Internet Engineering Task Force R. G. Cole
INTERNET-DRAFT D. H. Shur
draft-ietf-ipatm-framework-doc-02 AT&T Bell Laboratories
C. Villamizar
ANS
April 11, 1995
IP over ATM: A Framework Document
Status of this Memo
This document is an internet draft. 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.
Internet Drafts may be updated, replaced, or obsoleted by other
documents at any time. It is not appropriate to use Internet Drafts
as reference material or to cite them other than as a "working draft"
or "work in progress". Please check the lid-abstracts.txt listing
contained in the internet-drafts shadow directories on nic.ddn.mil,
nnsc.nsf.net, nic.nordu.net, ftp.nisc.src.com, or munnari.oz.au to
learn the current status of any Internet Draft.
Abstract
The discussions of the IP over ATM working group over the last several
years have produced a diverse set of proposals, some of which are
no longer under active consideration. A categorization is provided
for the purpose of focusing discussion on the various proposals
for IP over ATM deemed of primary interest by the IP over ATM
working group. The intent of this framework is to help clarify the
differences between proposals and identify common features in order to
promote convergence to a smaller and more mutually compatible set of
standards. In summary, it is hoped that this document, in classifying
ATM approaches and issues will help to focus the IP over ATM working
group's direction.
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
1 Introduction
The IP over ATM Working Group of the Internet Engineering Task
Force (IETF) is chartered to develop standards for routing and
forwarding IP packets over ATM sub-networks. This document provides
a classification/taxonomy of IP over ATM options and issues and then
describes various proposals in these terms.
The remainder of this memorandum is organized as follows:
o Section 2 defines several terms relating to networking and
internetworking,
o Section 3 discusses the parameters for a taxonomy of the
different ATM models under discussion.
o Section 4 discusses the options for low level encapsulation.
o Section 5 discusses tradeoffs between connection oriented and
connectionless approaches.
o Section 6 discusses the various means of providing direct
connections across IP subnet boundaries,
o Section 7 discusses the proposal to extend IP routing to better
accommodate direct connections across IP subnet boundaries,
o Section 8 identifies several prominent IP over ATM proposals that
have been discussed within the IP over ATM Working Group and
their relationship to the framework described in this document,
o Section 9 addresses the relationship between the documents
developed in the IP over ATM working group and the various models
discussed, and
2 Definitions and Terminology
We define several terms:
A Host or End System (ES): A host delivers/receives IP packets
to/from other systems, but does not relay IP packets.
A Router or Intermediate System (IS): A router delivers/receives
IP packets to/from other systems, and relays IP packets among
systems.
Cole, Shur, Villamizar Expires October 11, 1995 [Page 2]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
IP Subnet: In an IP subnet, all members of the subnet are able
to reach all other members of the subnet directly, without
forwarding to another entity.
Bridged IP Subnet: A bridged IP subnet is one in which two or
more physically disjoint media are made to appear as a single IP
subnet. There are two basic types of bridging, media access
control (MAC) level, and proxy ARP (see section 6). Bridging is
not discussed due to scaling problems associated with including a
large bridged component in a large internetwork.
A Broadcast Subnet: A broadcast network supports an arbitrary
number of hosts and routers and additionally is capable of
transmitting a single IP packet to all of these systems.
A Non-Broadcast Multiple Access (NBMA) Subnet: An NBMA
supports an arbitrary number of hosts and routers but
does not support a convenient multi-destination connectionless
transmission facility, as does a broadcast subnetwork.
A Multicast Capable Subnet: A multicast capable subnet supports
a facility to send a packet which reaches a subset of the
destinations on the subnet. Multicast setup may be sender
initiated, or leaf initiated. ATM UNI 3.0 [4] and UNI 3.1
support only sender initiated while IP supports leaf initiated
join. UNI 4.0 will support leaf initiated join.
An End-to-End path: An end-to-end path consists of two hosts which
can communicate with one another over an arbitrary number of
routers and subnets.
An internetwork: An internetwork (small ``i'') is the concatenation
of networks, often of various different media and lower level
encapsulations, to form an integrated larger network supporting
communication between any of the hosts on any of the component
networks. The Internet (big ``I'') is a specific well known
global concatenation of (over 40,000 at the time of writing)
component networks.
IP forwarding: IP forwarding is the process of receiving a packet
and using a very low overhead decision process, determining how
to forward the packet.
IP routing: IP routing is the exchange of information that takes
place in order to have available the information necessary to
make a correct IP forwarding decision.
IP address resolution: A somewhat static mapping exists between
IP address on the local IP subnet and media address on the
local subnet. This mapping is known as IP address resolution.
An address resolution protocol (ARP) is a protocol supporting
Cole, Shur, Villamizar Expires October 11, 1995 [Page 3]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
address resolution (and usually nothing else).
In order to support end-to-end connectivity, two techniques are used.
One involves allowing direct connectivity across classic IP subnet
boundaries supported by certain NBMA media, which includes ATM. The
other involves IP routing and IP forwarding. In essence, the former
technique is extending IP address resolution beyond the boundaries of
the IP subnet, while the latter is interconnecting IP subnets.
Large internetworks, and in particular the Internet, are unlikely to
be composed of a single media, or a star topology, with a single media
at the center. Within a large network supporting a common media,
typically any large NBMA such as ATM, IP routing and IP forwarding
must always be accommodated if the internetwork is larger than the
NBMA and connected to the the NBMA in multiple places with redundant
diversely located interconnections.
