Network Working Group S. Floyd, Editor
Request for Comments: 3426 Internet Architecture Board
Category: Informational November 2002
General Architectural and Policy Considerations
Status of this Memo
This memo provides information for the Internet community. It does
not specify an Internet standard of any kind. Distribution of this
memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2002). All Rights Reserved.
Abstract
This document suggests general architectural and policy questions
that the IETF community has to address when working on new standards
and protocols. We note that this document contains questions to be
addressed, as opposed to guidelines or architectural principles to be
followed.
1. Introduction
This document suggests general architectural and policy questions to
be addressed in our work in the IETF. This document contains
questions to be addressed, as opposed to guidelines or architectural
principles to be followed. These questions are somewhat similar to
the "Security Considerations" currently required in IETF documents
[RFC2316].
This document is motivated in part by concerns about a growing lack
of coherence in the overall Internet architecture. We have moved
from a world of a single, coherent architecture designed by a small
group of people, to a world of complex, intricate architecture to
address a wide-spread and heterogeneous environment. Because
individual pieces of the architecture are often designed by
sub-communities, with each sub-community having its own set of
interests, it is necessary to pay increasing attention to how each
piece fits into the larger picture, and to consider how each piece is
chosen. For example, it is unavoidable that each of us is inclined
to solve a problem at the layer of the protocol stack and using the
tools that we understand the best; that does not necessarily mean
that this is the most appropriate layer or set of tools for solving
this problem in the long-term.
Floyd Informational [Page 1]
RFC 3426 Architectural and Policy Considerations November 2002
Our assumption is that this document will be used as suggestions (but
not a checklist!) of issues to be addressed by IETF members in
chartering new working groups, in working in working groups, and in
evaluating the output from other working groups. This document is
not a primer on how to design protocols and architectures, and does
not provide answers to anything.
2. Relationship to "Architectural Principles of the Internet"
RFC 1958 [RFC1958] outlines some architectural principles of the
Internet, as "guidelines that have been found useful in the past, and
that may be useful to those designing new protocols or evaluating
such designs." An example guideline is that "it is also generally
felt that end-to-end functions can best be realized by end-to-end
protocols." Similarly, an example design issue from [RFC1958] is that
"heterogeneity is inevitable and must be supported by design."
In contrast, this document serves a slightly different purpose, by
suggesting additional architectural questions to be addressed. Thus,
one question suggested in this document is the following: "Is this
proposal the best long-term solution to the problem? If not, what
are the long-term costs of this solution, in terms of restrictions on
future development, if any?" This question could be translated to a
roughly equivalent architectural guideline, as follows: "Identify
whether the proposed protocol is a long-term solution or a short-term
solution, and identify the long-term costs and the exit strategy for
any short-term solutions."
In contrast, other questions are more open-ended, such as the
question about robustness: "How robust is the protocol, not just to
the failure of nodes, but also to compromised or malfunctioning
components, imperfect or defective implementations, etc?" As a
community, we are still learning about the degree of robustness that
we are able to build into our protocols, as well as the tools that
are available to ensure this robustness. Thus, there are not yet
clear architectural guidelines along the lines of "Ensure that your
protocol is robust against X, Y, and Z."
3. Questions
In this section we list some questions to ask in designing protocols.
Each question is discussed more depth in the rest of this paper. We
aren't suggesting that all protocol design efforts should be required
to explicitly answer all of these questions; some questions will be
more relevant to one document than to another. We also aren't
suggesting that this is a complete list of architectural concerns.
Floyd Informational [Page 2]
RFC 3426 Architectural and Policy Considerations November 2002
DESIGN QUESTIONS:
Justifying the Solution:
* Why are you proposing this solution, instead of proposing something
else, or instead of using existing protocols and procedures?
Interactions between Layers:
* Why are you proposing a solution at this layer of the protocol
stack, rather than at another layer? Are there solutions at other
layers of the protocol stack as well?
* Is this an appropriate layer in terms of correctness of function,
data integrity, performance, ease of deployment, the diagnosability
of failures, and other concerns?
* What are the interactions between layers, if any?
Long-term vs. Short-term Solutions:
* Is this proposal the best long-term solution to the problem?
* If not, what are the long-term costs of this solution, in terms of
restrictions on future development, if any? What are the
requirements for the development of longer-term solutions?
The Whole Picture vs. Building Blocks:
* Have you considered the larger context, while appropriately
restricting your own design efforts to one part of the whole?
* Are there parts of the overall solution that will have to be
provided by other IETF Working Groups or by other standards bodies?
EVALUATION QUESTIONS:
Weighing Benefits against Costs:
* How do the architectural benefits of a proposed new protocol
compare against the architectural costs, if any? Have the
architectural costs been carefully considered?
Robustness:
* How robust is the protocol, not just to the failure of nodes, but
also to compromised or malfunctioning components, imperfect or
defective implementations, etc?
Floyd Informational [Page 3]
RFC 3426 Architectural and Policy Considerations November 2002
* Does the protocol take into account the realistic conditions of the
current or future Internet (e.g., packet drops and packet corruption;
packet reordering; asymmetric routing; etc.)?
