that the peer is in fact who it claims to be, andverify that it is authorized to carry on the exchange.More generally, in securing the channel, one wants to verify that
messages putatively received from a peer were in fact received from
the peer, and given that they are, to only carry on transactions
with peers that one trusts. This is analogous to how one responds to
a salesman at the front door - one asks who the salesman represents,
seeks a credential as proof, and then asks one self whether one
wants to deal with that company. Only if all indications are
positive does one carry on a transaction.Unfortunately, even trusted peers can be the purveyors of
incorrect or malicious content; having secured the channel, one also
wants to secure the information exchanged through the channel. In
electronic mail and other database exchanges, it may be necessary to
be able to verify the identity of the sender and the correctness of
the content long after the information exchange has occurred - for
example, if a contract is exchanged that is secured by digital
signatures, one will wish to be able to verify those signatures at
least throughout the lifetime of the contract, and probably a long
time after that.The third "A" in "AAA" is Accounting. This service is especially
important for Internet Service Providers; the related service of
auditing is important for enterprises. RADIUS and DIAMETER are
commonly used to realize these services.At several layers, there is a question of confidentiality. If one
is putting one's credit card in a transaction, one wants application
layer privacy, which might be supplied by an encrypting application
or transport layer protocol. If one is trying to hide one's network
structure, one might additionally want to encrypt the network layer
header.While these are not critical to the design of a specific system,
they are important to running a network. We therefore bring them
up.While not critical to running a network, certain functions are
made a lot easier if numeric addresses can be replaced with mnemonic
names. This facilitates renumbering of networks, which happens, and
generally improves the manageability and serviceability of the
network. DNS has a set of security extensions called DNSSEC, which
can be used to provide strong cryptographic authentication to that
protocol.Network management has proven to be a difficult problem; there
are many solutions, and each has proponents with solid arguments for
their viewpoint. In the IETF, we have two major network management
solutions for device operation: SNMP, which is ASN.1-encoded and is
primarily used for monitoring of system variables in a polled
architecture, and NetConf, which is XML-encoded and primarily used
for device configuration.Another aspect of network management is the initial provisioning
and configuration of hosts. Address assignment and other
configuration is discussed in . Smart
Grid deployments will require additional identity authentication and
authorization as well as other provisioning and configuration that
may not be within the scope of DHCP and Neighbor Discovery. While
the IP Protocol Suite does not have specific solutions for secure
provisioning and configuration, these problems have been solved
using IP protocols in specifications such as DOCSIS 3.0.In this section, having briefly laid out the architecture and some of
the problems that the architecture tries to address, we introduce
specific protocols that might be appropriate to various use cases. In
each place, the options are in the protocols used - one wants to select
the right privacy, AAA, transport, and network solutions in each
case.As noted, a key consideration in security solutions is a good
threat analysis coupled with appropriate mitigations at each
layer.In the Internet Protocol Suite there are several approaches to
AAA issues; generally, one chooses one of them for a purpose. As
they have different attack surfaces and protection domains, they
require some thought in application. Two important ones are the IP
Security Architecture, which protects IP datagrams, and Transport
Layer Security, which protects the information that the Transport
delivers.The Security Architecture for the Internet
Protocol is a set of control and data protocols that are
implemented between IPv4 and its Transport layer, or in IPv6's
Security extension header. It allows transport layer sessions (which
underlie applications) to communicate in a way that is designed to
prevent eavesdropping, tampering, or message forgery. The
architecture is spelled out in a number of additional specifications
for specific components: the Encapsulating
Security Payload (ESP), the Internet
Security Association and Key Management Protocol (ISAKMP),
Internet Key Exchange (IKEv2) , Cryptographic Algorithms, and the use of
Advanced Encryption Standard (AES)
.In the transport mode, IPsec ESP encrypts the transport layer and
the application data. In the tunnel mode, which is frequently used
for Virtual Private Networks, one also encrypts the Internet
Protocol, and encapsulates the encrypted data inside a second IP
header directed to the intended decryptor.Transport Layer Security and Datagram Transport Layer Security are mechanisms that travel