Routing information in a very large internetwork can be quite
dynamic due to the high probability that something somewhere in a
very large internetwork is changing state. The address resolution
space consumption and resource consumption due to state change, or
maintenance of state information is rarely a problem in classic IP
subnets. It can become a problem in large bridged networks or
in proposals that attempt to extend address resolution beyond the
IP subnet. Scaling properties of address resolution and routing
proposals, with respect to state information and state change, must be
considered.
3 Parameters Common to IP Over ATM Proposals
Much of the discussion of IP over ATM has been confounded by
distinctions made between local area networks (LANs), and wide area
networks (WANs) that do not necessarily hold. The distinction
between a LAN, MAN and WAN is a matter of geographic dispersion.
Geographic dispersion affects performance due to increased propagation
delay. However, LANs such as the major Internet traffic interconnect
sites have multiple administrative authorities, currently exclusively
support routers providing transit to multihomed internets, currently
rely on PVCs and static address resolution, and rely heavily on IP
routing. This is not typical of LANs, but emphasizes the point that
prior characterization of LANs do not necessarily hold. Similarly,
WANs such as those under consideration by numerous large IP providers,
do not conform to prior characterizations of ATM WANs.
The following characteristics of the IP over ATM internetwork may be
independent of geographic dispersion (LAN, MAN, or WAN).
Cole, Shur, Villamizar Expires October 11, 1995 [Page 4]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
o The size of the IP over ATM internetwork (number of nodes).
o The size of ATM IP subnets (LIS) in the ATM Internetwork.
o Single IP subnet vs multiple IP subnet ATM internetworks.
o Single or multiple administrative authority
o Presence of routers providing transit to multihomed internets
o The presence or absence of dynamic address resolution.
o The presence or absence of an IP routing protocol
IP over ATM should therefore be characterized by:
o Encapsulations below the IP level.
o Degree to which a connection oriented lower level is available
and utilized.
o Type of address resolution at the IP subnet level (static or
dynamic).
o Degree to which address resolution is extended beyond the IP
subnet boundary.
o The type of routing (if any) supported above the IP level.
ATM-specific attributes of particular importance include:
o The different types of services provided by the ATM Adaptation
Layers (AAL). These specify whether a connection oriented or a
connectionless service is provided, the Quality-of-Service, the
connection-mode, etc. The models discussed within this document
assume connection-oriented, best effort services over AAL Type 5.
o The type of virtual circuits used, i.e., PVCs versus SVCs. The
PVC environment requires the use of either static tables for
ATM-to-IP address mapping or the use of inverse ARP, while the
SVC environment requires ARP functionality to be provided.
o The type of support for multicast services. If point-to-point
services only are available, then a server for IP multicast is
required. If point-to-multipoint services are available, then
IP multicast can be supported via meshes of point-to-multipoint
Cole, Shur, Villamizar Expires October 11, 1995 [Page 5]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
connections (although use of a server may be necessary due to
limits on the number of multipoint VCs able to be supported) or
to maintain the leaf initiated join semantics.
o The presence of logical link identifiers (VPI/VCIs) and the
various information element (IE) encodings within the ATM SVC
signaling specification, i.e., the ATM Forum UNI version 3.1.
This allows a VC originator to specify a range of "layer"
entities as the destination "AAL User". The AAL specifications
do not prohibit any particular "layer X" from attaching directly
to a local AAL service. Taken together these points imply a
range of methods for encapsulation of upper layer protocols over
ATM. For example, while LLC/SNAP encapsulation is one approach
(the default), it is also possible to bind virtual circuits
to higher level entities in the TCP/IP protocol stack. Some
examples of the latter are single VC per protocol binding, TULIP,
and TUNIC, discussed further in Section 4.
o The number and type of ATM administrative domains/networks, and
type of addressing used within an administrative domain/network.
In particular, in the single domain/network case, all attached
systems may be safely assumed to be using a single common
addressing format, while in the multiple domain case, attached
stations may not all be using the same common format,
with corresponding implications on address resolution. (See
Appendix A for a discussion of some of the issues that arise
when multiple ATM address formats are used in the same logical
IP subnet (LIS).) Also security/authentication is much more of a
concern in the multiple domain case.
IP over ATM proposals do not universally accept that IP routing over
an ATM network is required. Certain proposals rely on the following
assumptions:
o The assumption of widespread deployment of ATM within premises-
based networks, private wide-area networks and public networks,
and
o The definition of interfaces, signaling and routing protocols
among private ATM networks.
The above assumptions amount to ubiquitous deployment of a seemless
ATM fabric which serves as the hub of a star topology around which
all other media is attached. There has been a great deal of
discussion over when, if ever, this will be a realistic assumption for
very large internetworks, such as the Internet. Advocates of such
approaches point out that even if these are not relevant to very large
Cole, Shur, Villamizar Expires October 11, 1995 [Page 6]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
internetworks such as the Internet, there may be a place for such
models in smaller internetworks, such as corporate networks.
The NHRP protocol (Section 8.2), not necessarily specific to ATM,
would be particularly appropriate for the case of ubiquitous ATM
deployment. NHRP permits the establishment of direct connections
across IP subnets in the ATM domain.
Proposals have begun to emerge which support routing in the ATM layer
with routing exchange on a peer basis above the IP layer. These
proposals remain useful in the event that the above assumptions do not
hold. The similar Peer Model and the Integrated Model (Section 8.4)
both propose an exchange of routing information with IP. Proposals
have not yet been put forward in which full IP routing attribute
exchange is supported.
4 Encapsulations and Lower Layer Identification
Data encapsulation, and the identification of VC endpoints, constitute
two important issues that are somewhat orthogonal to the issues of
network topology and routing. The relationship between these two
issues is also a potential sources of confusion. In conventional
LAN technologies the 'encapsulation' wrapped around a packet of
data typically defines the (de)multiplexing path within source and
destination nodes (e.g. the Ethertype field of an Ethernet packet).
