RFC 5772
A Set of Possible Requirements for a Future Routing Architecture
Pages: 68
Historic
Historic
Part 1 of 3 – Pages 1 to 24
Internet Research Task Force (IRTF) A. Doria Request for Comments: 5772 LTU Category: Historic E. Davies ISSN: 2070-1721 Folly Consulting F. Kastenholz BBN Technologies February 2010 A Set of Possible Requirements for a Future Routing ArchitectureAbstract
The requirements for routing architectures described in this document were produced by two sub-groups under the IRTF Routing Research Group (RRG) in 2001, with some editorial updates up to 2006. The two sub- groups worked independently, and the resulting requirements represent two separate views of the problem and of what is required to fix the problem. This document may usefully serve as part of the recommended reading for anyone who works on routing architecture designs for the Internet in the future. The document is published with the support of the IRTF RRG as a record of the work completed at that time, but with the understanding that it does not necessarily represent either the latest technical understanding or the technical consensus of the research group at the date of publication. Status of This Memo This document is not an Internet Standards Track specification; it is published for the historical record. This document defines a Historic Document for the Internet community. This document is a product of the Internet Research Task Force (IRTF). The IRTF publishes the results of Internet-related research and development activities. These results might not be suitable for deployment. This RFC represents the individual opinion(s) of one or more members of the Routing Research Group of the Internet Research Task Force (IRTF). Documents approved for publication by the IRSG are not a candidate for any level of Internet Standard; see Section 2 of RFC 5741. Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at http://www.rfc-editor.org/info/rfc5772.
Copyright Notice Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document.Table of Contents
1. Background ......................................................4 2. Results from Group A ............................................5 2.1. Group A - Requirements for a Next Generation Routing and Addressing Architecture ....................................5 2.1.1. Architecture ........................................6 2.1.2. Separable Components ................................6 2.1.3. Scalable ............................................7 2.1.4. Lots of Interconnectivity ..........................10 2.1.5. Random Structure ...................................10 2.1.6. Multi-Homing .......................................11 2.1.7. Multi-Path .........................................11 2.1.8. Convergence ........................................12 2.1.9. Routing System Security ............................14 2.1.10. End Host Security .................................16 2.1.11. Rich Policy .......................................16 2.1.12. Incremental Deployment ............................19 2.1.13. Mobility ..........................................19 2.1.14. Address Portability ...............................20 2.1.15. Multi-Protocol ....................................20 2.1.16. Abstraction .......................................20 2.1.17. Simplicity ........................................21 2.1.18. Robustness ........................................21 2.1.19. Media Independence ................................22 2.1.20. Stand-Alone .......................................22 2.1.21. Safety of Configuration ...........................23 2.1.22. Renumbering .......................................23 2.1.23. Multi-Prefix ......................................23 2.1.24. Cooperative Anarchy ...............................23 2.1.25. Network-Layer Protocols and Forwarding Model ......23 2.1.26. Routing Algorithm .................................23 2.1.27. Positive Benefit ..................................24 2.1.28. Administrative Entities and the IGP/EGP Split .....24 2.2. Non-Requirements ..........................................25 2.2.1. Forwarding Table Optimization ......................25
2.2.2. Traffic Engineering ................................25
2.2.3. Multicast ..........................................25
2.2.4. Quality of Service (QoS) ...........................26
2.2.5. IP Prefix Aggregation ..............................26
2.2.6. Perfect Safety .....................................26
2.2.7. Dynamic Load Balancing .............................27
2.2.8. Renumbering of Hosts and Routers ...................27
2.2.9. Host Mobility ......................................27
2.2.10. Backward Compatibility ............................27
3. Requirements from Group B ......................................27
3.1. Group B - Future Domain Routing Requirements ..............28
3.2. Underlying Principles .....................................28
3.2.1. Inter-Domain and Intra-Domain ......................29
3.2.2. Influences on a Changing Network ...................29
3.2.3. High-Level Goals ...................................31
3.3. High-Level User Requirements ..............................35
3.3.1. Organizational Users ...............................35
3.3.2. Individual Users ...................................37
3.4. Mandated Constraints ......................................38
3.4.1. The Federated Environment ..........................39
3.4.2. Working with Different Sorts of Networks ...........39
3.4.3. Delivering Resilient Service .......................39
3.4.4. When Will the New Solution Be Required? ............40
3.5. Assumptions ...............................................40
3.6. Functional Requirements ...................................42
3.6.1. Topology ...........................................43
3.6.2. Distribution .......................................44
3.6.3. Addressing .........................................48