within the transport layer PDU, meaning that they readily traverse
network address translators and secure the information exchanges
without securing the datagrams exchanged or the transport layer
itself. Each allows client/server applications to communicate in a
way that is designed to prevent eavesdropping, tampering, or message
forgery.The S/MIME specification was originally specified as
an extension to SMTP Mail to provide evidence that the putative
sender of an email message in fact sent it, and that the content
received was in fact the content that was sent. As its name
suggests, by extension this is a way of securing any object that can
be exchanged, by any means, and has become one of the most common
ways to secure an object.Other approaches also exist, such as the use of digital
signatures on XML-encoded files, as jointly standardized by W3C and
the IETF .Here we mention both IPv4 and IPv6. The reader is warned: IPv4 is
running out of address space, and IPv6 has positive reasons that one
might choose it apart from the IPv6 space such as the address
autoconfiguration facility and its ability to support an arbitrarily
large number of hosts in a subnet. As such, the IETF recommends that
one always choose IPv6 support, and additionally choose IPv4 support
in the near term.IPv4, with the Internet Control Message Protocol,
constitutes the traditional protocol implemented throughout the
Internet. IPv4 provides for transmission of datagrams from source to
destination hosts, which are identified by fixed length
addresses.IPv4 also provides for fragmentation and reassembly of long
datagrams, if necessary, for transmission through "small packet"
networks. ICMP, which is a separate protocol implemented along with
IPv4, enables the network to report errors and other issues to hosts
that originate problematic datagrams.IPv4 addresses are administratively assigned, usually using
automated methods, and assigned using the Dynamic Host Configuration Protocol
(DHCP). On most interface types, neighboring equipment
identify each other's addresses using Address Resolution Protocol (ARP).Routing for the IPv4 Internet is done by routing applications
that exchange connectivity information and build semi-static
destination routing databases. If a datagram is directed to a
given destination address, the address is looked up in the routing
database, and the most specific ("longest") prefix found that
contains it is used to identify the next hop router, or the end
system it will be delivered to. This is not generally implemented
on hosts, although it can be; generally, a host sends datagrams to
a router on its local network, and the router carries out the
intent.IETF specified routing protocols include RIP Version 2, OSI
IS-IS for IPv4, OSPF Version
2, and BGP-4. Active research
exists in mobile ad hoc routing and other routing paradigms; these
result in new protocols and modified forwarding paradigms.IPv4 was originally specified as a unicast (point to point)
protocol, and was extended to support multicast in . This uses the Internet Group Management Protocol to enable applications to join multicast
groups, and for most applications uses Source-Specific Multicast for routing and
delivery of multicast messages.An experiment carried out in IPv4 that is not core to the
architecture but may be of interest in the Smart Grid is the
development of so-called "Reliable Multicast". This is "so-called"
because it is not "reliable" in the strict sense of the word - it
is perhaps better described as "enhanced reliability". A best
effort network by definition can lose traffic, duplicate it, or
reorder it, something as true for multicast as for unicast.
Research in "Reliable Multicast" technology intends to improve the
probability of delivery of multicast traffic.In that research, the IETF imposed guidelines on the research community
regarding what was desirable. Important results from that research
include a number of papers and several proprietary protocols
including some that have been used in support of business
operations. RFCs in the area include The
Use of Forward Error Correction (FEC) in Reliable Multicast
, the Negative-acknowledgment
(NACK)-Oriented Reliable Multicast (NORM) Protocol , and
the Selectively Reliable Multicast
Protocol (SRMP) . These are considered experimental.IPv6, with the Internet Control Message Protocol "v6",
constitutes the next generation protocol for the Internet. IPv6
provides for transmission of datagrams from source to destination
hosts, which are identified by fixed length addresses.IPv6 also provides for fragmentation and reassembly of long
datagrams by the originating host, if necessary, for transmission
through "small packet" networks. ICMPv6, which is a separate
protocol implemented along with IPv6, enables the network to report
errors and other issues to hosts that originate problematic
datagrams.An IPv6 Address may be
administratively assigned using DHCPv6 in a manner similar to the way IPv4
addresses are, but may also be autoconfigured, facilitating
network management. Autoconfiguration procedures are defined in
and .