Choice of the protocol endpoint within the packet's destination node
is essentially carried 'in-band'.
As the multiplexing is pushed towards ATM and away from LLC/SNAP
mechanism, a greater burden will be placed upon the call setup and
teardown capacity of the ATM network. This may result in some
questions being raised regarding the scalability of these lower level
multiplexing options.
With the ATM Forum UNI version 3.1 service the choice of endpoint
within a destination node is made 'out of band' - during the Call
Setup phase. This is quite independent of any in-band encapsulation
mechanisms that may be in use. The B-LLI Information Element allows
Layer 2 or Layer 3 entities to be specified as a VC's endpoint. When
faced with an incoming SETUP message the Called Party will search
locally for an AAL User that claims to provide the service of the
layer specified in the B-LLI. If one is found then the VC will be
accepted (assuming other conditions such as QoS requirements are also
met).
An obvious approach for IP environments is to simply specify the
Internet Protocol layer as the VCs endpoint, and place IP packets into
AAL--SDUs for transmission. This is termed 'VC multiplexing' or 'Null
Cole, Shur, Villamizar Expires October 11, 1995 [Page 7]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Figure 1: AAL5 PDU is an encapsulated IP packet
Encapsulation', because it involves terminating a VC (through an AAL
instance) directly on a layer 3 endpoint. However, this approach
has limitations in environments that need to support multiple layer 3
protocols between the same two ATM level endpoints. Each pair of
layer 3 protocol entities that wish to exchange packets require their
own VC.
RFC--1483 [6] notes that VC multiplexing is possible, but focuses
on describing an alternative termed 'LLC/SNAP Encapsulation'. This
allows any set of protocols that may be uniquely identified by an
LLC/SNAP header to be multiplexed onto a single VC. Figure 1 shows
how this works for IP packets - the first 3 bytes indicate that
the payload is a Routed Non-ISO PDU, and the Organizationally Unique
Identifier (OUI) of 0x00-00-00 indicates that the Protocol Identifier
(PID) is derived from the EtherType associated with IP packets
(0x800). ARP packets are multiplexed onto a VC by using a PID of
0x806 instead of 0x800.
Whatever layer terminates a VC carrying LLC/SNAP encapsulated traffic
must know how to parse the AAL--SDUs in order to retrieve the packets.
The recently approved signalling standards for IP over ATM are more
explicit, noting that the default SETUP message used to establish IP
over ATM VCs must carry a B-LLI specifying an ISO 8802/2 Layer 2
(LLC) entity as each VCs endpoint. More significantly, there is no
information carried within the SETUP message about the identity of
the layer 3 protocol that originated the request - until the packets
begin arriving the terminating LLC entity cannot know which one or
more higher layers are packet destinations.
Taken together, this means that hosts require a protocol entity to
register with the host's local UNI 3.1 management layer as being an
LLC entity, and this same entity must know how to handle and generate
LLC/SNAP encapsulated packets. The LLC entity will also require
Cole, Shur, Villamizar Expires October 11, 1995 [Page 8]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Figure 2: LLC/SNAP encapsulation allows more than just IP or ARP per
VC.
mechanisms for attaching to higher layer protocols such as IP and ARP.
Figure 2 attempts to show this, and also highlights the fact that
such an LLC entity might support many more than just IP and ARP. In
fact the combination of RFC 1483 LLC/SNAP encapsulation, LLC entities
terminating VCs, and suitable choice of LLC/SNAP values, can go along
way towards providing an integrated approach to building multiprotocol
networks over ATM.
The processes of actually establishing AAL Users, and identifying them
to the local UNI 3.1 management layers, are still undefined and are
likely to be very dependent on operating system environments.
Two encapsulations have been discussed within the IP over ATM working
group which differ from those given in RFC--1483 [6]. These
have the characteristic of largely or totally eliminating IP header
overhead. These models were discussed in the July 1993 IETF meeting
in Amsterdam, but have not been fully defined by the working group.
Given a single hop configuration between Sub-ES, both Sub-ES assume
that following name resolution, address resolution, and SVC signaling,
an implicit binding is established between entities in the two
Sub-ESes. In this case full IP headers (and in particular source and
destination addresses) are not required in each data packet.
o The first model is "TCP and UDP over Lightweight IP" (TULIP)
in which only the IP protocol field is carried in each packet,
everything else being bound at call set-up time. In this case
Cole, Shur, Villamizar Expires October 11, 1995 [Page 9]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
the implicit binding is between the IP entities in each Sub-ES.
Since there is no further routing problem once the binding
is established, since AAL5 can indicate packet size, since
fragmentation cannot occur, and since ATM signaling will handle
exception conditions, the absence of all other IP header fields
and of ICMP should not be an issue. Entry to TULIP mode would
occur as the last stage in SVC signaling, by a simple extension
to the encapsulation negotiation described in RFC--1755 [11].
TULIP changes nothing in the abstract architecture of the IP
model, since each subnet ES still has an IP address which is
resolved to an ATM address. It simply uses the point-to-point
property of VCs to allow the elimination of some per-packet
overhead. The use of TULIP could in principle be negotiated on a
per-SVC basis or configured on a per-PVC basis.
o The second model is "TCP and UDP over a Nonexistent IP
Connection" (TUNIC). In this case no network-layer information is
carried in each packet, everything being bound at virtual circuit
set-up time. The implicit binding is between two applications
using either TCP or UDP directly over AAL5 on a dedicated VC. If
this can be achieved, the IP protocol field has no useful dynamic
function. However, in order to achieve binding between two
applications, the use of a well-known port number in classical
IP or in TULIP mode may be necessary during call set-up. This
is a subject for further study and would require significant
extensions to the use of SVC signaling described in RFC--1755
[11].