3.6.4. Statistics Support .................................50
3.6.5. Management Requirements ............................50
3.6.6. Provability ........................................51
3.6.7. Traffic Engineering ................................52
3.6.8. Support for Middleboxes ............................54
3.7. Performance Requirements ..................................54
3.8. Backward Compatibility (Cutover) and Maintainability ......55
3.9. Security Requirements .....................................56
3.10. Debatable Issues .........................................57
3.10.1. Network Modeling ..................................58
3.10.2. System Modeling ...................................58
3.10.3. One, Two, or Many Protocols .......................59
3.10.4. Class of Protocol .................................59
3.10.5. Map Abstraction ...................................59
3.10.6. Clear Identification for All Entities .............60
3.10.7. Robustness and Redundancy .........................60
3.10.8. Hierarchy .........................................60
3.10.9. Control Theory ....................................61
3.10.10. Byzantium ........................................61
3.10.11. VPN Support ......................................61
3.10.12. End-to-End Reliability ...........................62
3.10.13. End-to-End Transparency ..........................62
4. Security Considerations ........................................62
5. IANA Considerations ............................................63
6. Acknowledgments ................................................63
7. Informative References .........................................65
1. Background
In 2001, the IRTF Routing Research Group (IRTF RRG) chairs, Abha
Ahuja and Sean Doran, decided to establish a sub-group to look at
requirements for inter-domain routing (IDR). A group of well-known
routing experts was assembled to develop requirements for a new
routing architecture. Their mandate was to approach the problem
starting from a blank slate. This group was free to take any
approach, including a revolutionary approach, in developing
requirements for solving the problems they saw in inter-domain
routing.
Simultaneously, an independent effort was started in Sweden with a
similar goal. A team, calling itself Babylon, with participation
from vendors, service providers, and academia assembled to understand
the history of inter-domain routing, to research the problems seen by
the service providers, and to develop a proposal of requirements for
a follow-on to the current routing architecture. This group's remit
required an evolutionary approach starting from current routing
architecture and practice. In other words, the group limited itself
to developing an evolutionary strategy. The Babylon group was later
folded into the IRTF RRG as Sub-Group B to distinguish it from the
original RRG Sub-Group A.
One of the questions that arose while the groups were working in
isolation was whether there would be many similarities between their
sets of requirements. That is, would the requirements that grew from
a blank sheet of paper resemble those that started with the
evolutionary approach? As can be seen from reading the two sets of
requirements, there were many areas of fundamental agreement but some
areas of disagreement.
There were suggestions within the RRG that the two teams should work
together to create a single set of requirements. Since these
requirements are only guidelines to future work, however, some felt
that doing so would risk losing content without gaining any
particular advantage. It is not as if any group, for example, the
IRTF RRG or the IETF Routing Area, was expected to use these
requirements as written and to create an architecture that met these
requirements. Rather, the requirements were in practice strong
recommendations for a way to proceed in creating a new routing architecture. In the end, the decision was made to include the results of both efforts, side by side, in one document. This document contains the two requirement sets produced by the teams. The text has received only editorial modifications; the requirements themselves have been left unaltered. Whenever the editors felt that conditions had changed in the few years since the text was written, an editors' note has been added to the text. In reading this document, it is important to keep in mind that all of these requirements are suggestions, which are laid out to assist those interested in developing new routing architectures. It is also important to remember that, while the people working on these suggestions have done their best to make intelligent suggestions, there are no guarantees. So a reader of this document should not treat what it says as absolute, nor treat every suggestion as necessary. No architecture is expected to fulfill every "requirement". Hopefully, though, future architectures will consider what is offered in this document. The IRTF RRG supported publication of this document as a historical record of the work completed on the understanding that it does not necessarily represent either the latest technical understanding or the technical consensus of the research group at the time of publication. The document has had substantial review by members of the two teams, other members of the IRTF RRG, and additional experts over the years. Finally, this document does not make any claims that it is possible to have a practical solution that meets all the listed requirements.2. Results from Group A
This section presents the results of the work done by Sub-Group A of the IRTF RRG during 2001-2002. The work originally appeared under the title: "Requirements For a Next Generation Routing and Addressing Architecture" and was edited by Frank Kastenholz.2.1. Group A - Requirements for a Next Generation Routing and Addressing Architecture
The requirements presented in this section are not presented in any particular order.