IPv6 neighbors identify each other's addresses using either Neighbor Discovery (ND) or SEcure Neighbor Discovery (SEND).Routing for the IPv6 Internet is done by routing applications
that exchange connectivity information and build semi-static
destination routing databases. If a datagram is directed to a
given destination address, the address is looked up in the routing
database, and the most specific ("longest") prefix found that
contains it is used to identify the next hop router, or the end
system it will be delivered to. This is not generally implemented
on hosts, although it can be; generally, a host sends datagrams to
a router on its local network, and the router carries out the
intent.IETF specified routing protocols include RIP for IPv6, IS-IS
for IPv6, OSPF for IPv6, and
BGP-4 for IPv6. Active research
exists in mobile ad hoc routing, routing in low power networks
(sensors and smart grids) and other routing paradigms; these
result in new protocols and modified forwarding paradigms.From its initial design, IPv6 has specified both unicast and
multicast datagram exchange. This uses the Multicast Listener Discovery Protocol
(MLDv2) to enable
applications to join multicast groups, and for most applications
uses Source-Specific Multicast for
routing and delivery of multicast messages.The IPv6 over Low-Power Wireless
Personal Area Networks RFC addresses IPv6 header
compression and subnet architecture in at least some sensor
networks, and may be appropriate to the Smart Grid AMI or other
sensor domains.The mechanisms experimentally developed for reliable multicast
in IPv4, discussed in , can be
used in IPv6 as well.In general, the layered architecture enables the Internet
Protocol Suite to run over any appropriate layer 2 architecture;
with tongue in cheek, specifications have been written and
demonstrated to work for the carriage of IP
by Carrier Pigeon (perhaps the
most common carrier known to man) and on barbed wire. The ability to change the link
or physical layer without having to rethink the network layer,
transports, or applications has been a great benefit in the
Internet.Examples of link layer adaptation technology include: IPv4 has run on each link
layer environment that uses the Ethernet header (which is to say
10 and 100 MBPS wired Ethernet, 1 and 10 GBPS wired Ethernet,
and the various versions of IEEE 802.11) using . IPv6 does the same using .The IETF has defined a serial line protocol,
the Point-to-Point Protocol (PPP),
that uses HDLC (bit-synchronous or byte synchronous) framing.
The IPv4 adaptation specification is , and the IPv6 adaptation specification
is . Current use of this protocol
is in traditional serial lines, authentication exchanges in DSL
networks using PPP Over Ethernet
(PPPoE), and in the Digital Signaling Hierarchy
(generally referred to as Packet-on-SONET/SDH) using PPP over SONET/SDH.The adaptation specification for
IPv6 transmission over IEEE 802.15.4 Networks is .Numerous other adaptation specifications exist, including ATM,
Frame Relay, X.25, other standardized and proprietary LAN
technologies, and others.In this we list several transports: UDP, TCP, SCTP, and DCCP. Of
these, UDP and TCP are best known and most widely used, due to
history. SCTP and DCCP were built for specific purposes more recently
and bear consideration at least for those purposes.Note that if it is appropriate, other transports can also be built.
This is largely a question of requirements.The User Datagram Protocol and the
Lightweight User Datagram Protocol are
properly not "transport" protocols in the sense of "a set of rules
governing the exchange or transmission of data electronically
between devices". They are labels that provide for multiplexing of
applications directly on the IP layer, with transport functionality
embedded in the application.From a historical perspective, one should note that many
simplistic exchange designs have been built using UDP, and many of
them have not worked all that well. The use of UDP really should be
treated as designing a new transport. More generally, advice on the
use of UDP in new applications has been compiled in the Unicast UDP Usage Guidelines for Application
Designers.Datagram Transport Layer Security
can be used to prevent eavesdropping, tampering, or message forgery.
Alternatively, UDP can run over IPsec.TCP is the predominant transport
protocol in use in the Internet, with a long history. It is
"reliable", as the term is used in protocol design: it delivers data
to its peer and provides acknowledgement to the sender, or it dies
trying. It has extensions for Congestion
Control and Explicit Congestion
Notification.The user interface for TCP is a byte stream interface - an
application using TCP might "write" to it several times only to have
the data compacted into a common segment and delivered as such to
its peer. For message-stream interfaces, we generally use the ISO Transport Service on TCP in the application.Transport Layer Security can be
used to prevent eavesdropping, tampering, or message forgery.
Alternatively, TCP can run over IPsec. Additionally, discusses mechanisms similar to SCTP and
DCCP's "cookie" approach that may be used to secure TCP sessions
against flooding attacks.TCP has supported ongoing research since it was written. As a
result, the End to End research group has published a Roadmap for TCP Specification Documents
which will guide expectations in that area.SCTP is a more recent reliable
transport protocol that can be imagined as a TCP-like context
containing multiple separate and independent message streams (as
opposed to TCP's byte streams). The design of SCTP includes
appropriate congestion avoidance behavior and resistance to flooding
and masquerade attacks. As it uses a message stream interface as
opposed to TCP's byte stream interface, it may also be more
appropriate for the ISO Transport Service than RFC 1006/2126.SCTP offers several delivery options. The basic service is
sequential non-duplicated delivery of messages within a stream, for
each stream in use. Since streams are independent, one stream may
pause due to head of line blocking while another stream in the same
session continues to deliver data. In addition, SCTP provides a
mechanism for bypassing the sequenced delivery service. User
messages sent using this mechanism are delivered to the SCTP user as
soon as they are received.SCTP implements a simple "cookie" mechanism intended to limit the
effectiveness of flooding attacks by mutual authentication. This
demonstrates that the application is connected to the same peer, but
does not identify the peer. Mechanisms also exist for Dynamic Address Reconfiguration, enabling
peers to change addresses during the session and yet retain
connectivity. Transport Layer Security
can be used to prevent eavesdropping, tampering, or message forgery.