TULIP/TUNIC can be presented as being on one end of a continuum
opposite the SNAP/LLC encapsulation, with various forms of null
encapsulation somewhere in the middle. The continuum is simply a
matter of how much is moved from in-stream demultiplexing to call
setup demultiplexing. The various encapsulation types are presented
in Table 1.
Encapsulations such as TULIP and TUNIC make assumptions with regard
to the desirability to support connection oriented flow. The
tradeoffs between connection oriented and connectionless are discussed
in Section 5.
5 Connection Oriented and Connectionless Tradeoffs
The connection oriented and connectionless approaches each offer
advantages and disadvantages. In the past, strong advocates of pure
connection oriented and pure connectionless architectures have argued
intensely. IP over ATM does not need to be purely connectionless or
Cole, Shur, Villamizar Expires October 11, 1995 [Page 10]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Encapsulation In setup message Demultiplexing
-------------+--------------------------+------------------------
SNAP/LLC _ nothing _ source and destination
_ _ address, protocol
_ _ family, protocol, ports
_ _
NULL encaps _ protocol family _ source and destination
_ _ address, protocol, ports
_ _
TULIP _ source and destination _ protocol, ports
_ address, protocol family _
_ _
TUNIC - A _ source and destination _ ports
_ address, protocol family _
_ protocol _
_ _
TUNIC - B _ source and destination _ nothing
_ address, protocol family _
_ protocol, ports _
Table 1: Summary of Encapsulation Types
purely connection oriented.
IP over ATM provides a connection oriented architecture (ATM)
underlying a connectionless architecture (IP), on top of which much of
the traffic is supported by a reliable end-to-end connection oriented
protocol (TCP). A fundamental question is to what degree is it
beneficial to map different flows above IP into separate connections
below IP. There is a broad spectrum of opinion on this.
As stated in section 4, at one end of the spectrum, IP would remain
highly connectionless and set up single VCs between routers which are
adjacent on an IP subnet and for which there was active traffic flow.
All traffic between the such routers would be multiplexed on a single
ATM VC. At the other end of the spectrum, a separate ATM VC would
be created for each identifiable flow. For every unique TCP or UDP
address and port pair encountered a new VC would be required. Part of
the intensity of early arguments has been over failure to recognize
that there is a middle ground.
ATM offers QoS and traffic management capabilities that are well
suited for certain types of services. It may be advantageous to use
separate ATM VC for such services. Other IP services such as DNS,
are ill suited for connection oriented delivery, due to their normal
very short duration (typically one packet in each direction). Short
duration transactions, even many using TCP, may also be poorly suited
for a connection oriented model due to setup and state overhead.
Cole, Shur, Villamizar Expires October 11, 1995 [Page 11]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
APPLICATION Pure Connection Oriented Approach
----------------+-------------------------------------------------
General _ Always set up a VC
_
Short Duration _ Set up a VC. Either hold the packet during VC
UDP (DNS) _ setup or drop it and await a retransmission.
_ Teardown on a timer basis.
_
Short Duration _ Set up a VC. Either hold packet(s) during VC
TCP (SMTP) _ setup or drop them and await retransmission.
_ Teardown on detection of FIN-ACK or on a timer
_ basis.
_
Elastic (TCP) _ Set up a VC same as above. No clear method to
Bulk Transfer _ set QoS parameters has emerged.
_
Real Time _ Set up a VC. QoS parameters are assumed to
(audio, video) _ preceed traffic in RSVP or be carried in some
_ form within the traffic itself.
Table 2: Connection Oriented vs. Connectionless - a) a pure
connection oriented approach
ATM QoS and traffic management capabilities may be poorly suited for
elastic traffic.
Work in progress is addressing how QoS requirements might be expressed
and how the local decisions might be made as to whether those
requirements are best and/or most cost effectively accomplished using
ATM or IP capabilities. Table 2, Table 3, and Table 4 describe
typical treatment of various types of traffic using a pure connection
oriented approach, middle ground approach, and pure connectionless
approach.
The above qualitative description of connection oriented vs
connectionless service serve only as examples to illustrate differing
approaches. Work in the area of an integrated service model, QoS
and resource reservation are related to but outside the scope of
the IP over ATM Work Group. This work falls under the Integrated
Services Work Group (int-serv) and Reservation Protocol Work Group
(rsvp), and will ultimately determine when direct connections will be
established. The IP over ATM Work Group can make more rapid progress
if concentrating solely on how direct connections are established.
Cole, Shur, Villamizar Expires October 11, 1995 [Page 12]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
APPLICATION Middle Ground
----------------+-------------------------------------------------
General _ Use RSVP or other indication which clearly
_ indicate a VC is needed and what QoS parameters
_ are appropriate.
_
Short Duration _ Forward hop by hop. RSVP is unlikely to preceed
UDP (DNS) _ this type of traffic.
_
Short Duration _ Forward hop by hop unless RSVP indicates
TCP (SMTP) _ otherwise. RSVP is unlikely to preceed this
_ type of traffic.
_
Elastic (TCP) _ Forward hop by hop unless RSVP indicates
Bulk Transfer _ otherwise. Derive QoS parameters from RSVP, if
_ RSVP indicates special treatment. A local
_ decision can be made as to whether the QoS is
_ better served by a separate VC.
_
Real Time _ Forward hop by hop unless RSVP indicates
(audio, video) _ otherwise. RSVP will indicate QoS requirements.
_ It is assumed RSVP will generally be used for
_ this case. A local decision can be made as to
_ whether the QoS is better served by a sepa-
rate VC.