Tragedy of the Commons:
* Is performance still robust if everyone is using this protocol?
Are there other potential impacts of widespread deployment that need
to be considered?
Protecting Competing Interests:
* Does the protocol protect the interests of competing parties (e.g.,
not only end-users, but also ISPs, router vendors, software vendors,
or other parties)?
Designing for Choice vs. Avoiding Unnecessary Complexity:
* Is the protocol designed for choice, to allow different players to
express their preferences where appropriate? At the other extreme,
does the protocol provide so many choices that it threatens
interoperability or introduces other significant problems?
Preserving Evolvability?
* Does the protocol protect the interests of the future, by
preserving the evolvability of the Internet? Does the protocol
enable future developments?
* If an old protocol is overloaded with new functionality, or reused
for new purposes, have the possible complexities introduced been
taken carefully into account?
* For a protocol that introduces new complexity to the Internet
architecture, how does the protocol add robustness and preserve
evolvability, and how does it also introduce new fragilities to the
system?
Internationalization:
* Where protocols require elements in text format, have the possibly
conflicting requirements of global comprehensibility and the ability
to represent local text content been properly weighed against each
other?
Floyd Informational [Page 4]
RFC 3426 Architectural and Policy Considerations November 2002
DEPLOYMENT QUESTIONS:
* Is the protocol deployable?
Each of these questions is discussed in more depth in the rest of
this paper.
4. Justifying the Solution
Question: Why are you proposing this solution, instead of proposing
something else, or instead of using existing protocols and
procedures?
4.1. Case study: Integrated and Differentiated Services
A good part of the work of developing integrated and differentiated
services has been to understand the problem to be solved, and to come
to agreement on the architectural framework of the solution, and on
the nature of specific services. Thus, when integrated services were
being developed, the specification of the Controlled Load [RFC2211]
and Guaranteed [RFC2212] services each required justification of the
need for that particular service, of low loss and bounded delay
respectively.
Later, when RFC 2475 on "An Architecture for Differentiated Services"
proposed a scalable, service differentiation architecture that
differs from the previously-defined architecture for integrated
services, the document also had to clearly justify the requirements
for this new architecture, and compare the proposed architecture to
other possible approaches [RFC2475]. Similarly, when the Assured
Forwarding [RFC2597] and Expedited Forwarding [RFC3246] Per-Hop
Behaviors of differentiated services were proposed, each service
required a justification of the need for that service in the
Internet.
5. Interactions between Layers
Questions: Why are you proposing a solution at this layer of the
protocol stack, rather than at another layer? Are there solutions at
other layers of the protocol stack as well?
Is this an appropriate layer in terms of correctness of function,
data integrity, performance, ease of deployment, the diagnosability
of failures, and other concerns?
What are the interactions between layers, if any?
Floyd Informational [Page 5]
RFC 3426 Architectural and Policy Considerations November 2002
5.1. Discussion: The End-to-End Argument
The classic 1984 paper on "End-To-End Arguments In System Design"
[SRC84] begins a discussion of where to place functions among modules
by suggesting that "functions placed at low levels of a system may be
redundant or of little value when compared with the cost of providing
them at that low level. Examples discussed in the paper include bit
error recovery, security using encryption, duplicate message
suppression, recovery from system crashes, and delivery
acknowledgement. Low level mechanisms to support these functions are
justified only as performance enhancements." The end-to-end
principle is one of the key architectural guidelines to consider in
choosing the appropriate layer for a function.
5.2. Case study: Endpoint Congestion Management
The goal of the Congestion Manager in Endpoint Congestion Management
is to allow multiple concurrent flows with the same source and
destination address to share congestion control state [RFC3124].
There has been a history of proposals for multiplexing flows at
different levels of the protocol stack; proposals have included
adding multiplexing at the HTTP (WebMux) and TCP (TCP Control Blocks)
layers, as well as below TCP (the Congestion Manager) [Multiplexing].
However, the 1989 article on "Layered Multiplexing Considered
Harmful" suggests that "the extensive duplication of multiplexing
functionality across the middle and upper layers is harmful and
should be avoided" [T89]. Thus, one of the key issues in providing
mechanisms for multiplexing flows is to determine which layer or
layers of the protocol stack are most appropriate for providing this
functionality. The natural tendency of each researcher is generally
to add functionality at the layer that they know the best; this does
not necessarily result in the most appropriate overall architecture.
5.3. Case study: Layering Applications on Top of HTTP
There has been considerable interest in layering applications on top
of HTTP [RFC3205]. Reasons cited include compatibility with widely-
deployed browsers, the ability to reuse client and server libraries,
the ability to use existing security mechanisms, and the ability to
traverse firewalls. As RFC 3205 discusses, "the recent interest in
layering new protocols over HTTP has raised a number of questions
when such use is appropriate, and the proper way to use HTTP in
contexts where it is appropriate." Thus, RFC 3205 addresses not only
the benefits of layering applications on top of HTTP, but also
evaluates the additional complexity and overhead of layering an
application on top of HTTP, compared to the costs of introducing a
special-purpose protocol.
Floyd Informational [Page 6]
RFC 3426 Architectural and Policy Considerations November 2002
The web page on "References on Layering and the Internet
Architecture" has pointers to additional papers discussing general
layering issues in the Internet architecture [Layering].
