📄 rfc1287.txt
字号:
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 RFC discusses 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
Clark, Chapin, Cerf, Braden, & Hobby [Page 1]
RFC 1287 Future of Internet Architecture December 1991
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
Clark, Chapin, Cerf, Braden, & Hobby [Page 2]
RFC 1287 Future of Internet Architecture December 1991
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.
Clark, Chapin, Cerf, Braden, & Hobby [Page 3]
RFC 1287 Future of Internet Architecture December 1991
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.
Clark, Chapin, Cerf, Braden, & Hobby [Page 4]
RFC 1287 Future of Internet Architecture December 1991
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
Clark, Chapin, Cerf, Braden, & Hobby [Page 5]
RFC 1287 Future of Internet Architecture December 1991
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
Clark, Chapin, Cerf, Braden, & Hobby [Page 6]
RFC 1287 Future of Internet Architecture December 1991
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.
Clark, Chapin, Cerf, Braden, & Hobby [Page 7]
RFC 1287 Future of Internet Architecture December 1991
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
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -