Network Working Group D. Clark
Request for Comments: 1287 MIT
L. Chapin
BBN
V. Cerf
CNRI
R. Braden
ISI
R. Hobby
UC Davis
December 1991
Towards the Future Internet Architecture
Status of this Memo
This informational RFCdiscusses important directions for possible
future evolution of the Internet architecture, and suggests steps
towards the desired goals. It is offered to the Internet community
for discussion and comment. This memo provides information for the
Internet community. It does not specify an Internet standard.
Distribution of this memo is unlimited.
Table of Contents
1. INTRODUCTION ................................................. 2
2. ROUTING AND ADDRESSING ....................................... 5
3. MULTI-PROTOCOL ARCHITECTURES ................................. 9
4. SECURITY ARCHITECTURE ........................................ 13
5 TRAFFIC CONTROL AND STATE .................................... 16
6. ADVANCED APPLICATIONS ........................................ 18
7. REFERENCES ................................................... 21
APPENDIX A. Setting the Stage .................................... 22
APPENDIX B. Group Membership ..................................... 28
Security Considerations .......................................... 29
Authors' Addresses ............................................... 29
1. INTRODUCTION
1.1 The Internet Architecture
The Internet architecture, the grand plan behind the TCP/IP
protocol suite, was developed and tested in the late 1970s by a
small group of network researchers [1-4]. Several important
features were added to the architecture during the early 1980's --
subnetting, autonomous systems, and the domain name system [5,6].
More recently, IP multicasting has been added [7].
Within this architectural framework, the Internet Engineering Task
Force (IETF) has been working with great energy and effectiveness
to engineer, define, extend, test, and standardize protocols for
the Internet. Three areas of particular importance have been
routing protocols, TCP performance, and network management.
Meanwhile, the Internet infrastructure has continued to grow at an
astonishing rate. Since January 1983 when the ARPANET first
switched from NCP to TCP/IP, the vendors, managers, wizards, and
researchers of the Internet have all been laboring mightily to
survive their success.
A set of the researchers who had defined the Internet architecture
formed the original membership of the Internet Activities Board
(IAB). The IAB evolved from a technical advisory group set up in
1981 by DARPA to become the general technical and policy oversight
body for the Internet. IAB membership has changed over the years
to better represent the changing needs and issues in the Internet
community, and more recently, to reflect the internationalization
of the Internet, but it has retained an institutional concern for
the protocol architecture.
The IAB created the Internet Engineering Task Force (IETF) to
carry out protocol development and engineering for the Internet.
To manage the burgeoning IETF activities, the IETF chair set up
the Internet Engineering Steering Group (IESG) within the IETF.
The IAB and IESG work closely together in ratifying protocol
standards developed within the IETF.
Over the past few years, there have been increasing signs of
strains on the fundamental architecture, mostly stemming from
continued Internet growth. Discussions of these problems
reverberate constantly on many of the major mailing lists.
1.2 Assumptions
The priority for solving the problems with the current Internet
architecture depends upon one's view of the future relevance of
TCP/IP with respect to the OSI protocol suite. One view has been
that we should just let the TCP/IP suite strangle in its success,
and switch to OSI protocols. However, many of those who have
worked hard and successfully on Internet protocols, products, and
service are anxious to try to solve the new problems within the
existing framework. Furthermore, some believe that OSI protocols
will suffer from versions of many of the same problems.
To begin to attack these issues, the IAB and the IESG held a one-
day joint discussion of Internet architectural issues in January
1991. The framework for this meeting was set by Dave Clark (see
Appendix A for his slides). The discussion was spirited,
provocative, and at times controversial, with a lot of soul-
searching over questions of relevance and future direction. The
major result was to reach a consensus on the following four basic
assumptions regarding the networking world of the next 5-10 years.
(1) The TCP/IP and OSI suites will coexist for a long time.
There are powerful political and market forces as well as
some technical advantages behind the introduction of the OSI
suite. However, the entrenched market position of the TCP/IP
protocols means they are very likely to continue in service
for the foreseeable future.
(2) The Internet will continue to include diverse networks and
services, and will never be comprised of a single network
technology.
Indeed, the range of network technologies and characteristics
that are connected into the Internet will increase over the
next decade.
(3) Commercial and private networks will be incorporated, but we
cannot eXPect the common carriers to provide the entire
service. There will be mix of public and private networks,
common carriers and private lines.
(4) The Internet architecture needs to be able to scale to 10**9
networks.
The historic exponential growth in the size of the Internet
will presumably saturate some time in the future, but
forecasting when is about as easy as forecasting the future
economy. In any case, responsible engineering requires an
architecture that is CAPABLE of expanding to a worst-case
size. The exponent "9" is rather fuzzy; estimates have
varied from 7 to 10.