Table 3: Connection Oriented vs. Connectionless - b) a middle ground
approach
Cole, Shur, Villamizar Expires October 11, 1995 [Page 13]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
APPLICATION Pure Connectionless Approach
----------------+-------------------------------------------------
General _ Always forward hop by hop. Use queueing
_ algorithms implemented at the IP layer to
_ support reservations such as those specified by
_ RSVP.
_
Short Duration _ Forward hop by hop.
UDP (DNS) _
_
Short Duration _ Forward hop by hop.
TCP (SMTP) _
_
Elastic (TCP) _ Forward hop by hop. Assume ability of TCP to
Bulk Transfer _ share bandwidth (within a VBR VC) works as well
_ or better than ATM traffic management.
_
Real Time _ Forward hop by hop. Assume that queueing
(audio, video) _ algorithms at the IP level can work as well or
_ better (due to support for predictive
_ reservation).
Table 4: Connection Oriented vs. Connectionless - c) a pure
connectionless approach
Cole, Shur, Villamizar Expires October 11, 1995 [Page 14]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
6 Crossing IP Subnet Boundaries
A single IP subnet will not scale well to a large size. Techniques
which extend the size of an IP subnet in other media include MAC layer
bridging, and proxy ARP bridging.
MAC layer bridging alone does not scale. Protocols such as ARP rely
on the media broadcast to exchange address resolution information.
Most bridges improve scaling by capturing ARP packets and retaining
the content, and distributing the information among bridging peers,
then repeating the ARP information gathered from ARP replies only
where explicit ARP requests are made. This technique is known as
proxy ARP.
Proxy ARP bridging improves scaling by reducing broadcast traffic, but
still suffers scaling problems. If the bridged IP subnet is part
of a larger internetwork, often a routing protocol is required to
indicate what destinations are beyond the IP subnet. This requires
running a routing protocol. Because internets of enormous size create
scaling problems for routing protocols, the component netowrks of such
large internets are often partitioned into areas, autonomous systems
or routing domains, and routing confederacies.
Scaling limits of the simple IP subnet require a large network to
be partitioned into smaller IP subnets. For NBMA media like ATM,
there are advantages to creating direct connections across the entire
underlying NBMA network. This leads to the need to create direct
connections across IP subnet boundaries.
Many of the problems and issues associated with this configuration
were originally being addressed in the IETF's IPLPDN working group and
the IP over ATM working group. This area is now being addressed
in the Routing over Large Clouds working group. Examples of work
performed in the IPLPDN working group include short-cut routing
(proposed by P. Tsuchiya) and directed ARP RFC--1433 [5] over SMDS
networks. The ROLC working group has produced the distributed ARP
server architectures and next hop resolution protocols in RFC--1735
[7], and NHRP [8]. Questions/Issues specifically related to defining
an capability to cross IP subnet boundaries include:
o How can routing be optimized across multiple logical IP subnets
over both a common ATM based and a non-ATM based infrastructure.
For example, in Figure 3, there are two gateways/routers between
the SVC-based WATM and the non-ATM subnet. The optimal path
from end-systems on any ATM-based subnet to the non ATM-based
subnet is a function of the routing state information of the two
routers.
o How to incorporate policy routing constraints.
Cole, Shur, Villamizar Expires October 11, 1995 [Page 15]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Figure 3: An SVC-based WATM configuration between End Systems which
in fact could be Hosts or Routers.
o What is the proper coupling between routing and address
resolution particularly with respect to off-subnet communication.
o What are the local procedures to be followed by hosts and
routers.
o Routing between end systems not sharing a common IP-level (or L3)
network prefix, but able to be directly connected at the NBMA
media level,
o Defining the details for an efficient address resolution
architecture including defining the procedures to be followed by
clients and servers (see RFC--1433 [5], RFC--1735 [7] and NHRP
[8]).
o How to identify the need for and accommodate special purpose SVCs
for control or routing and high bandwidth data transfers.
For ATM (unlike other NBMA media), an additional complexity in
supporting IP routing over these ATM internets lies in the
multiplicity of address formats in UNI 3.0 [4]. NSAP modeled address
formats only are supported on "private ATM" networks, while either 1)
E.164 only, 2) NSAP modeled formats only, or 3) both are supported on
"public ATM" networks. Further, while both the E.164 and NSAP modeled
address formats are to be considered as network points of attachment,
it seems that E.164 only networks are to be considered as subordinate
to "private networks", in some sense. This leads to some confusion in
defining an ARP mechanism in supporting all combinations of end-to-end
scenarios (refer to the discussion in Appendix A on the possible
Cole, Shur, Villamizar Expires October 11, 1995 [Page 16]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
scenarios to be supported by ARP).
Figure 3 shows an example end-to-end configuration consisting of
four components, three of which are ATM technology based, while
the fourth is a standard IP subnet based on non-ATM technology.
Since end-systems (either hosts or routers) attached to the ATM-based
networks may communicate directly via ATM (subject to policy
constraints), in this model such nodes may also communicate directly
at the IP level without necessarily needing an intermediate router,
even if end-systems do not share a common IP-level network prefix.
Communication with end-systems on the non-ATM-based Classical IP
subnet takes place via a router, following the Classical IP model (see
Section 8.1 below).
7 Extensions to IP Routing
RFC--1620 [3] describes the problems and issues associated with direct
connections across IP subnet boundaries in greater detail, as well as
possible solution approaches. The ROLC WG has identified persistent
routing loop problems that can occur if the techniques used to
accomplish direct connections across IP subnet boundaries are used to
carry reachability information for destinations reachable by more than
one router attached to the NBMA media.