6. Long-term vs. Short-term Solutions
Questions: Is this proposal the best long-term solution to the
problem?
If not, what are the long-term costs of this solution, in terms of
restrictions on future development, if any? What are the
requirements for the development of longer-term solutions?
6.1. Case study: Traversing NATs
In order to address problems with NAT middleboxes altering the
external address of endpoints, various proposals have been made for
mechanisms where an originating process attempts to determine the
address (and port) by which it is known on the other side of a NAT.
This would allow an originating process to be able to use address
data in the protocol exchange, or to advertise an external address
from which it will receive connections.
The IAB in [RFC3424] has outlined reasons why these proposals can be
considered, at best, short-term fixes to specific problems, and the
specific issues to be carefully evaluated before standardizing such
proposals. These issues include the identification of the
limited-scope problem to be fixed, the description of an exit
strategy for the short-term solution, and the description of the
longer-term problems left unsolved by the short-term solution.
7. Looking at the whole picture vs. making a building block
For a complex protocol which interacts with protocols from other
standards bodies as well as from other IETF working groups, it can be
necessary to keep in mind the overall picture while, at the same
time, breaking out specific parts of the problem to be standardized
in particular working groups.
Question: Have you considered the larger context, while restricting
your own design efforts to one part of the whole?
Question: Are there parts of the overall solution that will have to
be provided by other IETF Working Groups or by other standards
bodies?
Floyd Informational [Page 7]
RFC 3426 Architectural and Policy Considerations November 2002
7.1. Case Study: The Session Initiation Protocol (SIP)
The Session Initiation Protocol (SIP) [RFC2543], for managing
connected, multimedia sessions, is an example of a complex protocol
that has been broken into pieces for standardization in other working
groups. SIP has also involved interaction with other standardization
bodies.
The basic SIP framework is being standardized by the SIP working
group. This working group has focused on the core functional
features of setting up, managing, and tearing down multimedia
sessions. Extensions are considered if they relate to these core
features.
The task of setting up a multimedia session also requires a
description of the desired multimedia session. This is provided by
the Session Description Protocol (SDP). SDP is a building block that
is supplied by the Multiparty Multimedia Session Control (MMUSIC)
working group. It is not standardized within the SIP working group.
Other working groups are involved in standardizing extensions to SIP
that fall outside of core functional features or applications. The
SIPPING working group is analyzing the requirements for SIP applied
to different tasks, and the SIMPLE working group is examining the
application of SIP to instant messaging and presence. The IPTEL
working group is defining a call processing language (CPL) that
interacts with SIP in various ways. These working groups
occasionally feed requirements back into the main SIP working group.
Finally, outside standardization groups have been very active in
providing the SIP working group with requirements. The Distributed
Call Signaling (DCS) group from the PacketCable Consortium, 3GPP, and
3GPP2 are all using SIP for various telephony-related applications,
and members of these groups have been involved in drafting
requirements for SIP. In addition, there are extensions of SIP which
are under consideration in these standardization bodies. Procedures
are under development in the IETF to address proposed extensions to
SIP, including a category of preliminary, private, or proprietary
extensions, and to provide for the safe management of the SIP
namespace [MBMWOR02].
8. Weighing architectural benefits against architectural costs
Questions: How do the architectural benefits of a proposed new
protocol compare against the architectural costs, if any? Have the
architectural costs been carefully considered?
Floyd Informational [Page 8]
RFC 3426 Architectural and Policy Considerations November 2002
8.1. Case Study: Performance-enhancing proxies (PEPs)
RFC 3135 [RFC3135] considers the relative costs and benefits of
placing performance-enhancing proxies (PEPs) in the middle of a
network to address link-related degradations. In the case of PEPs,
the potential costs include disabling the end-to-end use of IP layer
security mechanisms; introducing a new possible point of failure that
is not under the control of the end systems; adding increased
difficulty in diagnosing and dealing with failures; and introducing
possible complications with asymmetric routing or mobile hosts. RFC
3135 carefully considers these possible costs, the mitigations that
can be introduced, and the cases when the benefits of
performance-enhancing proxies to the user are likely to outweigh the
costs.
8.2. Case Study: Open Pluggable Edge Services (OPES)
One of the issues raised by middleboxes in the Internet involves the
end-to-end integrity of data. This is illustrated in the recent
question of chartering the Open Pluggable Edge Services (OPES)
Working Group. Open Pluggable Edge Services are services that would
be deployed as application-level intermediaries in the network, for
example, at a web proxy cache between the origin server and the
client. These intermediaries would transform or filter content, with
the explicit consent of either the content provider or the end user.
One of the architectural issues that arose in the process of
chartering the OPES Working Group concerned the end-to-end integrity
of data. As an example, it was suggested that "OPES would reduce
both the integrity, and the perception of integrity, of
communications over the Internet, and would significantly increase
uncertainly about what might have been done to content as it moved
through the network", and that therefore the risks of OPES outweighed
the benefits [CDT01].
As one consequence of this debate, the IAB wrote a document on "IAB
Architectural and Policy Considerations for OPES", considering both
the potential architectural benefits and costs of OPES [RFC3238].