1.3 Beginning a Planning Process
Another result of the IAB and IESG meeting was the following list
of the five most important areas for architectural evolution:
(1) Routing and Addressing
This is the most urgent architectural problem, as it is
directly involved in the ability of the Internet to continue
to grow successfully.
(2) Multi-Protocol Architecture
The Internet is moving towards widespread support of both the
TCP/IP and the OSI protocol suites. Supporting both suites
raises difficult technical issues, and a plan -- i.e., an
architecture -- is required to increase the chances of
success. This area was facetiously dubbed "making the
problem harder for the good of mankind."
Clark had observed that translation gateways (e.g., mail
gateways) are very much a fact of life in Internet operation
but are not part of the architecture or planning. The group
discussed the possibility of building the architecture around
the partial connectivity that such gateways imply.
(3) Security Architecture
Although military security was considered when the Internet
architecture was designed, the modern security issues are
much broader, encompassing commercial requirements as well.
Furthermore, experience has shown that it is difficult to add
security to a protocol suite unless it is built into the
architecture from the beginning.
(4) Traffic Control and State
The Internet should be extended to support "real-time"
applications like voice and video. This will require new
packet queueing mechanisms in gateways -- "traffic control"
-- and additional gateway state.
(5) Advanced Applications
As the underlying Internet communication mechanism matures,
there is an increasing need for innovation and
standardization in building new kinds of applications.
The IAB and IESG met again in June 1991 at SDSC and devoted three
full days to a discussion of these five topics. This meeting,
which was called somewhat perversely the "Architecture Retreat",
was convened with a strong resolve to take initial steps towards
planning evolution of the architecture. Besides the IAB and IESG,
the group of 32 people included the members of the Research
Steering Group (IRSG) and a few special guests. On the second
day, the Retreat broke into groups, one for each of the five
areas. The group membership is listed in Appendix B.
This document was assembled from the reports by the chairs of
these groups. This material was presented at the Atlanta IETF
meeting, and appears in the minutes of that meeting [8].
2. ROUTING AND ADDRESSING
Changes are required in the addressing and routing structure of IP to
deal with the anticipated growth and functional evolution of the
Internet. We expect that:
o The Internet will run out of certain classes of IP network
addresses, e.g., B addresses.
o The Internet will run out of the 32-bit IP address space
altogether, as the space is currently subdivided and managed.
o The total number of IP network numbers will grow to the point
where reasonable routing algorithms will not be able to perform
routing based upon network numbers.
o There will be a need for more than one route from a source to a
destination, to permit variation in TOS and policy conformance.
This need will be driven both by new applications and by diverse
transit services. The source, or an agent acting for the
source, must control the selection of the route options.
2.1 Suggested Approach
There is general agreement on the approach needed to deal with
these facts.
(a) We must move to an addressing scheme in which network numbers
are aggregated into larger units as the basis for routing.
An example of an aggregate is the Autonomous System, or the
Administrative Domain (AD).
Aggregation will accomplish several goals: define regions
where policy is applied, control the number of routing
elements, and provide elements for network management. Some
believe that it must be possible to further combine
aggregates, as in a nesting of ADs.
(b) We must provide some efficient means to compute common
routes, and some general means to compute "special" routes.
The general approach to special routes will be some form of
route setup specified by a "source route".
There is not full agreement on how ADs may be expected to be
aggregated, or how routing protocols should be organized to deal
with the aggregation boundaries. A very general scheme may be
used [ref. Chiappa], but some prefer a scheme that more restricts
and defines the expected network model.
To deal with the address space exhaustion, we must either expand
the address space or else reuse the 32 bit field ("32bf") in
different parts of the net. There are several possible address
formats that might make sense, as described in the next section.
Perhaps more important is the question of how to migrate to the
new scheme. All migration plans will require that some routers
(or other components inside the Internet) be able to rewrite
headers to accommodate hosts that handle only the old or format or
only the new format. Unless the need for such format conversion
can be inferred algorithmically, migration by itself will require
some sort of setup of state in the conversion element.
We should not plan a series of "small" changes to the
architecture. We should embark now on a plan that will take us
past the exhaustion of the address space. This is a more long-
range act of planning than the Internet community has undertaken
recently, but the problems of migration will require a long lead
time, and it is hard to see an effective way of dealing with some
of the more immediate problems, such as class B exhaustion, in a
way that does not by itself take a long time. So, once we embark
on a plan of change, it should take us all the way to replacing
the current 32-bit global address space. (This conclusion is
subject to revision if, as is always possible, some very clever
idea surfaces that is quick to deploy and gives us some breathing
room. We do not mean to discourage creative thinking about
short-term actions. We just want to point out that even small
changes take a long time to deploy.)
Conversion of the address space by itself is not enough. We must
at the same time provide a more scalable routing architecture, and
tools to better manage the Internet. The proposed approach is to
ADs as the unit of aggregation for routing. We already have
partial means to do this. IDPR does this. The OSI version of BGP
(IDRP) does this. BGP could evolve to do this. The additional
facility needed is a global table that maps network numbers to
ADs.
