Network Working Group D. Thaler Request for Comments: 5218 B. Aboba Category: Informational IAB July 2008
What Makes for a Successful Protocol?
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.
Abstract
The Internet community has specified a large number of protocols to date, and these protocols have achieved varying degrees of success. Based on case studies, this document attempts to ascertain factors that contribute to or hinder a protocol's success. It is hoped that these observations can serve as guidance for future protocol work.
One of the goals of the Internet Engineering Task Force (IETF) is to define protocols that successfully meet their implementation and deployment goals. Based on case studies, this document identifies some of the factors influencing success and failure of protocol designs. It is hoped that this document will be of use to the following audiences:
o IESG members deciding whether to charter a Working Group to do work on a specific protocol;
o Working Group participants selecting among protocol proposals;
o Document authors developing a new protocol specification;
o Anyone evaluating the success of a protocol experiment.
In discussing the factors that help or hinder the success of a protocol, we need to first define what we mean by "success". A protocol can be successful and still not be widely deployed, if it meets its original goals. However, in this document, we consider a successful protocol to be one that both meets its original goals and is widely deployed. Note that "widely deployed" does not mean "inter-domain"; successful protocols (e.g., DHCP [RFC2131]) may be widely deployed solely for intra-domain use.
The following are examples of successful protocols:
Two major dimensions on which a protocol can be evaluated are scale and purpose. When designed, a protocol is intended for some range of purposes and was designed for use on a particular scale.
According to these metrics, a "successful" protocol is one that is used for its original purpose and at the originally intended scale. A "wildly successful" protocol far exceeds its original goals, in terms of purpose (being used in scenarios far beyond the initial design), in terms of scale (being deployed on a scale much greater than originally envisaged), or both. That is, it has overgrown its bounds and has ventured out "into the wild".
Another example of a wildly successful protocol is IPv4. Although it was designed for all purposes ("Everything over IP and IP over Everything"), it has been deployed on a far greater scale than that for which it was originally designed; the limited address space only became an issue after it had already vastly surpassed its original design.
Another example of a successful protocol is ARP. Originally intended for a more general purpose (namely, resolving network layer addresses to link layer addresses, regardless of the media type or network layer protocol), ARP was widely deployed for a narrower scope of uses
Thaler & Aboba Informational [Page 4]
RFC 5218 Protocol Success July 2008
(resolution of IPv4 addresses to Ethernet MAC addresses), but then was adopted for other uses such as detecting network attachment (Detecting Network Attachment in IPv4 (DNAv4) [RFC4436]). Also, like IPv4, ARP is deployed on a much greater scale (in terms of number of machines, but not number on the same subnet) than originally expected.
Wild success can be both good and bad. A wildly successful protocol is so useful that it can solve more problems or address more scenarios or devices. This may indicate that it is time to revise the protocol to better accommodate the new design space.
However, if a protocol is used for a purpose other than what it was designed for:
o There may be undesirable side effects because of design decisions that are appropriate for the originally intended purpose, but inappropriate for the new purpose.
o There may be performance problems if the protocol was not designed to scale to the extent to which it was deployed.
o Implementers may attempt to add or change functionality to work around the design limitations without complete understanding of their effect on the overall protocol behavior and invariants.
o Wildly successful protocols become high value targets for attackers because of their popularity and the potential for exploitation of uses or extensions that are less well understood and tested than the original protocol.
A wildly successful protocol is therefore vulnerable to "death by success", collapsing as a result of attacks or scaling limitations.
Failure, or the lack of success, cannot be determined before allowing sufficient time to pass (e.g., 5-10 years for an average protocol). Failure criteria include:
o No mainstream implementations. There is little or no support in hosts, routers, or other classes of relevant devices.
o No deployment. Devices that support the protocol are not deployed, or if they are, then the protocol is not enabled.
o No use. While the protocol may be deployed, there are no applications or scenarios that actually use the protocol.
At the time a protocol is first designed, the three above conditions hold, which is why it is important to allow sufficient time to pass before evaluating the success or failure of a protocol.