This document did not recommend specific solutions or mandate
specific functional requirements, but instead included
recommendations of issues such as concerns about data integrity that
OPES solutions standardized in the IETF should be required to
address.
Floyd Informational [Page 9]
RFC 3426 Architectural and Policy Considerations November 2002
9. General Robustness Questions
Questions: How robust is the protocol, not just to the failure of
nodes, but also to compromised or malfunctioning components,
imperfect or defective implementations, etc?
Does the protocol take into account the realistic conditions of the
current or future Internet (e.g., packet drops and packet corruption,
packet reordering, asymmetric routing, etc.)?
9.1. Discussion: Designing for Robustness
Robustness has long been cited as one of the overriding goals of the
Internet architecture [Clark88]. The robustness issues discussed in
[Clark88] largely referred to the robustness of packet delivery even
in the presence of failed routers; today robustness concerns have
widened to include a goal of robust performance in the presence of a
wide range of failures, buggy code, and malicious actions.
As [ASSW02] argues, protocols need to be designed somewhat
defensively, to maximize robustness against inconsistencies and
errors. [ASSW02] discusses several approaches for increasing
robustness in protocols, such as verifying information whenever
possible; designing interfaces that are conceptually simple and
therefore less conducive to error; protecting resources against
attack or overuse; and identifying and exposing errors so that they
can be repaired.
Techniques for verifying information range from verifying that
acknowledgements in TCP acknowledge data that was actually sent, to
providing mechanisms for routers to verify information in routing
messages.
Techniques for protecting resources against attack range from
preventing "SYN flood" attacks by designing protocols that don't
allocate resources for a single SYN packet, to partitioning resources
(e.g., preventing one flow or aggregate from using all of the link
bandwidth).
9.2. Case Study: Explicit Congestion Notification (ECN)
The Internet is based on end-to-end congestion control, and
historically the Internet has used packet drops as the only method
for routers to indicate congestion to the end nodes. ECN [RFC3168]
is a recent addition to the IP architecture to allow routers to set a
bit in the IP packet header to inform end-nodes of congestion,
instead of dropping the packet.
Floyd Informational [Page 10]
RFC 3426 Architectural and Policy Considerations November 2002
The first, Experimental specification of ECN [RFC3168] contained an
extensive discussion of the dangers of a rogue or broken router
"erasing" information from the ECN field in the IP header, thus
preventing indications of congestion from reaching the end-nodes. To
add robustness, the standards-track specification [RFC3168] specified
an additional codepoint in the IP header's ECN field, to use for an
ECN "nonce". The development of the ECN nonce was motivated by
earlier research on specific robustness issues in TCP [SCWA99]. RFC
3168 explains that the addition of the codepoint "is motivated
primarily by the desire to allow mechanisms for the data sender to
verify that network elements are not erasing the CE codepoint, and
that data receivers are properly reporting to the sender the receipt
of packets with the CE codepoint set, as required by the transport
protocol." Supporting mechanisms for the ECN nonce are needed in the
transport protocol to ensure robustness of delivery of the ECN-based
congestion indication.
In contrast, a more difficult and less clear-cut robustness issue for
ECN concerns the differential treatment of packets in the network by
middleboxes, based on the TCP header's ECN flags in a TCP SYN packet
[RFC3360]. The issue concerns "ECN-setup" SYN packets, that is, SYN
packets with ECN flags set in the TCP header to negotiate the use of
ECN between the two TCP end-hosts. There exist firewalls in the
network that drop "ECN-setup" SYN packets, others that send TCP Reset
messages, and yet others that zero ECN flags in TCP headers. None of
this was anticipated by the designers of ECN, and RFC 3168 added
optional mechanisms to permit the robust operation of TCP in the
presence of firewalls that drop "ECN-setup" SYN packets. However,
ECN is still not robust to all possible scenarios of middleboxes
zeroing ECN flags in the TCP header. Up until now, transport
protocols have been standardized independently from the mechanisms
used by middleboxes to control the use of these protocols, and it is
still not clear what degree of robustness is required from transport
protocols in the presence of the unauthorized modification of
transport headers in the network. These and similar issues are
discussed in more detail in [RFC3360].
10. Avoiding Tragedy of the Commons
Question: Is performance still robust if everyone is using the new
protocol? Are there other potential impacts of widespread deployment
that need to be considered?
10.1. Case Study: End-to-end Congestion Control
[RFC2914] discusses the potential for congestion collapse if flows
are not using end-to-end congestion control in a time of high
congestion. For example, if a new transport protocol was proposed
Floyd Informational [Page 11]
RFC 3426 Architectural and Policy Considerations November 2002
that did not use end-to-end congestion control, it might be easy to
show that an individual flow using the new transport protocol would
perform quite well as long as all of the competing flows in the
network were using end-to-end congestion control. To fully evaluate
the new transport protocol, it is necessary to look at performance
when many flows are competing, all using the new transport protocol.