For several reasons (special routes and address conversion, as
well as accounting and resource allocation), we are moving from a
"stateless" gateway model, where only precomputed routes are
stored in the gateway, to a model where at least some of the
gateways have per-connection state.
2.2 Extended IP Address Formats
There are three reasonable choices for the extended IP address
format.
A) Replace the 32 bit field (32bf) with a field of the same size
but with different meaning. Instead of being globally
unique, it would now be unique only within some smaller
region (an AD or an aggregate of ADs). Gateways on the
boundary would rewrite the address as the packet crossed the
boundary.
Issues: (1) addresses in the body of packets must be found
and rewritten; (2) the host software need not be changed; (3)
some method (perhaps a hack to the DNS) must set up the
address mappings.
This scheme is due to Van Jacobson. See also the work by
Paul Tsuchiya on NAT.
B) Expand the 32bf to a 64 bit field (or some other new size),
and use the field to hold a global host address and an AD for
that host.
This choice would provide a trivial mapping from the host to
the value (the AD) that is the basis of routing. Common
routes (those selected on the basis of destination address
without taking into account the source address as well) can
be selected directly from the packet address, as is done
today, without any prior setup.
3) Expand the 32bf to a 64 bit field (or some other new size),
and use the field as a "flat" host identifier. Use
connection setup to provide routers with the mapping from
host id to AD, as needed.
The 64 bits can now be used to simplify the problem of
allocating host ids, as in Ethernet addresses.
Each of these choices would require an address re-writing module
as a part of migration. The second and third require a change to
the IP header, so host software must change.
2.3 Proposed Actions
The following actions are proposed:
A) Time Line
Construct a specific set of estimates for the time at which
the various problems above will arise, and construct a
corresponding time-line for development and deployment of a
new addressing/routing architecture. Use this time line as a
basis for evaluating specific proposals for changes. This is
a matter for the IETF.
B) New Address Format
Explore the options for a next generation address format and
develop a plan for migration. Specifically, construct a
prototype gateway that does address mapping. Understand the
complexity of this task, to guide our thinking about
migration options.
C) Routing on ADs
Take steps to make network aggregates (ADs) the basis of
routing. In particular, explore the several options for a
global table that maps network numbers to ADs. This is a
matter for the IETF.
D) Policy-Based Routing
Continue the current work on policy based routing. There are
several specific objectives.
- Seek ways to control the complexity of setting policy
(this is a human interface issue, not an algorithm
complexity issue).
- Understand better the issues of maintaining connection
state in gateways.
- Understand better the issues of connection state setup.
E) Research on Further Aggregation
Explore, as a research activity, how ADs should be aggregated
into still larger routing elements.
- Consider whether the architecture should define the
"role" of an AD or an aggregate.
- Consider whether one universal routing method or
distinct methods should be used inside and outside ADs
and aggregates.
Existing projects planned for DARTnet will help resolve several of
these issues: state in gateways, state setup, address mapping,
accounting and so on. Other experiments in the R&D community also
bear on this area.
3. MULTI-PROTOCOL ARCHITECTURE
Changing the Internet to support multiple protocol suites leads to
three specific architectural questions:
o How exactly will we define "the Internet"?
o How would we architect an Internet with n>1 protocol suites,
regardless of what the suites are?
o Should we architect for partial or filtered connectivity?
o How to add explicit support for application gateways into the
architecture?
3.1 What is the "Internet"?
It is very difficult to deal constructively with the issue of "the
multi-protocol Internet" without first determining what we believe
"the Internet" is (or should be). We distinguish "the Internet",
a set of communicating systems, from "the Internet community", a
set of people and organizations. Most people would accept a loose
definition of the latter as "the set of people who believe
themselves to be part of the Internet community". However, no
such "sociological" definition of the Internet itself is likely to
be useful.
Not too long ago, the Internet was defined by IP connectivity (IP
and ICMP were - and still are - the only "required" Internet
protocols). If I could PING you, and you could PING me, then we
were both on the Internet, and a satisfying working definition of
the Internet could be constructed as a roughly transitive closure
of IP-speaking systems. This model of the Internet was simple,
uniform, and - perhaps most important - testable. The IP-
connectivity model clearly distinguished systems that were "on the
Internet" from those that were not.
As the Internet has grown and the technology on which it is based
has gained widespread commercial acceptance, the sense of what it
means for a system to be "on the Internet" has changed, to
include:
* Any system that has partial IP connectivity, restricted by
policy filters.
* Any system that runs the TCP/IP protocol suite, whether or
not it is actually Accessible from other parts of the
Internet.
* Any system that can exchange RFC-822 mail, without the
intervention of mail gateways or the transformation of mail
objects.
* Any system with e-mail connectivity to the Internet, whether
or not a mail gateway or mail object transformation is
required.
These definitions of "the Internet", are still based on the
original concept of connectivity, just "moving up the stack".
We propose instead a new definition of the Internet, based on a
different unifying concept:
* "Old" Internet concept: IP-based.
The organizing principle is the IP address, i.e., a common
network address space.
* "New" Internet concept: Application-based.
The organizing principle is the domain name system and
Directories, i.e., a common - albeit necessarily multiform -
application name space.
This suggests that the idea of "connected status", which has
traditionally been tied to the IP address(via network numbers,
should instead be coupled to the names and related identifying
information contained in the distributed Internet directory.
A naming-based definition of "the Internet" implies a much larger
Internet community, and a much more dynamic (and unpredictable)
operational Internet. This argues for an Internet architecture
based on adaptability (to a broad spectrum of possible future
developments) rather than anticipation.
3.2 A Process-Based Model of the Multiprotocol Internet
Rather than specify a particular "multi-protocol Internet",
embracing a pre-determined number of specific protocol
architectures, we propose instead a process-oriented model of the
Internet, which accommodates different protocol architectures
according to the traditional "things that work" principle.
A process-oriented Internet model includes, as a basic postulate,
the assertion that there is no *steady-state* "multi-protocol
Internet". The most basic forces driving the evolution of the
Internet are pushing it not toward multi-protocol diversity, but
toward the original state of protocol-stack uniformity (although
it is unlikely that it will ever actually get there). We may
represent this tendency of the Internet to evolve towards
homogeneity as the most "thermodynamically stable" state by
describing four components of a new process-based Internet
architecture:
Part 1: The core Internet architecture
This is the traditional TCP/IP-based architecture. It is the
"magnetic center" of Internet evolution, recognizing that (a)
homogeneity is still the best way to deal with diversity in
an internetwork, and (b) IP connectivity is still the best
basic model of the Internet (whether or not the actual state
of IP ubiquity can be achieved in practice in a global
operational Internet).
"In the beginning", the Internet architecture consisted only of
this first part. The success of the Internet, however, has
carried it beyond its uniform origins; ubiquity and uniformity
have been sacrificed in order to greatly enrich the Internet "gene
pool".
Two additional parts of the new Internet architecture express the
ways in which the scope and extent of the Internet have been
expanded.
Part 2: Link sharing
Here physical resources -- transmission media, network
interfaces, perhaps some low-level (link) protocols -- are
shared by multiple, non-interacting protocol suites. This
part of the architecture recognizes the necessity and
convenience of coexistence, but is not concerned with
interoperability; it has been called "ships in the night" or
"S.I.N.".
Coexisting protocol suites are not, of course, genuinely
isolated in practice; the ships passing in the night raise
issues of management, non-interference, coordination, and
fairness in real Internet systems.
Part 3: Application interoperability
Absent ubiquity of interconnection (i.e., interoperability of
the "underlying stacks"), it is still possible to achieve
ubiquitous application functionality by arranging for the
essential semantics of applications to be conveyed among
disjoint communities of Internet systems. This can be
accomplished by application relays, or by user agents that
present a uniform virtual access method to different
application services by expressing only the shared semantics.
This part of the architecture emphasizes the ultimate role of
the Internet as a basis for communication among applications,
rather than as an end in itself. To the extent that it
enables a population of applications and their users to move
from one underlying protocol suite to another without
unacceptable loss of functionality, it is also a "transition
enabler".
Adding parts 2 and 3 to the original Internet architecture is at
best a mixed blessing. Although they greatly increase the scope
of the Internet and the size of the Internet community, they also
introduce significant problems of complexity, cost, and
management, and they usually represent a loss of functionality
(particularly with respect to part 3). Parts 2 and 3 represent
unavoidable, but essentially undesirable, departures from the
homogeneity represented by part 1. Some functionality is lost,
and additional system complexity and cost is endured, in order to
expand the scope of the Internet. In a perfect world, however,
the Internet would evolve and expand without these penalties.
There is a tendency, therefore, for the Internet to evolve in
favor of the homogeneous architecture represented by part 1, and
away from the compromised architectures of parts 2 and 3. Part 4
expresses this tendency.
Part 4: Hybridization/Integration.
Part 4 recognizes the desirability of integrating similar
elements from different Internet protocol architectures to
form hybrids that reduce the variability and complexity of
the Internet system. It also recognizes the desirability of
leveraging the existing Internet infrastructure to facilitate
the absorption of "new stuff" into the Internet, applying to
"new stuff" the established Internet practice of test,
evaluate, adopt.
This part expresses the tendency of the Internet, as a
system, to attempt to return to the original "state of grace"
represented by the uniform architecture of part 1. It is a
force acting on the evolution of the Internet, although the
Internet will never actually return to a uniform state at any
point in the future.