2.1.1. Architecture
The new routing and addressing protocols, data structures, and algorithms need to be developed from a clear, well thought-out, and documented architecture. The new routing and addressing system must have an architectural specification that describes all of the routing and addressing elements, their interactions, what functions the system performs, and how it goes about performing them. The architectural specification does not go into issues such as protocol and data structure design. The architecture should be agnostic with regard to specific algorithms and protocols. Doing architecture before doing detailed protocol design is good engineering practice. This allows the architecture to be reviewed and commented upon, with changes made as necessary, when it is still easy to do so. Also, by producing an architecture, the eventual users of the protocols (the operations community) will have a better understanding of how the designers of the protocols meant them to be used.2.1.2. Separable Components
The architecture must place different functions into separate components. Separating functions, capabilities, and so forth into individual components and making each component "stand alone" is generally considered by system architects to be "A Good Thing". It allows individual elements of the system to be designed and tuned to do their jobs "very well". It also allows for piecemeal replacement and upgrading of elements as new technologies and algorithms become available. The architecture must have the ability to replace or upgrade existing components and to add new ones, without disrupting the remaining parts of the system. Operators must be able to roll out these changes and additions incrementally (i.e., no "flag days"). These abilities are needed to allow the architecture to evolve as the Internet changes. The architecture specification shall define each of these components, their jobs, and their interactions.
Some thoughts to consider along these lines are:
o Making topology and addressing separate subsystems. This may
allow highly optimized topology management and discovery without
constraining the addressing structure or physical topology in
unacceptable ways.
o Separate "fault detection and healing" from basic topology. From
Mike O'Dell:
Historically the same machinery is used for both. While
attractive for many reasons, the availability of exogenous
topology information (i.e., the intended topology) should, it
seems, make some tasks easier than the general case of starting
with zero knowledge. It certainly helps with recovery in the
case of constraint satisfaction. In fact, the intended
topology is a powerful way to state certain kinds of policy.
[ODell01]
o Making policy definition and application a separate subsystem,
layered over the others.
The architecture should also separate topology, routing, and
addressing from the application that uses those components. This
implies that applications such as policy definition, forwarding, and
circuit and tunnel management are separate subsystems layered on top
of the basic topology, routing, and addressing systems.
2.1.3. Scalable
Scaling is the primary problem facing the routing and addressing
architecture today. This problem must be solved and it must be
solved for the long term.
The architecture must support a large and complex network. Ideally,
it will serve our needs for the next 20 years. Unfortunately:
1. we do not know how big the Internet will grow over that time, and
2. the architecture developed from these requirements may change the
fundamental structure of the Internet and therefore its growth
patterns. This change makes it difficult to predict future
growth patterns of the Internet.
As a result, we can't quantify the requirement in any meaningful way.
Using today's architectural elements as a mechanism for describing
things, we believe that the network could grow to:
1. tens of thousands of ASs
Editors' Note: As of 2005, this level had already been
reached.
2. tens to hundreds of millions of prefixes, during the lifetime of
this architecture.
These sizes are given as a "flavor" for how we expect the Internet to
grow. We fully believe that any new architecture may eliminate some
current architectural elements and introduce new ones.
A new routing and addressing architecture designed for a specific
network size would be inappropriate. First, the cost of routing
calculations is based only in part on the number of ASs or prefixes
in the network. The number and locations of the links in the network
are also significant factors. Second, past predictions of Internet
growth and topology patterns have proven to be wildly inaccurate, so
developing an architecture to a specific size goal would at best be
shortsighted.
Editors' Note: At the time of these meetings, the BGP statistics
kept at sites such as www.routeviews.org either did not exist or
had been running for only a few months. After 5 years of
recording public Internet data trends in AS growth, routing table
growth can be observed (past) with some short-term prediction. As
each year of data collection continues, the ability to observe and
predict trends improves. This architecture work pointed out the
need for such statistics to improve future routing designs.
Therefore, we will not make the scaling requirement based on a
specific network size. Instead, the new routing and addressing
architecture should have the ability to constrain the increase in
load (CPU, memory space and bandwidth, and network bandwidth) on ANY
SINGLE ROUTER to be less than these specific functions:
1. The computational power and memory sizes required to execute the
routing protocol software and to contain the tables must grow
more slowly than hardware capabilities described by Moore's Law,
doubling every 18 months. Other observations indicate that
memory sizes double every 2 years or so.
2. Network bandwidth and latency are some key constraints on how
fast routing protocol updates can be disseminated (and therefore
how fast the routing system can adapt to changes). Raw network
bandwidth seems to quadruple every 3 years or so. However, it
seems that there are some serious physics problems in going
faster than 40 Gbit/s (OC768); we should not expect raw network
link speed to grow much beyond OC768. On the other hand, for
economic reasons, large swathes of the core of the Internet will
still operate at lower speeds, possibly as slow as DS3.
Editors' Note: Technology is running ahead of imagination and
higher speeds are already common.
Furthermore, in some sections of the Internet, even lower speed
links are found. Corporate access links are often T1, or slower.
Low-speed radio links exist. Intra-domain links may be T1 or
fractional-T1 (or slower).
Therefore, the architecture must not make assumptions about the
bandwidth available.
3. The speeds of high-speed RAMs (Static RAMs (SRAMs), used for
caches and the like) are growing, though slowly. Because of
their use in caches and other very specific applications, these
RAMs tend to be small, a few megabits, and the size of these RAMs
is not increasing very rapidly.