The lack of a value chain can make it difficult for a new protocol to progress from implementation to deployment to use. While the term "chicken-and-egg" problem is sometimes used to describe the lack of a value chain, the lack of implementation, deployment, or use is not the cause of failure, it is merely a symptom.
There are many strategies that have been used in the past for overcoming the initial lack of implementations, deployment, and use, although none of these guarantee success. For example:
o Address a critical and imminent problem. If the need is severe enough, the industry is incented to adopt it as soon as implementations exist, and well-known need is sufficient to motivate implementations. For example, NAT provided an immediate address sharing capability to the individual deploying it (Appendix A.8). Thus, when creating a protocol, consider whether it can be easily tailored or expanded to directly target a critical problem; if it only solves part of the problem, consider what would be needed in addition.
o Provide a "killer app" with low deployment costs. This strategy can be used to generate demand where none existed before. See the HTTP case study in Appendix A.1 for an example.
o Provide value for existing unmodified applications. This solves the chicken-and-egg problem by ensuring that use exists as soon as the protocol is deployed, and therefore, the benefit can be realized immediately. See the Wired Equivalent Privacy (WEP) case study in Appendix A.6 for an example.
Thaler & Aboba Informational [Page 6]
RFC 5218 Protocol Success July 2008
o Reduce complexity and cost by narrowing the intended purpose and/or scope to an area where it is easiest to succeed. This may allow removing complexity that is not required for the narrow purpose. Removing complexity reduces the cost of implementation and deployment to where the resulting cost may be very low compared to the benefit. For example, link-scoped multicast is far more successful than, say, inter-domain multicast (see Appendix A.4).
o A government or other entity may provide incentives or disincentives that motivate implementation and deployment. For example, specific cryptographic algorithms may be mandated. As another example, Japan started an economic incentive program to generate IPv6 [RFC2460] implementations and deployment.
As we will see, such strategies are often successful because they directly target the top success factors.
In this section, we identify factors that contribute to success and "wild" success.
Note that a successful protocol will not necessarily include all the success factors, and some success factors may be present even in failed designs. Nevertheless, experience appears to indicate that the presence of success factors seems to improve the probability of success.
The success factors, and their relative importance, were suggested by a series of case studies (Appendix A).
It is critical to the success of a protocol that the benefits of deploying the protocol (monetary or otherwise) outweigh the costs, which include:
o Hardware cost: Protocols that don't require hardware changes are easier to deploy than those that do. Overlay networks are one way to avoid requiring hardware changes. However, often hardware updates are required even for protocols whose functionality could be provided solely in software. Vendors often implement new
Thaler & Aboba Informational [Page 7]
RFC 5218 Protocol Success July 2008
functionality only within later branches of the code tree, which may only run on new hardware. As a result, the safest way to avoid hardware upgrade cost is to design for backward compatibility with both existing hardware and software.
o Operational interference: Protocols that don't require changes to other operational processes and tools are easier to deploy than ones that do. For example, IPsec [RFC4301] interferes with NetFlow [RFC3954] deep packet inspection, which can be important to operators.
o Retraining: Protocols that have no configuration, or are very easy to configure/manage, are cheaper to deploy.
o Business dependencies: Protocols that don't require changes to a business model (whether for implementers or deployers) are easier to deploy than ones that do. There are costs associated with changing billing and accounting systems and retraining of associated personnel, and in addition, the assumptions on which the previous business model was based may change. For example, some time ago many service providers had business models built around dial-up with an assumption that machines were not connected all the time; protocols that desired always-on connectivity required the business model to change since the networks were not optimized for always-on. Similarly, some service providers have business models that assume that upstream bandwidth is underutilized; peer-to-peer protocols may require this business model to change. Finally, many service providers have business models based on charging for the amount of bandwidth consumed on the link to a customer; multicast protocols interfere with this business model since they provide a way for a customer to consume less bandwidth on the source link by sending multicast traffic, as opposed to paying more to source many unicast streams, without having some other mechanism to cover the cost of replication in the network (e.g., router CPU, downstream link bandwidth, extra management). Multicast protocols also complicate business models based on charging the source of traffic based on the amount of multicast replication, since the source may not be able to predict the cost until a bill is received.