If all of the competing flows were using the more aggressive
transport protocol in a time of high congestion, the result could be
high steady-state packet drop rates and reduced overall throughput,
with busy links carrying packets that will only be dropped
downstream. This can be viewed as a form of tragedy of the commons,
when a practice that is productive if done by only one person (e.g.,
adding a few more sheep to the common grazing pasture) is instead
counter-productive when done by everyone [H68].
11. Balancing Competing Interests
Question: Does the protocol protect the interests of competing
parties (e.g., not only end-users, but also ISPs, router vendors,
software vendors, or other parties)?
11.1. Discussion: balancing competing interests
[CWSB02] discusses the role that competition between competing
interests plays in the evolution of the Internet, and takes the
position that the role of Internet protocols is to design the playing
field in this competition, rather than to pick the outcome. The IETF
has long taken the position that it can only design the protocols,
and that often two competing approaches will be developed, with the
marketplace left to decide between them [A02]. (It has also been
suggested that "the marketplace" left entirely to itself does not
always make the best decisions, and that working to identify and
adopt the technically best solution is sometimes helpful. Thus,
while the role of the marketplace should not be ignored, the
decisions of the marketplace should also not be held as sacred or
infallible.)
An example cited in [CWSB02] of modularization in support of
competing interests is the decision to use codepoints in the IP
header to select QoS, rather than binding QoS to other properties
such as port numbers. This separates the structural and economic
issues related to QoS from technical issues of protocols and port
numbers, and allows space for a wide range of structural and pricing
solutions to emerge.
There have been perceived problems over the years of individuals
adding unnecessary complexity to IETF protocols, increasing the
barrier to other implementations of those protocols. Clearly such
Floyd Informational [Page 12]
RFC 3426 Architectural and Policy Considerations November 2002
action would not be protecting the interests of open competition.
Underspecified protocols can also serve as an unnecessary barrier to
open competition.
12. Designing for Choice vs. Avoiding Unnecessary Complexity:
Is the protocol designed for choice, to allow different players to
express their preferences where appropriate? At the other extreme,
does the protocol provide so many choices that it threatens
interoperability or introduces other significant problems?
12.1. Discussion: the importance of choice
[CWSB02] suggests that "the fundamental design goal of the Internet
is to hook computers together, and since computers are used for
unpredictable and evolving purposes, making sure that the users are
not constrained in what they can do is doing nothing more than
preserving the core design tenet of the Internet. In this context,
user empowerment is a basic building block, and should be embedded
into all mechanism whenever possible."
As an example of choice, "the design of the mail system allows the
user to select his SMTP server and his POP server" [CWSB02]. More
open-ended questions about choice concern the design of mechanisms
that would enable the user to choose the path at the level of
providers, or to allow users to choose third-party intermediaries
such as web caches, or providers for Open Pluggable Edge Services
(OPES). [CWSB02] also notes that the issue of choice itself reflects
competing interests. For example, ISPs would generally like to lock
in customers, while customers would like to preserve their ability to
change among providers.
At the same time, we note that excessive choice can lead to "kitchen
sink" protocols that are inefficient and hard to understand, have too
much negotiation, or have unanticipated interactions between options.
For example, [P99] notes that excessive choice can lead to difficulty
in ensuring interoperability between two independent, conformant
implementations of the protocol.
The dangers of excessive options are also discussed in [MBMWOR02],
which gives guidelines for responding to the "continuous flood" of
suggestions for modifications and extensions to SIP (Session
Initiation Protocol). In particular, the SIP Working Group is
concerned that proposed extensions have general use, and do not
provide efficiency at the expense of simplicity or robustness.
[MBMWOR02] suggests that other highly extensible protocols developed
in the IETF might also benefit from more coordination of extensions.
Floyd Informational [Page 13]
RFC 3426 Architectural and Policy Considerations November 2002
13. Preserving evolvability?
Does the protocol protect the interests of the future, by preserving
the evolvability of the Internet? Does the protocol enable future
developments?
If an old protocol is overloaded with new functionality, or reused
for new purposes, have the possible complexities introduced been
taken into account?
For a protocol that introduces new complexity to the Internet
architecture, does the protocol add robustness and preserve
evolvability? Does it also introduce unwanted new fragilities to the
system?
13.1. Discussion: evolvability
There is an extensive literature and an ongoing discussion about the
evolvability, or lack of evolvability, of the Internet
infrastructure; the web page on "Papers on the Evolvability of the
Internet Infrastructure" has pointers to some of this literature
[Evolvability]. Issues range from the evolvability and overloading
of the DNS; the difficulties of the Internet in evolving to
incorporate multicast, QoS, or IPv6; the difficulties of routing in
meeting the demands of a changing and expanding Internet; and the
role of firewalls and other middleboxes in limiting evolvability.
[CWSB02] suggests that among all of the issues of evolvability,
"keeping the net open and transparent for new applications is the
most important goal." In the beginning, the relative transparency of
the infrastructure was sufficient to ensure evolvability, where a
"transparent" network simply routes packets from one end-node to
another. However, this transparency has become more murky over time,
as cataloged in [RFC3234], which discusses the ways that middleboxes
interact with existing protocols and increase the difficulties in
diagnosing failures. [CWSB02] realistically suggests the following
guideline: "Failures of transparency will occur - design what happens
then," where examples of failures of transparency include firewalls,
application filtering, connection redirection, caches, kludges to
DNS, and the like. Thus, maintaining evolvability also requires
mechanisms for allowing evolution in the face of a lack of
transparency of the infrastructure itself.