According to this dynamic process model, running X.400 mail over
RFC1006 on a TCP/IP stack, integrated IS-IS routing, transport
gateways, and the development of a single common successor to the
IP and CLNP protocols are all examples of "good things". They
represent movement away from the non-uniformity of parts 2 and 3
towards greater homogeneity, under the influence of the "magnetic
field" asserted by part 1, following the hybridization dynamic of
part 4.
4. SECURITY ARCHITECTURE
4.1 Philosophical Guidelines
The principal themes for development of an Internet security
architecture are simplicity, testability, trust, technology and
security perimeter identification.
* There is more to security than protocols and cryptographic
methods.
* The security architecture and policies should be simple
enough to be readily understood. Complexity breeds
misunderstanding and poor implementation.
* The implementations should be testable to determine if the
policies are met.
* We are forced to trust hardware, software and people to make
any security architecture function. We assume that the
technical instruments of security policy enforcement are at
least as powerful as modern personal computers and work
stations; we do not require less capable components to be
self-protecting (but might apply external remedies such as
link level encryption devices).
* Finally, it is essential to identify security perimeters at
which protection is to be effective.
4.2 Security Perimeters
There were four possible security perimeters: link level,
net/subnet level, host level, and process/application level. Each
imposes different requirements, can admit different techniques,
and makes different assumptions about what components of the
system must be trusted to be effective.
Privacy Enhanced Mail is an example of a process level security
system; providing authentication and confidentiality for SNMP is
another example. Host level security typically means applying an
external security mechanism on the communication ports of a host
computer. Network or subnetwork security means applying the
external security capability at the gateway/router(s) leading from
the subnetwork to the "outside". Link-level security is the
traditional point-to-point or media-level (e.g., Ethernet)
encryption mechanism.
There are many open questions about network/subnetwork security
protection, not the least of which is a potential mismatch between
host level (end/end) security methods and methods at the
network/subnetwork level. Moreover, network level protection does
not deal with threats arising within the security perimeter.
Applying protection at the process level assumes that the
underlying scheduling and operating system mechanisms can be
trusted not to prevent the application from applying security when
appropriate. As the security perimeter moves downward in the
system architecture towards the link level, one must make many
assumptions about the security threat to make an argument that
enforcement at a particular perimeter is effective. For example,
if only link-level encryption is used, one must assume that
attacks come only from the outside via communications lines, that
hosts, switches and gateways are physically protected, and the
people and software in all these components are to be trusted.
4.3 Desired Security Services
We need authenticatable distinguished names if we are to implement
discretionary and non-discretionary access control at application
and lower levels in the system. In addition, we need enforcement
for integrity (anti-modification, anti-spoof and anti-replay
defenses), confidentiality, and prevention of denial-of-service.
For some situations, we may also need to prevent repudiation of
message transmission or to prevent covert channels.
We have some building blocks with which to build the Internet
security system. Cryptographic algorithms are available (e.g.,
Data Encryption Standard, RSA, El Gamal, and possibly other public
key and symmetric key algorithms), as are hash functions such as
MD2 and MD5.
We need Distinguished Names (in the OSI sense) and are very much
in need of an infrastructure for the assignment of such
identifiers, together with widespread directory services for
making them known. Certificate concepts binding distinguished
names to public keys and binding distinguished names to
capabilities and permissions may be applied to good advantage.
At the router/gateway level, we can apply address and protocol
filters and other configuration controls to help fashion a
security system. The proposed OSI Security Protocol 3 (SP3) and
Security Protocol 4 (SP4) should be given serious consideration as
possible elements of an Internet security architecture.
Finally, it must be observed that we have no good solutions to
safely storing secret information (such as the secret component of
a public key pair) on systems like PCs or laptop computers that
are not designed to enforce secure storage.
4.4 Proposed Actions
The following actions are proposed.
A) Security Reference Model
A Security Reference Model for the Internet is needed, and it
should be developed expeditiously. This model should
establish the target perimeters and document the objectives
of the security architecture.
B) Privacy-Enhanced Mail (PEM)
For Privacy Enhanced Mail, the most critical steps seem to be
the installation of (1) a certificate generation and
management infrastructure, and (2) X.500 directory services
to provide access to public keys via distinguished names.
Serious attention also needs to be placed on any limitations
imposed by patent and export restrictions on the deployment
of this system.
C) Distributed System Security
We should examine security methods for distributed systems
applications, in both simple (client/server) and complex
(distributed computing environment) cases. For example, the
utility of certificates granting permissions/capabilities to
objects bound to distinguished names should be examined.
D) Host-Level Security
SP4 should be evaluated for host-oriented security, but SP3
should also be considered for this purpose.
E) Application-Level Security
We should implement application-level security services, both
for their immediate utility (e.g., PEM, SNMP authentication)
and also to gain valuable practical experience that can
inform the refinement of the Internet security architecture.
5. TRAFFIC CONTROL AND STATE
In the present Internet, all IP datagrams are treated equally. Each
datagram is forwarded independently, regardless of any relationship
it has to other packets for the same connection, for the same
application, for the same class of applications, or for the same user
class. Although Type-of-Service and Precedence bits are defined in
the IP header, these are not generally implemented, and in fact it is
not clear how to implement them.
It is now widely accepted that the future Internet will need to
support important applications for which best-effort is not
sufficient -- e.g., packet video and voice for teleconferencing.
This will require some "traffic control" mechanism in routers,
controlled by additional state, to handle "real-time" traffic.
5.1 Assumptions and Principles
o ASSUMPTION: The Internet will need to support performance
guarantees for particular subsets of the traffic.
Unfortunately, we are far from being able to give precise meanings
to the terms "performance", "guarantees", or "subsets" in this
statement. Research is still needed to answer these questions.
o The default service will continue to be the current "best-
effort" datagram delivery, with no service guarantees.
o The mechanism of a router can be separated into (1) the
forwarding path and (2) the control computations (e.g.,
routing) which take place in the background.
The forwarding path must be highly optimized, sometimes with
hardware-assist, and it is therefore relatively costly and
difficult to change. The traffic control mechanism operates
in the forwarding path, under the control of state created by
routing and resource control computations that take place in
background. We will have at most one shot at changing the
forwarding paths of routers, so we had better get it right
the first time.
o The new extensions must operate in a highly heterogeneous
environment, in which some parts will never support
guarantees. For some hops of a path (e.g., a high-speed
LAN), "over-provisioning" (i.e., excess capacity) will allow
adequate service for real-time traffic, even when explicit
resource reservation is unavailable.
o Multicast distribution is probably essential.
5.2 Technical Issues
There are a number of technical issues to be resolved, including:
o Resource Setup
To support real-time traffic, resources need to be reserved
in each router along the path from source to destination.
Should this new router state be "hard" (as in connections) or
"soft" (i.e., cached state)?
o Resource binding vs. route binding
Choosing a path from source to destination is traditionally
performed using a dynamic routing protocol. The resource
binding and the routing might be folded into a single complex
process, or they might be performed essentially
independently. There is a tradeoff between complexity and
efficiency.
o Alternative multicast models
IP multicasting uses a model of logical addressing in which
targets attach themselves to a group. In ST-2, each host in
a multicast session includes in its setup packet an explicit
list of target addresses. Each of these approaches has
advantages and drawbacks; it is not currently clear which
will prevail for n-way teleconferences.
o Resource Setup vs. Inter-AD routing
Resource guarantees of whatever flavor must hold across an
arbitrary end-to-end path, including multiple ADs. Hence,
any resource setup mechanism needs to mesh smoothly with the
path setup mechanism incorporated into IDPR.
o Accounting
The resource guarantee subsets ("classes") may be natural
units for accounting.
5.3 Proposed Actions
The actions called for here are further research on the technical
issues listed above, followed by development and standardization
of appropriate protocols. DARTnet, the DARPA Research Testbed
network, will play an important role in this research.
6. ADVANCED APPLICATIONS
One may ask: "What network-based applications do we want, and why
don't we have them now?" It is easy to develop a large list of
potential applications, many of which would be based on a
client/server model. However, the more interesting part of the
question is: "Why haven't people done them already?" We believe the
answer to be that the tools to make application writing easy just do
not exist.
To begin, we need a set of common interchange formats for a number of
data items that will be used across the network. Once these common
data formats have been defined, we need to develop tools that the
applications can use to move the data easily.
6.1 Common Interchange Formats
The applications have to know the format of information that they
are exchanging, for the information to have any meaning. The
following format types are to concern:
(1) Text - Of the formats in this list, text is the most stable,
but today's international Internet has to address the needs
of character sets other than USASCII.
(2) Image - As we enter the "Multimedia Age", images will become
increasingly important, but we need to agree on how to
represent them in packets.
(3) Graphics - Like images, vector graphic information needs a
common definition. With such a format we could exchange
things like architectural blueprints.
(4) Video - Before we can have a video window running on our
workstation, we need to know the format of that video
information coming over the network.
(5) Audio/Analog - Of course, we also need the audio to go with
the video, but such a format would be used for representation
of all types of analog signals.
(6) Display - Now that we are opening windows on our workstation,
we want to open a window on another person's workstation to
show her some data pertinent to the research project, so now
we need a common window display format.
(7) Data Objects - For inter-process communications we need to
agree on the formats of things like integers, reals, strings,
etc.
Many of these formats are being defined by other, often several
other, standards organizations. We need to agree on one format
per category for the Internet.
6.2 Data Exchange Methods
Applications will require the following methods of data exchange.
(1) Store and Forward
Not everyone is on the network all the time. We need a
standard means of providing an information flow to
sometimes-connected hosts, i.e., we need a common store-and-
forward service. Multicasting should be included in such a
service.