On the other hand, the speed of "large" memories (Dynamic RAMs
(DRAMs)) is increasing even slower than that for the high-speed
RAMs. This is because the development of these RAMs is driven by
the PC market, where size is very important, and low speed can be
made up for by better caches.
Memory access rates should not be expected to increase
significantly.
Editors' Note: Various techniques have significantly increased
memory bandwidth. 800 MHz is now possible, compared with less
than 100 MHz in the year 2000. This does not, however,
contradict the next paragraph, but rather just extends the
timescales somewhat.
The growth in resources available to any one router will eventually
slow down. It may even stop. Even so, the network will continue to
grow. The routing and addressing architecture must continue to scale
in even this extreme condition. We cannot continue to add more
computing power to routers forever. Other strategies must be
available. Some possible strategies are hierarchy, abstraction, and
aggregation of topology information.
2.1.4. Lots of Interconnectivity
The new routing and addressing architecture must be able to cope with a high degree of interconnectivity in the Internet. That is, there are large numbers of alternate paths and routes among the various elements. Mechanisms are required to prevent this interconnectivity (and continued growth in interconnectivity) from causing tables, compute time, and routing protocol traffic to grow without bound. The "cost" to the routing system of an increase in complexity must be limited in scope; sections of the network that do not see, or do not care about, the complexity ought not pay the cost of that complexity. Over the past several years, the Internet has seen an increase in interconnectivity. Individual end sites (companies, customers, etc.), ISPs, exchange points, and so on, all are connecting to more "other things". Companies multi-home to multiple ISPs, ISPs peer with more ISPs, and so on. These connections are made for many reasons, such as getting more bandwidth, increased reliability and availability, policy, and so on. However, this increased interconnectivity has a price. It leads to more scaling problems as it increases the number of AS paths in the networks. Any new architecture must assume that the Internet will become a denser mesh. It must not assume, nor can it dictate, certain patterns or limits on how various elements of the network interconnect. Another facet of this requirement is that there may be multiple valid, loop-free paths available to a destination. See Section 2.1.7 for a further discussion. We wryly note that one of the original design goals of IP was to support a large, heavily interconnected network, which would be highly survivable (such as in the face of a nuclear war).2.1.5. Random Structure
The routing and addressing architecture must not place any constraints on or make assumptions about the topology or connectedness of the elements comprising the Internet. The routing and addressing architecture must not presume any particular network structure. The network does not have a "nice" structure. In the past, we used to believe that there was this nice "backbone/tier-1/ tier-2/end-site" sort of hierarchy. This is not so. Therefore, any new architecture must not presume any such structure.
Some have proposed that a geographic addressing scheme be used, requiring exchange points to be situated within each geographic "region". There are many reasons why we believe this to be a bad approach, but those arguments are irrelevant. The main issue is that the routing architecture should not presume a specific network structure.2.1.6. Multi-Homing
The architecture must provide multi-homing for all elements of the Internet. That is, multi-homing of hosts, subnetworks, end-sites, "low-level" ISPs, and backbones (i.e., lots of redundant interconnections) must be supported. Among the reasons to multi-home are reliability, load sharing, and performance tuning. The term "multi-homing" may be interpreted in its broadest sense -- one "place" has multiple connections or links to another "place". The architecture must not limit the number of alternate paths to a multi-homed site. When multi-homing is used, it must be possible to use one, some (more than one but less than all), or all of the available paths to the multi-homed site. The multi-homed site must have the ability to declare which path(s) are used and under what conditions (for example, one path may be declared "primary" and the other "backup" and to be used only when the primary fails). A current problem in the Internet is that multi-homing leads to undue increases in the size of the BGP routing tables. The new architecture must support multi-homing without undue routing table growth.2.1.7. Multi-Path
As a corollary to multi-homing, the architecture must allow for multiple paths from a source to a destination to be active at the same time. These paths need not have the same attributes. Policies are to be used to disseminate the attributes and to classify traffic for the different paths. There must be a rich "language" for specifying the rules for classifying the traffic and assigning classes of traffic to different paths (or prohibiting it from certain paths). The rules should allow traffic to be classified based upon, at least, the following:
o IPv6 FlowIDs, o Diffserv Code Point (DSCP) values, o source and/or destination prefixes, or o random selections at some probability. A mechanism is needed that allows operators to plan and manage the traffic load on the various paths. To start, this mechanism can be semi-automatic or even manual. Eventually, it ought to become fully automatic. When multi-path forwarding is used, options must be available to preserve packet ordering where appropriate (such as for individual TCP connections). Please refer to Section 2.2.7 for a discussion of dynamic load- balancing and management over multiple paths.2.1.8. Convergence
The speed of convergence (also called the "stabilization time") is the time it takes for a router's routing processes to reach a new, stable, "solution" (i.e., forwarding information base) after a change someplace in the network. In effect, what happens is that the output of the routing calculations stabilizes -- the Nth iteration of the software produces the same results as the N-1th iteration. The speed of convergence is generally considered to be a function of the number of subnetworks in the network and the amount of connections between those networks. As either number grows, the time it takes to converge increases. In addition, a change can "ripple" back and forth through the system. One change can go through the system, causing some other router to change its advertised connectivity, causing a new change to ripple through. These oscillations can take a while to work their way out of the network. It is also possible that these ripples never die out. In this situation, the routing and addressing system is unstable; it never converges. Finally, it is more than likely that the routers comprising the Internet never converge simply because the Internet is so large and complex. Assume it takes S seconds for the routers to stabilize on a solution for any one change to the network. Also, assume that changes occur, on average, every C seconds. Because of the size and complexity of the Internet, C is now less than S. Therefore, if a
change, C1, occurs at time T, the routing system would stabilize at
time T+S, but a new change, C2, will occur at time T+C, which is
before T+S. The system will start processing the new change before
it's done with the old.