Similarly, there are many types of benefits, including:
o Relieving pain: Protocols that drastically lower costs (monetary or otherwise) that exist prior to deploying the protocol are easier to show direct benefit from, since they address a burning need.
Thaler & Aboba Informational [Page 8]
RFC 5218 Protocol Success July 2008
o Enabling new scenarios: Protocols that enable new capabilities, scenarios, or user experiences can provide significant value, although the benefit may be harder to realize, as there may be more risk involved.
o Incremental improvements: Protocols that provide incremental improvements (e.g., better video quality) generate a small benefit, and hence can be successful as long as the cost is small.
There are at least two example cases of cost/benefits tradeoffs. In the first case, even upon initial deployment, the benefit outweighs the cost. In the second case, there is an upfront cost that outweighs the initial benefit, but the benefit grows over time (e.g., as more nodes or applications support it). The former model is much easier to get initial deployment, but over time both can be successful. The second model has a danger for the initial deployments, that if others don't deploy the protocol then the initial deployers have lost value, and so they must take on some risk in deploying the protocol.
Success most easily comes when the natural incentive structure is aligned with the deployment requirements. That is, those who are required to deploy, manage, or configure something are the same as those who gain the most benefit. In summary, it is best if there is significant positive net value at each organization where a change is required.
A protocol is incrementally deployable if early adopters gain some benefit even though the rest of the Internet does not support the protocol. There are several aspects to this.
Protocols that can be deployed by a single group or team (e.g., intra-domain) have a greater chance of success than those that require cooperation across organizations (or, in the worst case require a "flag day" where everyone has to change simultaneously). For example, protocols that don't require changes to infrastructure (e.g., router changes, service provider support, etc.) have a greater chance of success than ones that do, since less coordination is needed, NAT being a canonical example. Similarly, protocols that provide benefit when only one end changes have a greater chance of success than ones that require both ends of communication to support the protocol.
Finally, protocol updates that are backward compatible with older implementations have a greater chance of success than ones that aren't.
Protocols with freely available implementation code have a greater chance of success than protocols without. Often, this is more important than any technical consideration. For example, it can be argued that when deciding between IPv4 and Internetwork Packet Exchange (IPX) [IPX], this was the determining factor, even though, in many ways, IPX was technically superior to IPv4. Similar arguments have been made for the success of RADIUS [RFC2865] over TACACS+ [TACACS+]. See Appendix A for further discussion.
Freedom from usage restrictions means that anyone who wishes to implement or deploy can do so without legal or financial hindrance. Within the IETF, this point often comes up when evaluating between technologies, one of which has known Intellectual Property associated with it. Often the industry chooses the one with no known Intellectual Property, even if it is technically inferior.
Open specification availability means the protocol specification is made available to anyone who wishes to use it. This is true for all Internet Drafts and RFCs, and it has contributed to the success of protocol specifications developed within or contributed to the IETF.
The various aspects of this factor include:
o World-wide distribution: Is the specification accessible from anywhere in the world?
o Unrestricted distribution: Are there no legal restrictions on getting the specification?
o Permanence: Does the specification remain even after the creator is gone?
o Stability: Is there a stable version of the specification that does not change?
This factor means that the protocol is maintained by open processes, mechanisms exist for public comment on the protocol, and the protocol maintenance process allows the participation of all constituencies that are affected by the protocol.
This factor means that the protocol follows good design principles that lead to ease of implementation and interoperability, such as those described in "Architectural Principles of the Internet" [RFC1958]. For example, simplicity, modularity, and robustness to failures are all key design factors. Similarly, clarity in specifications is another aspect of good technical design that facilitates interoperability and ease of implementation. However, experience shows that good technical design has minimal impact on initial success compared with other factors.