One of the ways for maintaining evolvability is for designers of new
mechanisms and protocols to be as clear as possible about the
assumptions that are being made about the rest of the network. New
Floyd Informational [Page 14]
RFC 3426 Architectural and Policy Considerations November 2002
mechanisms that make unwarranted assumptions about the network can
end up placing unreasonable constraints on the future evolution of
the network.
13.2. Discussion: overloading
There has been a strong tendency in the last few years to overload
some designs with new functionality, with resulting operational
complexities. Extensible protocols could be seen as one of the tools
for providing evolvability. However, if protocols and systems are
stretched beyond their reasonable design parameters, then scaling,
reliability, or security issues could be introduced. Examples of
protocols that have been seen as either productively extended, or as
dangerously overloaded, or both, include DNS [K02,RFC3403], MPLS
[A02a], and BGP [B02]. In some cases, overloading or extending a
protocol may reduce total complexity and deployment costs by avoiding
the creation of a new protocol; in other cases a new protocol might
be the simpler solution.
We have a number of reusable technologies, including component
technologies specifically designed for re-use. Examples include
SASL, BEEP and APEX. TCP and UDP can also be viewed as reusable
transport protocols, used by a range of applications. On the other
hand, re-use should not go so far as to turn a protocol into a Trojan
Horse, as has happened with HTTP [RFC3205].
13.3. Discussion: complexity, robustness, and fragility
[WD02] gives a historical account of the evolution of the Internet as
a complex system, with particular attention to the tradeoffs between
complexity, robustness, and fragility. [WD02] describes the
robustness that follows from the simplicity of a connectionless,
layered, datagram infrastructure and a universal logical addressing
scheme, and, as case studies, describes the increasing complexity of
TCP and of BGP. The paper describes a complexity/robustness spiral
of an initially robust design and the appearance of fragilities,
followed by modifications for more robustness that themselves
introduce new fragilities. [WD02] conjectures that "the Internet is
only now beginning to experience an acceleration of this
complexity/robustness spiral and, if left unattended, can be fully
expected to experience arcane, irreconcilable, and far-reaching
robustness problems in the not-too-distant future." Citing [WD02],
[BFM02] focuses on the ways that complexity increases capital and
operational expenditures in carrier IP network, and views complexity
as the primary mechanism that impedes efficient scaling.
Floyd Informational [Page 15]
RFC 3426 Architectural and Policy Considerations November 2002
14. Internationalization
Where protocols require elements in text format, have the possibly
conflicting requirements of global comprehensibility and the ability
to represent local text content been properly weighed against each
other?
14.1. Discussion: internationalization
RFC 1958 [RFC1958] included a simple statement of the need for a
common language:
"Public (i.e. widely visible) names should be in case-independent
ASCII. Specifically, this refers to DNS names, and to protocol
elements that are transmitted in text format."
The IETF has studied character set issues in general [RFC 2130] and
made specific recommendations for the use of a standardized approach
[RFC 2277]. The situation is complicated by the fact that some uses
of text are hidden entirely in protocol elements and need only be
read by machines, while other uses are intended entirely for human
consumption. Many uses lie between these two extremes, which leads
to conflicting implementation requirements.
For the specific case of DNS, the Internationalized Domain Name
working group is considering these issues. As stated in the charter
of that working group, "A fundamental requirement in this work is to
not disturb the current use and operation of the domain name system,
and for the DNS to continue to allow any system anywhere to resolve
any domain name." This leads to some very strong requirements for
backwards compatibility with the existing ASCII-only DNS. Yet since
the DNS has come to be used as if it was a directory service, domain
names are also expected to be presented to users in local character
sets.
This document does not attempt to resolve these complex and difficult
issues, but simply states this as an issue to be addressed in our
work. The requirement that names encoded in a text format within
protocol elements be universally decodable (i.e. encoded in a
globally standard format with no intrinsic ambiguity) does not seem
likely to change. However, at some point, it is possible that this
format will no longer be case-independent ASCII.
Floyd Informational [Page 16]
RFC 3426 Architectural and Policy Considerations November 2002
15. Deployability
Question: Is the protocol deployable?
15.1. Discussion: deployability
It has been suggested that the failure to understand deployability
considerations in the current environment is one of the major
weakness of the IETF. As examples of deployment difficulties, RFC
2990 [RFC2990] discusses deployment difficulties with Quality of
Service (QoS) architectures, various documents of the MBONE
Deployment Working Group address deployment problems with IP
Multicast, and the IPv6 Working Group discusses a wealth of issues
related to the deployment of IPv6. [CN02] discusses how the
deployment of Integrated Services has been limited by factors such as
the failure to take into account administrative boundaries.
Additional papers on difficulties in the evolution of the Internet
architecture are available from [Evolvability].
Issues that can complicate deployment include whether the protocol is
compatible with pre-existing standards, and whether the protocol is
compatible with the installed base. For example, a transport
protocol is more likely to be deployable if it performs correctly and
reasonably robustly in the presence of dropped, reordered,
duplicated, delayed, and rerouted packets, as all of this can occur
in the current Internet.