(2) Global File Systems
Much of the data access over the network can be broken down
to simple file access. If you had a real global file system
where you access any file on the Internet (assuming you have
permission), would you ever need FTP?
(3) Inter-process Communications
For a true distributed computing environment, we need the
means to allow processes to exchange data in a standard
method over the network. This requirement encompasses RPC,
APIs, etc.
(4) Data Broadcast
Many applications need to send the same information to many
other hosts. A standard and efficient method is needed to
accomplish this.
(5) Database Access
For good information exchange, we need to have a standard
means for accessing databases. The Global File System can get
you to the data, but the database access methods will tell
you about its structure and content.
Many of these items are being addressed by other organizations,
but for Internet interoperability, we need to agree on the methods
for the Internet.
Finally, advanced applications need solutions to the problems of
two earlier areas in this document. From the Traffic Control and
State area, applications need the ability to transmit real-time
data. This means some sort of expectation level for data delivery
within a certain time frame. Applications also require global
authentication and access control systems from the Security area.
Much of the usefulness of today's Internet applications is lost
due to the lack of trust and security. This needs to be solved
for tomorrow's applications.
7. REFERENCES
[1] Cerf, V. and R. Kahn, "A Protocol for Packet Network
Intercommunication," IEEE Transactions on Communication, May
1974.
[2] Postel, J., Sunshine, C., and D. Cohen, "The ARPA Internet
Protocol," Computer Networks, Vol. 5, No. 4, July 1981.
[3] Leiner, B., Postel, J., Cole, R., and D. Mills, "The DARPA
Internet Protocol Suite," Proceedings INFOCOM 85, IEEE,
Washington DC, March 1985. Also in: IEEE Communications
Magazine, March 1985.
[4] Clark, D., "The Design Philosophy of the DARPA Internet
Protocols", Proceedings ACM SIGCOMM '88, Stanford, California,
August 1988.
[5] Mogul, J., and J. Postel, "Internet Standard Subnetting
Procedure", RFC950, USC/Information Sciences Institute, August
1985.
[6] Mockapetris, P., "Domain Names - Concepts and Facilities", RFC
1034, USC/Information Sciences Institute, November 1987.
[7] Deering, S., "Host Extensions for IP Multicasting", RFC1112,
Stanford University, August 1989.
[8] "Proceedings of the Twenty-First Internet Engineering Task
Force", Bell-South, Atlanta, July 29 - August 2, 1991.
APPENDIX A: Setting the Stage
Slide 1
WHITHER THE INTERNET?
OPTIONS FOR ARCHITECTURE
IAB/IESG -- Jan 1990
David D. Clark
__________________________________________________________________
Slide 2
SETTING THE TOPIC OF DISCUSSION
Goals:
o Establish a common frame of understanding for
IAB, IESG and the Internet community.
o Understand the set of problems to be solved.
o Understand the range of solutions open to us.
o Draw some conclusions, or else
"meta-conclusions".
__________________________________________________________________
Slide 3
SOME CLAIMS -- MY POSITION
We have two different goals:
o Make it possible to build "The Internet"
o Define a protocol suite called Internet
Claim: These goals have very different implications.
The protocols are but a means, though a powerful one.
Claim: If "The Internet" is to succeed and grow, it will
require specific design efforts. This need will continue
for at least another 10 years.
Claim: Uncontrolled growth could lead to chaos.
Claim: A grass-roots solution seems to be the only
means to success. Top-down mandates are powerless.
__________________________________________________________________
Slide 4
OUTLINE OF PRESENTATION
1) The problem space and the solution space.
2) A set of specific questions -- discussion.
3) Return to top-level questions -- discussion.
4) Plan for action -- meta discussion.
Try to separate functional requirements from technical approach.
Understand how we are bounded by our problem space and our
solution space.
Is architecture anything but protocols?
__________________________________________________________________
Slide 5
WHAT IS THE PROBLEM SPACE?
Routing and addressing:
How big, what topology, and what routing model?
Getting big:
User services, what technology for host and nets?
Divestiture of the Internet:
Accounting, controlling usage and fixing faults.
New services:
Video? Transactions? Distributed computing?
Security:
End node or network? Routers or relays?
__________________________________________________________________
Slide 6
BOUNDING THE SOLUTION SPACE
How far can we migrate from the current state?
o Can we change the IP header (except to OSI)?
o Can we change host requirements in mandatory ways?
o Can we manage a long-term migration objective?
- Consistent direction vs. diverse goals, funding.
Can we assume network-level connectivity?
o Relays are the wave of the future (?)
o Security a key issue; along with conversion.
o Do we need a new "relay-based" architecture?
How "managed" can/must "The Internet" be?
o Can we manage or constrain connectivity?
What protocols are we working with? One or many?
__________________________________________________________________
Slide 7
THE MULTI-PROTOCOL INTERNET
"Making the problem harder for the good of mankind."