Protocols that are extensible are more likely to be wildly successful in terms of being used for purposes outside their original design. An extensible protocol may carry general purpose payloads/options, or may be easy to add a new payload/option type. Such extensibility is desirable for protocols that intend to apply to all purposes (like IP). However, for protocols designed for a specialized purpose, extensibility should be carefully considered before including it.
Protocols that have no inherent limit near the edge of the originally envisioned scale are more likely to be wildly successful in terms of scale. For example, IPv4 had no inherent limit near its originally envisioned scale; the address space limit was not hit until it was already wildly successful in terms of scale. Another type of inherent limit would be a performance "knee" that causes a meltdown (e.g., a broadcast storm) once a scaling limit is passed.
The more successful a protocol becomes, the more attractive a target it will be. Protocols with security flaws may still become wildly successful provided that they are extensible enough to allow the flaws to be addressed in subsequent revisions. Examples include Secure SHell version 1 (SSHv1) and IEEE 802.11 with WEP. However, the combination of security flaws and limited extensibility tends to be deadly. For example, some early server-based multiplayer games ultimately failed due to insufficient protections against cheating, even though they were initially successful.
The case studies described in Appendix A indicate that the most important initial success factors are filling a real need and being incrementally deployable. When there are competing proposals of comparable benefit and deployability, open specifications and code become significant success factors. Open source availability is initially more important than open specification maintenance.
In most cases, technical quality was not a primary factor in initial success. Indeed, many successful protocols would not pass IESG review today. Technically inferior proposals can win if they are openly available. Factors that do not seem to be significant in determining initial success (but may affect wild success) include good design, security, and having an open specification maintenance process.
Many of the case studies concern protocols originally developed outside the IETF, which the IETF played a role in improving only after initial success was certain. While the IETF focuses on design quality, which is not a factor in determining initial protocol success, once a protocol succeeds, a good technical design may be key to it staying successful, or in dealing with wild success. Allowing extensibility in an initial design enables initial shortcomings to be addressed.
Security vulnerabilities do not seem to limit initial success, since vulnerabilities often become interesting to attackers only after the protocol becomes widely deployed enough to become a useful target. Finally, open specification maintenance is not important to initial success since many successful protocols were initially developed outside the IETF or other standards bodies, and were only standardized later.
In light of our conclusions, we recommend that the following questions be asked when evaluating protocol designs:
o Does the protocol exhibit one or more of the critical initial success factors?
o Are there implementers who are ready to implement the technology in ways that are likely to be deployed?
o Are there customers (especially high-profile customers) who are ready to deploy the technology?
o Are there potential niches where the technology is compelling?
Thaler & Aboba Informational [Page 12]
RFC 5218 Protocol Success July 2008
o If so, can complexity be removed to reduce cost?
o Is there a potential killer app? Or can the technology work underneath existing unmodified applications?
o Is the protocol sufficiently extensible to allow potential deficiencies to be addressed in the future?
o If it is not known whether the protocol will be successful, should the market decide first? Or should the IETF work on multiple alternatives and let the market decide among them? Are there factors listed in this document that may predict which is more likely to succeed?
In the early stages (e.g., BOFs, design of new protocols), evaluating the initial success factors is important in facilitating success. Similarly, efforts to revise unsuccessful protocols should evaluate whether the initial success factors (or enough of them) were present, rather than focusing on wild success, which is not yet a problem. For a revision of a successful protocol, on the other hand, focusing on the wild success factors is more appropriate.
This document discusses attributes that affect the success of protocols. It has no specific security implications. Recommendations on security in protocol design can be found in [RFC3552].
[RFC0793] Postel, J., "Transmission Control Protocol", STD 7, RFC 793, September 1981.
[RFC0826] Plummer, D., "Ethernet Address Resolution Protocol: Or converting network protocol addresses to 48.bit Ethernet address for transmission on Ethernet hardware", STD 37, RFC 826, November 1982.