Alternatively, SCTP can run over IPsec.DCCP is an "unreliable" transport
protocol (e.g., one that does not guarantee message delivery) that
provides bidirectional unicast connections of congestion-controlled
unreliable datagrams. DCCP is suitable for applications that
transfer fairly large amounts of data and that can benefit from
control over the tradeoff between timeliness and reliability.DCCP implements a simple "cookie" mechanism intended to limit the
effectiveness of flooding attacks by mutual authentication. This
demonstrates that the application is connected to the same peer, but
does not identify the peer. Datagram
Transport Layer Security can be used to prevent
eavesdropping, tampering, or message forgery. Alternatively, DCCP
can run over IPsec.To facilitate network management and operations, the Internet
Community has defined the Domain Name System
(DNS). Names are hierarchical:
a name like example.com is found registered with a .com registrar,
and within the associated network other names like
baldur.cincinatti.example.com can be defined, with obvious
hierarchy. Security extensions, which all a registry to sign the
records it contains and as a result demonstrate their authenticity,
are defined by the DNS Security Extensions .Similarly unrequired but useful is the ability for a device to
update its own DNS record. One could imagine a sensor, for example,
that is using Stateless Address
Autoconfiguration to create an address to associate it with a
name using DNS Dynamic Update or DNS Secure Dynamic Update.As discussed in and , IPv6 address assignment can be accomplished
using autoconfiguration but can also be accomplished using DHCP or DHCPv6. The best argument for the use of
autoconfiguration is a large number of systems that require little
more than a random number as an address; the argument for DHCP is
administrative control.There are other parameters that may need to be allocated to
hosts, and these do require administrative configuration; examples
include the address of one's DNS server, keys if Secure DNS is in
use, and others.The Network Time Protocol was originally designed by Dave Mills
of the University of Delaware and CSNET, for the purpose of
implementing a temporal metric in the Fuzzball Routing Protocol and
generally coordinating time experiments. The current versions of the
time protocol are the Network Time
Protocol, which is designed for synchronization to within a
few microseconds, and the Simple Network Time
Protocol which is used to set real time clocks to within a
few milliseconds. The former is more precise, but relies on frequent
exchanges; the latter is less precise and lower overhead.NTP is currently being updated in .The Session Initiation
Protocol was originally developed to manage
Voice-on-IP and Video-on-IP sessions on the Internet, as an open
source alternative to H.323. Via extensions, it is generally used
for session setup for a variety of purposes and for very quick
management exchanges.The Internet was originally structured in such a way that any host
could directly connect to any other host for which it could determine an
IP address. That was very quickly found to have issues, and folks found
ways to change that. To understand the implications, one must
understand, at a high level, the business structure of the Internet.The Internet, whose name implies that it is a network of networks,
may be understood as a number of interconnected businesses. Central to
its business structure are the networks that provide connectivity to
other networks, called "Transit Providers". These networks sell bulk
bandwidth to each other, and to other networks as customers. Around the
periphery of these networks, one finds companies, schools, and other
networks that provide services directly to individuals. These might
generally be divided into "Enterprise Networks" and "Access Networks";
Enterprise networks provide "free" connectivity to their own employees
or members, and also provide them a set of services including electronic
mail, web services, and so on. Access Networks sell broadband
connectivity (DSL, Cable Modem, 802.11 wireless or 3GPP wireless), or
"dial" services including PSTN dial-up and ISDN, to subscribers. The
subscribers are typically either residential or small office/home office
(SOHO) customers. Residential customers are generally entirely dependent
on their access provider for all services, while a SOHO buys some
services from the access provider and may provide others for itself.