Proposals have begun to emerge which simplify router and host
procedures for the case where direct connections across IP subnet
boundaries are desired, yet routing protocols are still needed and
further aggregation of routing information is desirable or needed.
8 IP Over ATM Proposals
8.1 The Classical IP Model
The Classical IP Model was suggested at the Spring 1993 IETF meeting
RFC--1577 [9] and retains the classical IP subnet architecture. This
model simply consists of cascading instances of IP subnets with
IP-level (or L3) routers at IP subnet borders. An example realization
of this model consists of a concatenation of two IP subnets. This is
shown in Figure 4. Forwarding IP packets over this Classical IP model
is straight forward using already well established routing techniques
and protocols.
SVC-based ATM IP subnets are simplified in that they:
Cole, Shur, Villamizar Expires October 11, 1995 [Page 17]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Figure 4: The Classical IP model as a concatenation of three separate
ATM IP subnets; two SVC LATM subnets and a single PVC WATM subnet.
o limit the number of hosts which must be directly connected at any
given time to those that may actually exchange traffic.
o The ATM network is capable of setting up connections between any
pair of hosts. Consistent with the standard IP routing algorithm
[RFC1009] connectivity to the ``outside'' world is achieved only
through a router, which may provide firewall functionality if so
desired.
o The IP subnet supports an efficient mechanism for address
resolution.
Issues addressed by the IP Over ATM Working Group, and some of the
resolutions, for this model are:
o Methods of encapsulation and multiplexing. This issue is
addressed in RFC--1483 [6], in which two methods of encapsulation
are defined, an LLC/SNAP and a per-VC multiplexing option.
o The definition of an address resolution server (defined in
RFC--1577).
o Defining the default MTU size. This issue is addressed in
RFC--1626 [2] which proposes the use of the MTU discovery
protocol (RFC--1191 [10]).
o Support for IP multicasting. In the summer of 1994, work
began on the issue of supporting IP multicasting over the SVC
LATM model. The proposal for IP multicasting is found in [1].
In order to support IP multicasting the ATM subnet must either
support point-to- multipoint SVCs, or multicast servers, or both.
o Defining interim SVC parameters, such as QoS parameters and
time-out values.
o Signaling and negotiations of parameters such as MTU size
Cole, Shur, Villamizar Expires October 11, 1995 [Page 18]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
and method of encapsulation. RFC--1755 [11] describes an
implementation agreement for routers signaling the ATM network
to establish SVCs initially based upon the ATM Forum's UNI
version 3.0 specification [4], and eventually to be ba sed
upon the ATM Forum's UNI version 3.1 and later specifications.
Topics addressed in RFC--1755 include (but are not limited to)
VC management procedures, e.g., when to time-out SVCs, QOS
parameters, service classes, explicit setup message formats for
various encapsulation methods, Sub-ES to Sub-ES negotiations,
etc.
RFC-1577 is also applicable to PVC-based subnets. Full mesh PVC
connectivity is required.
For more information see RFC--1577 [9].
8.2 The ROLC NHRP Model
The Next Hop Resolution Protocol (NHRP) [8], currently a draft defined
by the Routing Over Large Clouds Working Group (rolc) performs
address resolution to accomplish direct connections across IP subnet
boundaries. NHRP can suppliment RFC--1577 ARP. There has been recent
discussion of replacing RFC--1577 ARP with NHRP. NHRP can also perform
a proxy address resolution to provide the address of the border router
serving a destination off of the NBMA which is only served by a
single router on the NBMA. NHRP cannot be used in this way to support
addresses learned from routers for which the same destinations may
be heard at other routers, without the risk of creating persistent
routing loops.
NHRP is one of several methods of available to provide direct
connections across IP subnet boundaries.
8.3 Ohta's "Conventional" Model
The conventional model is layered and has the same network layer
architecture as the standard IP model, hence it retains the IP
subnet architecture. At the link layer, all types of ATM subnets
are allowed. The basic difference between the Classical and the
"Conventional" model is whether it is assumed that a router can relay
IP packets cell by cell or not. Link layers are connected by
routers but, when the two link layers are ATM, the routers are capable
of forwarding cell-by-cell. This could provide a latency advantage
depending on whether cell interleaving from multiple IP packets is
allowed or not (e.g., an ATM AAL such as AAL3/4 would allow for
Cole, Shur, Villamizar Expires October 11, 1995 [Page 19]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
cell interleaving, while AAL5 would not). This cell forwarding is
accomplished through a higher level mapping, above the ATM VCI layer.
8.4 The Peer and Integrated Models
The Classical IP model requires that IP routing be independent of
the details of connectivity in an ATM network. If ATM routing
processes are present within the network, this implies that the user
needs to install and manage two independent routing hierarchies (one
for ATM, and one for IP). The Integrated model (proposed and under
study within the Multiprotocol group of ATM Forum) considers a single
routing protocol to be used for both IP and for ATM. A single routing
information exchange is used to distribute topological information.
The routing computation used to calculate routes for IP will take into
account the topology, including link and node characteristics, of both
the IP and ATM networks and calculates an optimal route for IP packets
over the combined topology.
Many issues exist with respect to the Integrated Routing proposal.
While the potential exists to compute "better" routes, overall routing
complexity (e.g., ensuring loop-free routing) is increased.
Peer Models place IP routers/gateways on a peer basis with
corresponding entities in an ATM cloud (where the ATM cloud may
consist of a set of ATM networks, inter-connected via UNI or P-NNI
interfaces). That is, ATM network entities and the attached IP hosts
or routers exchange call routing information on a peer basis. Within
the ATM cloud, ATM network level addressing (NSAP-style), call routing
and packet formats are used. Peer models are being discussed in the
Multiprotocol SWG of the ATM Forum.