[RFC0959] Postel, J. and J. Reynolds, "File Transfer Protocol", STD 9, RFC 959, October 1985.
[RFC1035] Mockapetris, P., "Domain names - implementation and specification", STD 13, RFC 1035, November 1987.
[RFC1058] Hedrick, C., "Routing Information Protocol", RFC 1058, June 1988.
[RFC1436] Anklesaria, F., McCahill, M., Lindner, P., Johnson, D., Torrey, D., and B. Alberti, "The Internet Gopher Protocol (a distributed document search and retrieval protocol)", RFC 1436, March 1993.
[RFC1661] Simpson, W., "The Point-to-Point Protocol (PPP)", STD 51, RFC 1661, July 1994.
[RFC1866] Berners-Lee, T. and D. Connolly, "Hypertext Markup Language - 2.0", RFC 1866, November 1995.
[RFC1958] Carpenter, B., "Architectural Principles of the Internet", RFC 1958, June 1996.
[RFC2131] Droms, R., "Dynamic Host Configuration Protocol", RFC 2131, March 1997.
[RFC2328] Moy, J., "OSPF Version 2", STD 54, RFC 2328, April 1998.
[RFC2460] Deering, S. and R. Hinden, "Internet Protocol, Version 6 (IPv6) Specification", RFC 2460, December 1998.
[RFC2616] Fielding, R., Gettys, J., Mogul, J., Frystyk, H., Masinter, L., Leach, P., and T. Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1", RFC 2616, June 1999.
Thaler & Aboba Informational [Page 14]
RFC 5218 Protocol Success July 2008
[RFC2821] Klensin, J., "Simple Mail Transfer Protocol", RFC 2821, April 2001.
[RFC2865] Rigney, C., Willens, S., Rubens, A., and W. Simpson, "Remote Authentication Dial In User Service (RADIUS)", RFC 2865, June 2000.
[RFC3022] Srisuresh, P. and K. Egevang, "Traditional IP Network Address Translator (Traditional NAT)", RFC 3022, January 2001.
[RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A., Peterson, J., Sparks, R., Handley, M., and E. Schooler, "SIP: Session Initiation Protocol", RFC 3261, June 2002.
[RFC3552] Rescorla, E. and B. Korver, "Guidelines for Writing RFC Text on Security Considerations", BCP 72, RFC 3552, July 2003.
[RFC3954] Claise, B., "Cisco Systems NetFlow Services Export Version 9", RFC 3954, October 2004.
[RFC4120] Neuman, C., Yu, T., Hartman, S., and K. Raeburn, "The Kerberos Network Authentication Service (V5)", RFC 4120, July 2005.
[RFC4251] Ylonen, T. and C. Lonvick, "The Secure Shell (SSH) Protocol Architecture", RFC 4251, January 2006.
[RFC4271] Rekhter, Y., Li, T., and S. Hares, "A Border Gateway Protocol 4 (BGP-4)", RFC 4271, January 2006.
[RFC4301] Kent, S. and K. Seo, "Security Architecture for the Internet Protocol", RFC 4301, December 2005.
[RFC4436] Aboba, B., Carlson, J., and S. Cheshire, "Detecting Network Attachment in IPv4 (DNAv4)", RFC 4436, March 2006.
[RFC4864] Van de Velde, G., Hain, T., Droms, R., Carpenter, B., and E. Klein, "Local Network Protection for IPv6", RFC 4864, May 2007.
[TACACS+] Carrel, D. and L. Grant, "The TACACS+ Protocol, Version 1.78", Work in Progress, January 1997.
Thaler & Aboba Informational [Page 15]
RFC 5218 Protocol Success July 2008
[WAP] Open Mobile Alliance, "Wireless Application Protocol Architecture Specification", <http:// www.openmobilealliance.org/tech/affiliates/ LicenseAgreement.asp?DocName=/wap/ wap-210-waparch-20010712-a.pdf>.
In this Appendix, we include several case studies to illustrate the importance of potential success factors. Many other equally good case studies could have been included, but, in the interests of brevity, only a sampling is included here that is sufficient to justify the conclusions in the body of this document.