Are we migrating, interoperating, or tolerating multiple protocols?
o Not all protocol suites will have same range of functionality
at the same time.
o "The Internet" will require specific functions.
Claim: Fundamental conflict (not religion or spite):
o Meeting aggressive requirements for the Internet
o Dealing with OSI migration.
Conclusion: One protocol must "lead", and the others must follow.
When do we "switch" to OSI?
Consider every following slide in this context.
__________________________________________________________________
Slide 8
ROUTING and ADDRESSING
What is the target size of "The Internet"?
o How do addresses and routes relate?
o What is the model of topology?
o What solutions are possible?
What range of policy routing is required?
o BGP and IDRP are two answers. What is the question?
o Fixed classes, or variable paths?
o Source controlled routing is a minimum.
How seamless is the needed support for mobile hosts?
o New address class, rebind to local address, use DNS?
Shall we push for Internet multicast?
__________________________________________________________________
Slide 9
GETTING BIG -- AN OLD TITLE
(Addressing and routing was on previous slide...)
What user services will be needed in the next 10 years?
o Can we construct a plan?
o Do we need architectural changes?
Is there a requirement for dealing better with ranges in
speed, packet sizes, etc.
o Policy to phase out fragmentation?
What range of hosts (things != Unix) will we support?
_________________________________________________________________
Slide 10
DEALING WITH DIVESTITURE
The Internet is composed of parts separately managed and
controlled.
What support is needed for network charging?
o No architecture implies bulk charges and re-billing, pay
for lost packets.
o Do we need controls to supply billing id or routing?
Requirement: we must support links with controlled sharing.
(Simple form is classes based on link id.)
o How general?
Is there an increased need for fault isolation? (I vote yes!)
o How can we find managers to talk to?
o Do we need services in hosts?
_________________________________________________________________
Slide 11
NEW SERVICES
Shall we support video and audio? Real time? What %?
o Need to plan for input from research. What quality?
o Target date for heads-up to vendors.
Shall we "better" support transactions?
o Will TCP do? VMTP? Presentation? Locking?
What application support veneers are coming?
o Distributed computing -- will it actually happen?
o Information networking?
__________________________________________________________________
Slide 12
SECURITY
Can we persist in claiming the end-node is the only line of defense?
o What can we do inside the network?
o What can ask the host to do?
Do we tolerate relays, or architect them?
Can find a better way to construct security boundaries?
Do we need global authentication?
Do we need new host requirements:
o Logging.
o Authentication.
o Management interfaces.
- Phone number or point of reference.
__________________________________________________________________
APPENDIX B: Group Membership
Group 1: ROUTING AND ADDRESSING
Dave Clark, MIT [Chair]
Hans-Werner Braun, SDSC
Noel Chiappa, Consultant
Deborah Estrin, USC
Phill Gross, CNRI
Bob Hinden, BBN
Van Jacobson, LBL
Tony Lauck, DEC.
Group 2: MULTI-PROTOCOL ARCHITECTURE
Lyman Chapin, BBN [Chair]
Ross Callon, DEC
Dave Crocker, DEC
Christian Huitema, INRIA
Barry Leiner,
Jon Postel, ISI
Group 3: SECURITY ARCHITECTURE
Vint Cerf, CNRI [Chair]
Steve Crocker, TIS
Steve Kent, BBN
Paul Mockapetris, DARPA
Group 4: TRAFFIC CONTROL AND STATE
Robert Braden, ISI [Chair]
Chuck Davin, MIT
Dave Mills, University of Delaware
Claudio Topolcic, CNRI
Group 5: ADVANCED APPLICATIONS
Russ Hobby, UCDavis [Chair]
Dave Borman, Cray Research
Cliff Lynch, University of California
Joyce K. Reynolds, ISI
Bruce Schatz, University of Arizona
Mike Schwartz, University of Colorado
Greg Vaudreuil, CNRI.
Security Considerations
Security issues are discussed in Section 4.
Authors' Addresses
David D. Clark
Massachusetts Institute of Technology
Laboratory for Computer Science
545 Main Street
Cambridge, MA 02139
Phone: (617) 253-6003
EMail: ddc@LCS.MIT.EDU
Vinton G. Cerf
Corporation for National Research Initiatives
1895 Preston White Drive, Suite 100
Reston, VA 22091
Phone: (703) 620-8990
EMail: vcerf@nri.reston.va.us
Lyman A. Chapin
Bolt, Beranek & Newman
Mail Stop 20/5b
150 Cambridge Park Drive
Cambridge, MA 02140
Phone: (617) 873-3133
EMail: lyman@BBN.COM
Robert Braden
USC/Information Sciences Institute
4676 Admiralty Way
Marina del Rey, CA 90292
Phone: (310) 822-1511
EMail:
braden@isi.edu
Russell Hobby
University of California
Computing Services
Davis, CA 95616
Phone: (916) 752-0236
EMail: rdhobby@ucdavis.edu