16. Conclusions
This document suggests general architectural and policy questions to
be addressed when working on new protocols and standards in the IETF.
The case studies in this document have generally been short
illustrations of how the questions proposed in the document have been
addressed in working groups in the past. However, we have generally
steered away from case studies of more controversial issues, where
there is not yet a consensus in the IETF community. Thus, we
side-stepped suggestions for adding a case study for IKE (Internet
Key Exchange) as an possible example of a protocol with too much
negotiation, or of adding a case study of network management
protocols as illustrating the possible costs of leaving things to the
marketplace to decide. We would encourage others to contribute case
studies of these or any other issues that may shed light on some of
the questions in this document; any such case studies could include
a careful presentation of the architectural reasoning on both sides.
Floyd Informational [Page 17]
RFC 3426 Architectural and Policy Considerations November 2002
we would conjecture that there is a lot to be learned, in terms of
the range of answers to the questions posed in this document, by
studying the successes, failures, and other struggles of the past.
We would welcome feedback on this document for future revisions.
Feedback could be send to the editor, Sally Floyd, at floyd@icir.org.
17. Acknowledgements
This document has borrowed text freely from other IETF RFCs, and has
drawn on ideas from [ASSW02], [CWSB02], [M01] and elsewhere. This
document has developed from discussions in the IAB, and has drawn
from suggestions made at IAB Plenary sessions and on the ietf general
discussion mailing list. The case study on SIP was contributed by
James Kempf, an early case study on Stresses on DNS was contributed
by Karen Sollins, and Keith Moore contributed suggestions that were
incorporated in a number of places in the document. The discussions
on Internationalization and on Overloading were based on an earlier
document by Brian Carpenter and Rob Austein. We have also benefited
from discussions with Noel Chiappa, Karen Sollins, John Wroclawski,
and others, and from helpful feedback from members of the IESG. We
specifically thank Senthilkumar Ayyasamy, John Loughney, Keith Moore,
Eric Rosen, and Dean Willis and others for feedback on various stages
of this document.
18. Normative References
19. Informative References
[A02] Harald Alvestrand, "Re: How many standards or
protocols...", email to the ietf discussion mailing
list, Message-id: <598204031.1018942481@localhost>,
April 16, 2002.
[A02a] Loa Andersson, "The Role of MPLS in Current IP
Network", MPLS World News, September 16, 2002. URL
"http://www.mplsworld.com/archi_drafts/focus/analy-
andersson.htm".
[ASSW02] T. Anderson, S. Shenker, I. Stoica, and D. Wetherall,
"Design Guidelines for Robust Internet Protocols",
HotNets-I, October 2002.
[BFM02] Randy Bush, Tim Griffin, and David Meyer, "Some
Internet Architectural Guidelines and Philosophy",
Work in Progress.
Floyd Informational [Page 18]
RFC 3426 Architectural and Policy Considerations November 2002
[B02] Hamid Ould-Brahim, Bryan Gleeson, Peter Ashwood-Smith,
Eric C. Rosen, Yakov Rekhter, Luyuan Fang, Jeremy De
Clercq, Riad Hartani, and Tissa Senevirathne, "Using
BGP as an Auto-Discovery Mechanism for Network-based
VPNs", Work in Progress.
[CDT01] Policy Concerns Raised by Proposed OPES Working Group
Efforts, email to the IESG, from the Center for
Democracy & Technology, August 3, 2001. URL
"http://www.imc.org/ietf-openproxy/mail-
archive/msg00828.html".
[Clark88] David D. Clark, The Design Philosophy of the DARPA
Internet Protocols, SIGCOMM 1988.
[CN02] Brian Carpenter, Kathleen Nichols, "Differentiated
Services in the Internet", Technical Report, February
2002, URL "http://www.research.ibm.com/resources/
paper_search.shtml".
[CWSB02] Clark, D., Wroslawski, J., Sollins, K., and Braden,
R., "Tussle in Cyberspace: Defining Tomorrow's
Internet", SIGCOMM 2002. URL
"http://www.acm.org/sigcomm/sigcomm2002/papers/
tussle.html".
[Evolvability] Floyd, S., "Papers on the Evolvability of the Internet
Infrastructure". Web Page, URL
"http://www.icir.org/floyd/evolution.html".
[H68] Garrett Hardin, "The Tragedy of the Commons", Science,
V. 162, 1968, pp. 1243-1248. URL
"http://dieoff.org/page95.htm".
[K02] John C. Klensin, "Role of the Domain Name System",
Work in Progress.
[Layering] Floyd, S., "References on Layering and the Internet
Architecture", Web Page, URL
"http://www.icir.org/floyd/layers.html".
[Multiplexing] S. Floyd, "Multiplexing, TCP, and UDP: Pointers to the
Discussion", Web Page, URL
"http://www.icir.org/floyd/tcp_mux.html".
Floyd Informational [Page 19]
RFC 3426 Architectural and Policy Considerations November 2002
[MBMWOR02] Mankin, A., Bradner, S., Mahy, R., Willis, D., Ott, J.
and B. Rosen, "Change Process for the Session
Initiation Protocol (SIP)", BCP 67, RFC 3427, November
2002.