Networks that sell transit services to nobody else - SOHO, residential,
and enterprise networks - are generally refereed to as "edge networks";
Transit Networks are considered to be part of the "core" of the
Internet, and access networks are between the two. This general
structure is depicted in .A specific example is shown in a traceroute from the author's home to
a school he can see nearby. Internet connectivity in passes through The author's home,Cox Communications, an Access Network using Cable Modem
technology,TransitRail, a commodity peering service for research and
education (R&E) networks,Corporation for Education Network Initiatives in California
(CENIC), a transit provider for educational networks, andthe University of California at Santa Barbara, which in this
context might be viewed as an access network for its students and
faculty or as an enterprise network.Another specific example could be shown in a traceroute from the
author's home to his employer. Internet connectivity in that case uses a
Virtual Private Network (VPN tunnel) from the author's home, crossing
Cox Cable (an Access Network) and Pacific Bell (a Transit Network), and
terminating in Cisco Systems (an Enterprise Network); a traceroute of
the path doesn't show that as it is invisible within the VPN and the
contents of the VPN are invisible, due to encryption, to the networks on
the path. Instead, the traceroute in
is entirely within Cisco's internal network.In both cases, it will be observed that the author's home internally
uses address space from the Address Allocation
for Private Internets, and other networks generally use public
address space. It will also be observed that on entry to UCSB, the
traceroute in terminates before
arriving at the target.Three middleware technologies are in obvious use here. These are the
use of a firewall, a Network Address Translator (NAT), and a Virtual
Private Network (VPN).Firewalls are generally sold as, and considered, a security
technology. A firewall imposes a border between two administrative
domains, which are usually a residential, SOHO, or enterprise network
and its access or transit provider. In its essence, a firewall is a data
diode, imposing a policy on what sessions may pass between a protected
domain and the rest of the Internet. Simple policies generally permit
sessions to be originated from the protected network but not from the
outside; more complex policies may permit additional sessions from the
outside, as electronic mail to a mail server or a web session to a web
server, and may prevent certain applications from global access even
though they are originated from the inside. Firewalls are controversial
in the Internet community; network managers often insist on them simply
because they impose a boundary; others point out that their value as a
security solution is debatable, as most attacks come from behind the
firewall and application layer attacks such as viruses carried in email
or Active X are invisible to them. In general, as a security solution,
they are justified as a defense in depth; while the end system must in
the end be responsible for its own security, a firewall can inhibit or
prevent certain kinds of attacks from certain quarters such as the
consumption of CPU time on a critical server. Key documents describing
firewall technology and the issues it poses include:IP Multicast and FirewallsBenchmarking Terminology for Firewall
PerformanceBehavior of and Requirements for Internet
FirewallsBenchmarking Methodology for Firewall
PerformanceMobile IPv6 and Firewalls: Problem
StatementNAT and Firewall Traversal Issues of Host
Identity Protocol CommunicationNetwork Address Translation is a technology that was developed in
response to ISP behaviors in the mid-1990's; when was published, many ISPs started handing out
single or small numbers of addresses, and edge networks were forced to
translate. In time, this became considered a good thing, or at least not
a bad thing; it amplified the public address space, and it was sold as
if it were a firewall. It of course is not; while traditional dynamic
NATs only translate between internal and external session address/aport
tuples during the detected duration of the session, that session state
may exist in the network much longer than it exists on the end system,
and as a result constitutes an attack vector. The design, value, and
limitations of network address translation are described in: IP Network Address Translator Terminology
and ConsiderationsTraditional IP Network Address
TranslatorProtocol Complications with the IP Network
Address TranslatorNetwork Address Translator Friendly
Application Design GuidelinesIAB Considerations for Network Address
TranslationIPsec-Network Address Translation
Compatibility RequirementsNetwork Address Translation Behavioral
Requirements for Unicast UDPState of Peer-to-Peer Communication across
Network Address TranslatorsIP Multicast Requirements for a Network
Address Translator and a Network Address Port TranslatorVirtual Private Networks come in many forms; what they have in common
is that they are generally tunneled over the internet backbone, so that
as in , connectivity appears to be
entirely within the edge network although it is in fact across a service
provider's network. Examples include IPsec tunnel-mode encrypted
tunnels, IP-in-IP or GRE tunnels, and MPLS LSPs.This memo asks the IANA for no new parameters.Note to RFC Editor: This section will have served its purpose if it
correctly tells IANA that no new assignments or registries are required,
or if those assignments or registries are created during the RFC
publication process. From the author"s perspective, it may therefore be
removed upon publication as an RFC at the RFC Editor's discretion.Security is addressed in some detail in and .Review comments were made by Andrew Yourtchenko, Ashok Narayanan,
Bernie Volz, Chris Lonvick, Dave McGrew, Dave Oran, David Su, Hemant
Singh, James Polk, John Meylor, Joseph Salowey, Julien Abeille, Kerry
Lynn, Magnus Westerlund, Murtaza Chiba, Paul Duffy, Paul Hoffman, Ralph
Droms, Russ White, and Toerless Eckert. Dave McGrew and Ralph Droms
suggested text.DOCSIS 3.0 MAC and Upper Layer Protocols Interface
Specification, CM-SP-MULPIv3.0-I10-090529Cisco SystemsEthernet over Barbed Wire, Arcnet, 100MB Token Ring,
100Base-VGAnylan and iSCSI ...Cisco Systems