Positive net value: HTTP [RFC2616] with HTML [RFC1866] provided substantially more value than Gopher [RFC1436] and FTP [RFC0959]. Among other things, HTML/HTTP provided support for forms, which opened the door for commercial uses of the technology. In this sense, it enabled new scenarios. Furthermore, it only required changes by entities that received benefits; hence, the cost and benefits were aligned.
Incremental deployability: Browsers and servers were incrementally deployable, but initial browsers were also backward compatible with existing protocols such as FTP and Gopher.
Open code availability: Server code was open. Client source code was initially open to academic use only.
Restriction-free: Academic use licenses were freely available. HTML encumbrance only surfaced later.
Open specification availability: Yes.
Open maintenance process: Not at first, but eventually. This illustrates that it is not necessary to have an open maintenance process at first to achieve success. The maintenance process becomes important after initial success.
Good technical design: Fair. Initially, there was no support for graphics, HTML was missing many SGML [ISO-8879] features, and HTTP 1.0 had issues with congestion control and proxy support. These sorts of issues would typically prevent IESG approval today. However, they did not prevent the protocol from becoming successful.
Extensible: Extensibility was excellent along multiple dimensions, including HTTP, HTML, graphics, forms, Java, JavaScript, etc.
No hard scalability bound: Excellent. There was no registration process, as there was with Gopher, which allowed it to scale much better.
Threats sufficiently mitigated: No. There was initially no support for confidentiality (e.g., protection of credit card numbers), and HTTP 1.0 had cleartext passwords in Basic auth.
HTML/HTTP addressed scenarios that no other protocol addressed. Since deployment was easy, the protocol quickly took off. Only after HTML/HTTP became successful did security become an issue. HTML/ HTTP's initial success occurred outside the IETF, although HTTP was later standardized and refined, addressing some of the limitations.
Positive net value: There were initially many competitors, including IPX, AppleTalk, NetBEUI, OSI, and DECNet. All of them had positive net value. However, NetBEUI and DECNet were not designed for internetworking, which limited scalability and eventually stunted their growth.
Incremental deployability: None of the competitors (including IPv4) had incremental deployability, although there were few enough nodes that a flag day was manageable at the time.
Open code availability: IPv4 had open code from BSD, whereas IPX did not. Many argue that this was the primary reason for IPv4's success.
Restriction-free: Yes for IPv4; No for IPX.
Open specification availability: Yes for IPv4; No for IPX.
Open maintenance process: Yes for IPv4; No for IPX.
Good technical design: The initial design of IPv4 was fair, but arguably IPX was initially better. Improvements to IPv4 such as DHCP came much later.
Initially, it wasn't clear that IPv4 would win. There were also other competitors, such as OSI. However, the Advanced Research Projects Agency (ARPA) funded IPv4 implementation on BSD and this open source initiative led to many others picking up IPv4, which ultimately made a difference in IPv4 succeeding rather than its competitors. Even though IPX initially had a technically superior design, IPv4 succeeded because of its openness.
Positive net value: SSH [RFC4251] provided greater value than competitors. Kerberized telnet required deployment of a Kerberos server. IPsec required a public key infrastructure (PKI) or pre- shared key authentication. While the benefits were comparable, the overall costs of the alternatives were much higher, and they potentially required deployment by entities that did not directly receive benefit. Hence, unlike the alternatives, the cost and benefits of SSH were aligned.
Incremental deployability: Yes, SSH required SSH clients and servers, but did not require a key distribution center (KDC) or credential pre-configuration.
Open code availability: Yes
Restriction-free: It is unclear whether SSH was truly restriction- free or not.
Open specification availability: Not at first, but eventually.
Open maintenance process: Not at first, but eventually.
Thaler & Aboba Informational [Page 19]
RFC 5218 Protocol Success July 2008
Good technical design: SSHv1 was fair. It had a number of technical issues that were addressed in SSHv2.