A rationale for these models is that due to the envisioned complexity
of future ATM networks, the routing complexities will be similar
to routing over complex internetworks. Rather than building in
redundancy in routing complexities at the network and subnet levels,
perhaps it is possible to embed within the ATM network fabric IS
(router) functionality. This is illustrated in Figure 5. Here the
term "Single ATM Domain" refers to a collection of ATM switches from a
single ATM switch vendor or a collection of ATM switches administered
by a single authority. The routing function of the ATM-IS is only
required during the connection set-up phase, and not in the data
transfer phase following the connection establishment.
There are numerous issues to be worked for this Peer Model including:
o Defining a routing architecture which
Cole, Shur, Villamizar Expires October 11, 1995 [Page 20]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Figure 5: The ATM Internet model embeds within the ATM cloud
network level routing features which in essence peers with traditional
gateways or routers.
-- scales to the large size expected of the eventual ATM
network, and
-- supports TOS routing and IP options, and
o Defining a functional architecture which
-- scales to the large size expected of the eventual ATM
network,
-- supports the current status of the ATM Forum's standards in
signalling, routing, etc., and
-- allows a transition from Classical IP subnet models to peer
models.
During the discussions of the IP over ATM working group, it was felt
that the problems with the end-to-end peer model were much harder than
any other model, and had more unresolved technical issues. While
encouraging interested individuals/companies to research this area, it
was not an initial priority of the working group to address these
issues. The ATM Forum Network Layer Multiprotocol Working Group is
discussing this model.
8.5 Transition Models
Finally, it is useful to consider transition models, lying somewhere
between the Classical IP Models and the Peer Models. Some possible
architectures for transition models have been suggested by Fong Liaw.
Others are possible, for example Figure 6 showing a Classical IP
Cole, Shur, Villamizar Expires October 11, 1995 [Page 21]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Figure 6: The ATM transition model assuming the presence of gateways
or routers between the ATM networks and the ATM peer networks.
transition model which assumes the presence of gateways between ATM
networks and ATM Peer networks.
9 Application of the Working Group's and Related Documents
The IP Over ATM Working Group has generated several Internet-Drafts
and RFCs. This section identifies the relationship of these and other
related documents to the various IP Over ATM Models identified in this
document. The Drafts and RFCs produced to date are the following
references, RFC--1483 [6], RFC--1577 [9], RFC--1626 [2], RFC--1755
[11], IPMC [1] and NHRP [8]. Table 5 gives a summary of these
documents and their relationship to the various IP Over ATM Models.
Acknowledgments:
This draft is the direct result of the numerous discussions of the
IP over ATM Working Group of the Internet Engineering Task Force.
The authors also had the benefit of several private discussions
with H. Nguyen of AT&T Bell Laboratories. Brian Carpenter of CERN
was kind enough to contribute the TULIP and TUNIC sections to this
draft. Grenville Armitage of Bellcore was kind enough to contribute
the sections on VC binding, encapsulations and the use of B-LLI
information elements to signal such bindings. The text of Appendix A
was pirated liberally from Anthony Alles' of Cisco posting on the IP
over ATM discussion list (and modified at the authors' discretion).
M. Ohta provided a description of the Conventional Model (again which
the authors modified at their discretion). This draft also has
Cole, Shur, Villamizar Expires October 11, 1995 [Page 22]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
Documents Summary
----------------+-------------------------------------------------
RFC-1483 _ How to identify/label multiple
_ packet/frame-based protocols multiplexed over
_ ATM AAL5. Applies to any model dealing with IP
_ over ATM AAL5.
_
RFC-1577 _ Model for transporting IP and ARP over ATM AAL5
_ in an IP subnet where all nodes share a common
_ IP network prefix. Includes ARP server/Inv-ARP
_ packet formats and procedures for SVC/PVC
_ subnets.
_
RFC-1626 _ Specifies default IP MTU size to be used with
_ ATM AAL5. Requires use of PATH MTU discovery.
_ Applies to any model dealing with IP over ATM
_ AAL5
_
RFC-1755 _ Defines how implementations of IP over ATM
_ should use ATM call control signaling
_ procedures, and recommends values of mandatory
_ and optional IEs focusing particularly on the
_ Classical IP model.
_
IPMC _ Defines how to support IP multicast in Classical
_ IP model using either (or both) meshes of
_ point-to-multipoint ATM VCs, or multicast
_ server(s).
_
NHRP _ Describes a protocol that can be used by hosts
_ and routers to determine the NBMA next hop
_ address of a destination in "NBMA connectivity"
_ of the sending node. If the destination is not
_ connected to the NBMA fabric, the IP and NBMA
_ addresses of preferred egress points are
_ returned.
Table 5: Summary of WG Documents
Cole, Shur, Villamizar Expires October 11, 1995 [Page 23]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
benefitted from numerous suggestions from John T. Amenyo of ANS, Joel
Halpern of Newbridge, and Andy Malis of Ascom-Timplex.
Authors' Addresses:
Robert G. Cole
AT&T Bell Laboratories
101 Crawfords Corner Road, Rm. 3L-533
Holmdel, NJ 07733
Phone: (908) 949-1950
Fax: (908) 949-8887
Email: rgc@qsun.att.com
David H. Shur
AT&T Bell Laboratories
101 Crawfords Corner Road, Rm. 1F-338
Holmdel, NJ 07733
Phone: (908) 949-6719
Fax: (908) 949-5775
Email: d.shur@att.com
Curtis Villamizar
Advanced Network & Services
100 Clearbrook Road
Elmsford, NY 10523
Email: curtis@ans.net
References
[1] G. Armitage. Support for multicast over uni 3.1 based
atm networks. Internet Draft (Work in Progress) draft-ietf-
ipatm-ipmc-04.txt, Internet Engineering Task Force, February
1995.