[M01] Tim Moors, A Critical Review of End-to-end Arguments
in System Design, 2001. URL
"http://uluru.poly.edu/~tmoors/".
[P99] Radia Perlman, "Protocol Design Folklore", in
Interconnections Second Edition: Bridges, Routers,
Switches, and Internetworking Protocols, Addison-
Wesley, 1999.
[RFC1958] Carpenter, B., "Architectural Principles of the
Internet", RFC 1958, June 1996.
[RFC2211] Wroclawski, J., "Specification of the Controlled Load
Quality of Service", RFC 2211, September 1997.
[RFC2212] Shenker, S., Partridge, C., and R. Guerin,
"Specification of Guaranteed Quality of Service", RFC
2212, September 1997.
[RFC2316] Bellovin, S., "Report of the IAB Security Architecture
Workshop", RFC 2316, April 1998.
[RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang,
Z. and W. Weiss, "An Architecture for Differentiated
Services", RFC 2475, December 1998.
[RFC2543] Handley, M., Schulzrinne, H., Schooler, B. and J.
Rosenberg, "SIP: Session Initiation Protocol", RFC
25434, March 1999.
[RFC2597] Heinanen, J., Baker, F., Weiss, W. and J. Wroclawski,
"Assured Forwarding PHB Group", RFC 2597, June 1999.
[RFC2990] Huston, G., "Next Steps for the IP QoS Architecture",
RFC 2990, November 2000.
[RFC3124] Balakrishnan, H. and S. Seshan, "The Congestion
Manager", RFC 3124, June 2001.
[RFC3135] Border, J., Kojo, M., Griner, J., Montenegro, G. and
Z. Shelby, "Performance Enhancing Proxies Intended to
Mitigate Link-Related Degradations", RFC 3135, June
2001.
Floyd Informational [Page 20]
RFC 3426 Architectural and Policy Considerations November 2002
[RFC3168] Ramakrishnan, K. K., Floyd, S. and D. Black, "The
Addition of Explicit Congestion Notification (ECN) to
IP", RFC 3168, September 2001.
[RFC3205] Moore, K., "On the use of HTTP as a Substrate", BCP
56, RFC 3205, February 2002.
[RFC3221] Huston, G., "Commentary on Inter-Domain Routing in the
Internet", RFC 3221, December 2001.
[RFC3234] Carpenter, B. and S. Brim, "Middleboxes: Taxonomy and
Issues", RFC 3234, February 2002.
[RFC3238] Floyd, S. and L. Daigle, "IAB Architectural and Policy
Considerations for Open Pluggable Edge Services", RFC
3238, January 2002.
[RFC3246] Davie, B., Charny, A., Bennet, J. C. R., Benson, K.,
Le Boudec, J. Y., Courtney, W., Davari, S., Firoiu, V.
and D. Stiliadis, "An Expedited Forwarding PHB (Per-
Hop Behavior)", RFC 3246, March 2002.
[RFC3360] Floyd, S., "Inappropriate TCP Resets Considered
Harmful", BCP 60, RFC 3360, August 2002.
[RFC3403] Mealling, M., "Dynamic Delegation Discovery System
(DDDS) Part Three: The Domain Name System (DNS)
Database", RFC 3403, October 2002.
[RFC3424] Daigle, L., "IAB Considerations for UNilateral Self-
Address Fixing (UNSAF)", RFC 3424, November 2002.
[SCWA99] Stefan Savage, Neal Cardwell, David Wetherall, Tom
Anderson, "TCP Congestion Control with a Misbehaving
Receiver", ACM Computer Communications Review, October
1999.
[SRC84] J. Saltzer, D. Reed, and D. D. Clark, "End-To-End
Arguments In System Design", ACM Transactions on
Computer Systems, V.2, N.4, p. 277-88. 1984.
[T89] D. Tennenhouse, "Layered Multiplexing Considered
Harmful", Protocols for High-Speed Networks, 1989.
[WD02] Walter Willinger and John Doyle, "Robustness and the
Internet: Design and Evolution", draft, March 2002,
URL "http://netlab.caltech.edu/internet/".
Floyd Informational [Page 21]
RFC 3426 Architectural and Policy Considerations November 2002
20. Security Considerations
This document does not propose any new protocols, and therefore does
not involve any security considerations in that sense. However,
throughout this document there are discussions of the privacy and
integrity issues and the architectural requirements created by those
issues.
21. IANA Considerations
There are no IANA considerations regarding this document.
Authors' Addresses
Internet Architecture Board
EMail: iab@iab.org
IAB Membership at time this document was completed:
Harald Alvestrand
Ran Atkinson
Rob Austein
Fred Baker
Leslie Daigle
Steve Deering
Sally Floyd
Ted Hardie
Geoff Huston
Charlie Kaufman
James Kempf
Eric Rescorla
Mike St. Johns
Floyd Informational [Page 22]
RFC 3426 Architectural and Policy Considerations November 2002
Full Copyright Statement
Copyright (C) The Internet Society (2002). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.
Floyd Informational [Page 23]