This is not to say that all routers are constantly processing
changes. The effects of changes are like ripples in a pond. They
spread outward from where they occur. Some routers will be
processing just C1, others C2, others both C1 and C2, and others
neither.
We have two separate scopes over which we can set requirements with
respect to convergence:
1. Single Change: In this requirement, a single change of any type
(link addition or deletion, router failure or restart, etc.) is
introduced into a stabilized system. No additional changes are
introduced. The system must re-stabilize within some measure of
bounded time. This requirement is a fairly abstract one as it
would be impossible to test in a real network. Definition of the
time constraints remains an open research issue.
2. System-Wide: Defining a single target for maximum convergence
time for the real Internet is absurd. As we mentioned earlier,
the Internet is large enough and diverse enough so that it is
quite likely that new changes are introduced somewhere before the
system fully digests old ones.
So, the first requirement here is that there must be mechanisms to
limit the scope of any one change's visibility and effects. The
number of routers that have to perform calculations in response to a
change is kept small, as is the settling time.
The second requirement is based on the following assumptions:
- the scope of a change's visibility and impact can be limited.
That is, routers within that scope know of the change and
recalculate their tables based on the change. Routers outside of
the scope don't see it at all.
- Within any scope, S, network changes are constantly occurring and
the average inter-change interval is Tc seconds.
- There are Rs routers within scope S.
- A subset of the destinations known to the routers in S, Ds, are
impacted by a given change.
- We can state that for Z% of the changes, within Y% of Tc seconds
after a change, C, X% of the Rs routers have their routes to Ds
settled to a useful answer (useful meaning that packets can get to
Ds, though perhaps not by the optimal path -- this allows some
"hunting" for the optimal solution).
X, Y, and Z are yet to be defined. Their definition remains a
research issue.
This requirement implies that the scopes can be kept relatively small
in order to minimize Rs and maximize Tc.
The growth rate of the convergence time must not be related to the
growth rate of the Internet as a whole. This implies that the
convergence time either:
1. not be a function of basic network elements (such as prefixes and
links/paths), and/or
2. that the Internet be continuously divisible into chunks that
limit the scope and effect of a change, thereby limiting the
number of routers, prefixes, links, and so on, involved in the
new calculations.
2.1.9. Routing System Security
The security of the Internet's routing system is paramount. If the
routing system is compromised or attacked, the entire Internet can
fail. This is unacceptable. Any new architecture must be secure.
Architectures by themselves are not secure. It is the implementation
of an architecture, its protocols, algorithms, and data structures
that are secure. These requirements apply primarily to the
implementation. The architecture must provide the elements that the
implementation needs to meet these security requirements. Also, the
architecture must not prevent these security requirements from being
met.
Security means different things to different people. In order for
this requirement to be useful, we must define what we mean by
security. We do this by identifying the attackers and threats we
wish to protect against. They are:
Masquerading
The system, including its protocols, must be secure against
intruders adopting the identity of other known, trusted
elements of the routing system and then using that position of
trust for carrying out other attacks. Protocols must use
cryptographically strong authentication.
Denial-of-Service (DoS) Attacks
The architecture and protocols should be secure against DoS
attacks directed at the routers.
The new architecture and protocols should provide as much
information as it can to allow administrators to track down
sources of DoS and Distributed DOS (DDoS) attacks.