[2] R. Atkinson. Default IP MTU for use over ATM AAL5. Request
for Comments (Experimental) RFC 1626, Internet Engineering Task
Force, may 1994.
[3] R. Braden, J. Postel, and Y. Rekhter. Internet architecture ex-
tensions for shared media. Request for Comments (Informational)
RFC 1620, Internet Engineering Task Force, may 1994.
Cole, Shur, Villamizar Expires October 11, 1995 [Page 24]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
[4] ATM Forum. ATM User-Network Interface Specification. Prentice
Hall, September 1993.
[5] J. Garrett, J. Hagan, and J. Wong. Directed ARP. Request
for Comments (Experimental) RFC 1433, Internet Engineering Task
Force, March 1993.
[6] J. Heinanen. Multiprotocol encapsulation over ATM adaptation
layer 5. Request for Comments (Proposed Standard) RFC 1483,
Internet Engineering Task Force, July 1993.
[7] J. Heinanen and R. Govindan. Nbma address resolution protocol
(narp). Request for Comments (Experimental) RFC 1735, Internet
Engineering Task Force, December 1994.
[8] D. Katz and D. Piscitello. Nbma next hop resolution
protocol (nhrp). Internet Draft (Work in Progress) draft-ietf-
rolc-nhrp-03.txt, Internet Engineering Task Force, November
1994.
[9] M. Laubach. Classical IP and ARP over ATM. Request for Comments
(Proposed Standard) RFC 1577, Internet Engineering Task Force,
January 1994.
[10] J. Mogul and S. Deering. Path MTU discovery. Request for
Comments (Draft Standard) RFC 1191, Internet Engineering Task
Force, November 1990.
[11] M. Perez, F. Liaw, D. Grossman, A. Mankin, and A. Hoffman.
Atm signalling support for ip over atm. Request for Comments
(Informational) RFC 1755, Internet Engineering Task Force,
January 1995.
A Potential Interworking Scenarios to be Supported by ARP
The architectural model of the VC routing protocol, being defined by
the Private Network-to-Network Interface (P-NNI) working group of the
ATM Forum, categorizes ATM networks into two types:
o Those that participate in the VC routing protocols and use NSAP
modeled addresses UNI 3.0 [4] (referred to as private networks,
for short), and
o Those that do not participate in the VC routing protocol.
Typically, but possibly not in all cases, public ATM networks
that use native mode E.164 addresses UNI 3.0 [4] will fall into
this later category.
Cole, Shur, Villamizar Expires October 11, 1995 [Page 25]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
The issue for ARP, then is to know what information must be returned
to allow such connectivity. Consider the following scenarios:
o Private host to Private Host, no intervening public transit
network(s): Clearly requires that ARP return only the NSAP
modeled address format of the end host.
o Private host to Private host, through intervening public
networks: In this case, the connection setup from host A to host
B must transit the public network(s). This requires that at
each ingress point to the public network that a routing decision
be made as to which is the correct egress point from that public
network to the next hop private ATM switch, and that the native
E.164 address of that egress point be found (finding this is a VC
routing problem, probably requiring configuration of the public
network links and connectivity information). ARP should return,
at least, the NSAP address of the endpoint in which case the
mapping of the NSAP addresses to the E.164 address, as specified
in [4], is the responsibility of ingress switch to the public
network.
o Private Network Host to Public Network Host: To get connectivity
between the public node and the private nodes requires the
same kind of routing information discussed above - namely, the
directly attached public network needs to know the (NSAP format)
ATM address of the private station, and the native E.164 address
of the egress point from the public network to that private
network (or to that of an intervening transit private network
etc.). There is some argument, that the ARP mechanism could
return this egress point native E.164 address, but this may
be considered inconsistent for ARP to return what to some is
clearly routing information, and to others is required signaling
information.
In the opposite direction, the private network node can use, and
should only get, the E.164 address of the directly attached public
node. What format should this information be carried in? This
question is clearly answered, by Note 9 of Annex A of UNI 3.0 [4],
Version 2.4 (passed by ballot recently to become Version 3.0), vis:
"A call originated on a Private UNI destined for an host which only
has a native (non-NSAP) E.164 address (i.e. a system directly
attached to a public network supporting the native E.164 format) will
code the Called Party number information element in the (NSAP) E.164
private ATM Address Format, with the RD, AREA, and ESI fields set to
zero. The Called Party Subaddress information element is not used."
Hence, in this case, ARP should return the E.164 address of the
public ATM station in NSAP format. This is essentially implying an
Cole, Shur, Villamizar Expires October 11, 1995 [Page 26]
INTERNET-DRAFT IP over ATM: A Framework Document April 11, 1995
algorithmic resolution between the native E.164 and NSAP addresses of
directly attached public stations.
o Public network host to Public network host, no intervening
private network: In this case, clearly the Q.93b requests would
use native E.164 address formats.
o Public network host to Public network host, intervening private
network: same as the case immediately above, since getting
to and through the private network is a VC routing, not an
addressing issue.
So several issues arise for ARP in supporting arbitrary connections
between hosts on private and public network. One is how to
distinguish between E.164 address and E.164 encoded NSAP modeled
address. Another is what is the information to be supplied by ARP,
e.g., in the public to private scenario should ARP return only the
private NSAP modeled address or both an E.164 address, for a point of
attachment between the public and private networks, along with the
private NSAP modeled address.
Cole, Shur, Villamizar Expires October 11, 1995 [Page 27]