No Bad Data
Any new architecture and protocols must provide protection
against the introduction of bad, erroneous, or misleading data
by attackers. Of particular importance, an attacker must not
be able to redirect traffic flows, with the intent of:
o directing legitimate traffic away from a target, causing a
denial-of-service attack by preventing legitimate data from
reaching its destination,
o directing additional traffic (going to other destinations
that are "innocent bystanders") to a target, causing the
target to be overloaded, or
o directing traffic addressed to the target to a place where
the attacker can copy, snoop, alter, or otherwise affect the
traffic.
Topology Hiding
Any new architecture and protocols must provide mechanisms to
allow network owners to hide the details of their internal
topologies, while maintaining the desired levels of service
connectivity and reachability.
Privacy
By "privacy" we mean privacy of the routing protocol exchanges
between routers.
When the routers are on point-to-point links, with routers at
each end, there may not be any need to encrypt the routing
protocol traffic as the possibility of a third party
intercepting the traffic is limited, though not impossible. We
do believe, however, that it is important to have the ability
to protect routing protocol traffic in two cases:
1. When the routers are on a shared network, it is possible
that there are hosts on the network that have been
compromised. These hosts could surreptitiously monitor the
protocol traffic.
2. When two routers are exchanging information "at a distance"
(over intervening routers and, possibly, across
administrative domain boundaries). In this case, the
security of the intervening routers, links, and so on,
cannot be assured. Thus, the ability to encrypt this
traffic is important.
Therefore, we believe that the option to encrypt routing
protocol traffic is required.
Data Consistency
A router should be able to detect and recover from any data
that is received from other routers that is inconsistent. That
is, it must not be possible for data from multiple routers,
none of which is malicious, to "break" another router.
Where security mechanisms are provided, they must use methods that
are considered to be cryptographically secure (e.g., using
cryptographically strong encryption and signatures -- no cleartext
passwords!).
Use of security features should not be optional (except as required
above). This may be "social engineering" on our part, but we believe
it to be necessary. If a security feature is optional, the
implementation of the feature must default to the "secure" setting.
2.1.10. End Host Security
The architecture must not prevent individual host-to-host
communications sessions from being secured (i.e., it cannot interfere
with things like IPsec).
2.1.11. Rich Policy
Before setting out policy requirements, we need to define the term.
Like "security", "policy" means different things to different people.
For our purposes, "policy" is the set of administrative influences
that alter the path determination and next-hop selection procedures
of the routing software.
The main motivators for influencing path and next-hop selection seem to be transit rules, business decisions, and load management. The new architecture must support rich policy mechanisms. Furthermore, the policy definition and dissemination mechanisms should be separated from the network topology and connectivity dissemination mechanisms. Policy provides input to and controls the generation of the forwarding table and the abstraction, filtering, aggregation, and dissemination of topology information. Note that if the architecture is properly divided into subsystems, then at a later time, new policy subsystems that include new features and capabilities could be developed and installed as needed. We divide the general area of policy into two sub-categories: routing information and traffic control. Routing Information Policies control what routing information is disseminated or accepted, how it is disseminated, and how routers determine paths and next-hops from the received information. Traffic Control Policies determine how traffic is classified and assigned to routes.2.1.11.1. Routing Information Policies
There must be mechanisms to allow network administrators, operators, and designers to control receipt and dissemination of routing information. These controls include, but are not limited to: - Selecting to which other routers routing information will be transmitted. - Specifying the "granularity" and type of transmitted information. The length of IPv4 prefixes is an example of granularity. - Selection and filtering of topology and service information that is transmitted. This gives different "views" of internal structure and topology to different peers. - Selecting the level of security and authenticity for transmitted information. - Being able to cause the level of detail that is visible for some portion of the network to reduce the farther you get from that part of the network. - Selecting from whom routing information will be accepted. This control should be "provisional" in the sense of "accept routes from "foo" only if there are no others available".
- Accepting or rejecting routing information based on the path the
information traveled (using the current system as an example, this
would be filtering routes based on an AS appearing anywhere in the
AS path). This control should be "use only if there are no other
paths available".
- Selecting the desired level of granularity for received routing
information (this would include, but is not limited to, things
similar in nature to the prefix-length filters widely used in the
current routing and addressing system).
- Selecting the level of security and authenticity of received
information in order for that information to be accepted.
- Determining the treatment of received routing information based on
attributes supplied with the information.
- Applying attributes to routing information that is to be
transmitted and then determining treatment of information (e.g.,
sending it "here" but not "there") based on those tags.
- Selection and filtering of topology and service information that
is received.
2.1.11.2. Traffic Control Policies
The architecture should provide mechanisms that allow network
operators to manage and control the flow of traffic. The traffic
controls should include, but are not limited to:
- The ability to detect and eliminate congestion points in the
network (by redirecting traffic around those points).
- The ability to develop multiple paths through the network with
different attributes and then assign traffic to those paths based
on some discriminators within the packets (discriminators include,
but are not limited to, IP addresses or prefixes, IPv6 flow ID,
DSCP values, and MPLS labels).
- The ability to find and use multiple, equivalent paths through the
network (i.e., they would have the "same" attributes) and allocate
traffic across the paths.
- The ability to accept or refuse traffic based on some traffic
classification (providing, in effect, transit policies).
Traffic classification must at least include the source and
destination IP addresses (prefixes) and the DSCP value. Other
fields may be supported, such as:
* Protocol and port-based functions,
* DSCP/QoS (Quality of Service) tuple (such as ports),
* Per-host operations (i.e., /32 s for IPv4 and /128 s for IPv6),
and
* Traffic matrices (e.g., traffic from prefix X and to prefix Y).
2.1.12. Incremental Deployment
The reality of the Internet is that there can be no Internet-wide
cutover from one architecture and protocol to another. This means
that any new architecture and protocol must be incrementally
deployable; ISPs must be able to set up small sections of the new
architecture, check it out, and then slowly grow the sections.
Eventually, these sections will "touch" and "squeeze out" the old
architecture.
The protocols that implement the architecture must be able to
interoperate at "production levels" with currently existing routing
protocols. Furthermore, the protocol specifications must define how
the interoperability is done.
We also believe that sections of the Internet will never convert over
to the new architecture. Thus, it is important that the new
architecture and its protocols be able to interoperate with "old
architecture" regions of the network indefinitely.
The architecture's addressing system must not force existing address
allocations to be redone: no renumbering!
2.1.13. Mobility
There are two kinds of mobility: host mobility and network mobility.
Host mobility is when an individual host moves from where it was to
where it is. Network mobility is when an entire network (or
subnetwork) moves.
The architecture must support network-level mobility. Please refer
to Section 2.2.9 for a discussion of host mobility.
Editors' Note: Since the time of this work, the Network Mobility
(NEMO) extensions to Mobile-IP [RFC3963] to accommodate mobile
networks have been developed.
2.1.14. Address Portability
One of the big "hot items" in the current Internet political climate
is portability of IP addresses (both v4 and v6). The short
explanation is that people do not like to renumber when changing
connection point or provider and do not trust automated renumbering
tools.
The architecture must provide complete address portability.
2.1.15. Multi-Protocol
The Internet is expected to be "multi-protocol" for at least the next
several years. IPv4 and IPv6 will co-exist in many different ways
during a transition period. The architecture must be able to handle
both IPv4 and IPv6 addresses. Furthermore, protocols that supplant
IPv4 and IPv6 may be developed and deployed during the lifetime of
the architecture. The architecture must be flexible and extensible
enough to handle new protocols as they arise.
Furthermore, the architecture must not assume any given relationships
between a topological element's IPv4 address and its IPv6 address.
The architecture must not assume that all topological elements have
IPv4 addresses/prefixes, nor can it assume that they have IPv6
addresses/prefixes.
The architecture should allow different paths to the same destination
to be used for different protocols, even if all paths can carry all
protocols.
In addition to the addressing technology, the architecture need not
be restricted to only packet-based multiplexing/demultiplexing
technology (such as IP); support for other multiplexing/
demultiplexing technologies may be added.
2.1.16. Abstraction
The architecture must provide mechanisms for network designers and
operators to:
o Group elements together for administrative control purposes,
o Hide the internal structure and topology of those groupings for
administrative and security reasons,
o Limit the amount of topology information that is exported from the
groupings in order to control the load placed on external routers,
o Define rules for traffic transiting or terminating in the
grouping.
The architecture must allow the current Autonomous System structure
to be mapped into any new abstraction schemes.
Mapping mechanisms, algorithms, and techniques must be specified.
2.1.17. Simplicity
The architecture must be simple enough so that someone who is
extremely knowledgeable in routing and who is skilled at creating
straightforward and simple explanations can explain all the important
concepts in less than an hour.
This criterion has been chosen since developing an objective measure
of complexity for an architecture can be very difficult and is out of
scope for this document.
The requirement is that the routing architecture be kept as simple as
possible. This requires careful evaluation of possible features and
functions with a merciless weeding out of those that "might be nice"
but are not necessary.
By keeping the architecture simple, the protocols and software used
to implement the architecture are simpler. This simplicity in turn
leads to:
1. Faster implementation of the protocols. If there are fewer bells
and whistles, then there are fewer things that need to be
implemented.
2. More reliable implementations. With fewer components, there is
less code, reducing bug counts, and fewer interactions between
components that could lead to unforeseen and incorrect behavior.
2.1.18. Robustness
The architecture, and the protocols implementing it, should be
robust. Robustness comes in many different flavors. Some
considerations with regard to robustness include (but are not limited
to):
o Continued correct operation in the face of:
* Defective (even malicious) trusted routers.
* Network failures. Whenever possible, valid alternate paths are
to be found and used.
o Failures must be localized. That is, the architecture must limit
the "spread" of any adverse effects of a misconfiguration or
failure. Badness must not spread.
Of course, the general robustness principle of being liberal in
what's accepted and conservative in what's sent must also be applied.
Original Editor's Note: Some of the contributors to this section
have argued that robustness is an aspect of security. I have
exercised editor's discretion by making it a separate section.
The reason for this is that to too many people "security" means
"protection from break-ins" and "authenticating and encrypting
data". This requirement goes beyond those views.
2.1.19. Media Independence
While it is an article of faith that IP operates over a wide variety
of media (such as Ethernet, X.25, ATM, and so on), IP routing must
take an agnostic view toward any "routing" or "topology" services
that are offered by the medium over which IP is operating. That is,
the new architecture must not be designed to integrate with any
media-specific topology management or routing scheme.
The routing architecture must assume, and must work over, the
simplest possible media.
The routing and addressing architecture can certainly make use of
lower-layer information and services, when and where available, and
to the extent that IP routing wishes.
2.1.20. Stand-Alone
The routing architecture and protocols must not rely on other
components of the Internet (such as DNS) for their correct operation.
Routing is the fundamental process by which data "finds its way
around the Internet" and most, if not all, of those other components
rely on routing to properly forward their data. Thus, routing cannot
rely on any Internet systems, services, or capabilities that in turn
rely on routing. If it did, a dependency loop would result.
2.1.21. Safety of Configuration
The architecture, protocols, and standard implementation defaults must be such that a router installed "out of the box" with no configuration, etc., by the operators will not cause "bad things" to happen to the rest of the routing system (e.g., no dial-up customers advertising routes to 18/8!).2.1.22. Renumbering
The routing system must allow topological entities to be renumbered.2.1.23. Multi-Prefix
The architecture must allow topological entities to have multiple prefixes (or the equivalent under the new architecture).2.1.24. Cooperative Anarchy
As RFC 1726[RFC1726] states: A major contributor to the Internet's success is the fact that there is no single, centralized, point of control or promulgator of policy for the entire network. This allows individual constituents of the network to tailor their own networks, environments, and policies to suit their own needs. The individual constituents must cooperate only to the degree necessary to ensure that they interoperate. This decentralization, called "cooperative anarchy", is still a key feature of the Internet today. The new routing architecture must retain this feature. There can be no centralized point of control or promulgator of policy for the entire Internet.2.1.25. Network-Layer Protocols and Forwarding Model
For the purposes of backward compatibility, any new routing and addressing architecture and protocols must work with IPv4 and IPv6 using the traditional "hop-by-hop" forwarding and packet-based multiplex/demultiplex models. However, the architecture need not be restricted to these models. Additional forwarding and multiplex/ demultiplex models may be added.2.1.26. Routing Algorithm
The architecture should not require a particular routing algorithm family. That is to say, the architecture should be agnostic about link-state, distance-vector, or path-vector routing algorithms.
2.1.27. Positive Benefit
Finally, the architecture must show benefits in terms of increased stability, decreased operational costs, and increased functionality and lifetime, over the current schemes. This benefit must remain even after the inevitable costs of developing and debugging the new protocols, enduring the inevitable instabilities as things get shaken out, and so on.2.1.28. Administrative Entities and the IGP/EGP Split
We explicitly recognize that the Internet consists of resources under control of multiple administrative entities. Each entity must be able to manage its own portion of the Internet as it sees fit. Moreover, the constraints that can be imposed on routing and addressing on the portion of the Internet under the control of one administration may not be feasibly extended to cover multiple administrations. Therefore, we recognize a natural and inevitable split between routing and addressing that is under a single administrative control and routing and addressing that involves multiple administrative entities. Moreover, while there may be multiple administrative authorities, the administrative authority boundaries may be complex and overlapping, rather than being a strict hierarchy. Furthermore, there may be multiple levels of administration, each with its own level of policy and control. For example, a large network might have "continental-level" administrations covering its European and Asian operations, respectively. There would also be that network's "inter-continental" administration covering the Europe-to-Asia links. Finally, there would be the "Internet" level in the administrative structure (analogous to the "exterior" concept in the current routing architecture). Thus, we believe that the administrative structure of the Internet must be extensible to many levels (more than the two provided by the current IGP/EGP split). The interior/exterior property is not absolute. The interior/exterior property of any point in the network is relative; a point on the network is interior with respect to some points on the network and exterior with respect to others. Administrative entities may not trust each other; some may be almost actively hostile toward each other. The architecture must accommodate these models. Furthermore, the architecture must not require any particular level of trust among administrative entities.
(next page on part 2)