The "leap of faith" trust model (accept an untrusted certificate the first time you connect) was initially criticized by "experts", but was popular with users. It provided vastly more functionality and didn't require a KDC and so was easy to deploy. These factors made SSH a clear winner.
A.4. Inter-Domain IP Multicast vs. Application Overlays
We now look at a protocol that has not been successful (i.e., has not met its original design goals) after a long period of time has passed. Note that this discussion applies only to inter-domain multicast, not intra-domain or intra-subnet multicast.
Positive net value: Unclear. When many receivers of the same stream exist, the benefit relieves pain near the sender, and in some cases enables new scenarios. However, when few receivers exist, the benefits are only incremental improvements when compared with multiple streams. While there was positive value in bandwidth savings, this was offset by the lack of viable business models, and lack of tools. Hence, the costs generally outweighed the benefits.
Furthermore, the costs are not necessarily aligned with the benefits. Inter-domain Multicast requires changes by (at least) applications, hosts, and routers. However, it is the applications that get the primary benefit. For application layer overlaps, on the other hand, only the applications need to change, and hence the cost is lower (and so are the benefits), and cost and benefits are aligned.
Incremental deployability: Poor for inter-domain multicast, since it required every router in the end-to-end path between a source and any receiver to support multicast. This severely limited the
Thaler & Aboba Informational [Page 20]
RFC 5218 Protocol Success July 2008
deployability of native multicast. Initially, the strategy was to use an overlay network (the Multicast Backbone (MBone)) to work around this. However, the overlay network eventually suffered from performance problems at high fan-out points, and so adding another node required more coordination with other organizations to find someone that was not overloaded and agreed to forward traffic on behalf of the new node.
Incremental deployability was good for application-layer overlays, since only the applications need to change. However, benefit only exists when the sender(s) and receivers both support the overlay mechanism.
Open code availability: Yes.
Restriction-free: Yes.
Open specification availability: Yes for inter-domain multicast. Application-layer overlays are not standardized, but left to each application.
Open maintenance process: Yes for inter-domain multicast. Application-layer overlays are not standardized, but left to each application.
Good technical design: This is debatable for inter-domain multicast. In many respects, the technical design is very efficient. In other respects, it results in per-connection state in all intermediate routers, which is questionable at best. Application-layer overlays do not have the disadvantage, but receive a smaller benefit.
No hard scalability bound: Inter-domain multicast had scalability issues in terms of number of groups, and in terms of number of sources. It scaled extremely well in terms of number of receivers. Application-layer overlays scale well in all dimensions, except that they do not scale as well as inter-domain multicast in terms of bandwidth since they still result in multiple streams over the same link.
Threats sufficiently mitigated: No for inter-domain-multicast, since untrusted hosts can create state in intermediate routers along an entire path. Better for application-layer multicast.
Because the benefits weren't enough to outweigh the costs for entities (service providers and application developers) to use it, instead the industry has tended to choose application overlays with replicated unicast.
Positive net value: Compared to competitors such as HTTP/TCP/IP, and NTT DoCoMo's i-mode [IMODE], the relative value of WAP is unclear. It suffered from a poor experience, and a lack of tools.
Incremental deployability: Poor. WAP required a WAP-to-HTTP proxy in the service provider and WAP support in phones; adding a new site often required participation by the service provider.
Open code availability: No.
Restriction-free: No. WAP has two patents with royalties required.
Open specification availability: No.
Open maintenance process: No, there was a US$27000 entrance fee.
Good technical design: No, a common complaint was that WAP was underspecified and hence interoperability was difficult.
There were a number of close competitors to WAP. Incremental deployability was easier with the competitors, and the restrictions on code and specification access were significant factors that hindered its ability to succeed.
Positive net value: Yes. WEP provided security when there was no alternative, and it only required changes by entities that got benefit.
Incremental deployability: Yes. Although one needed to configure both the access point and stations, each wireless network could independently deploy WEP.
Open code availability: Essentially no, because of Rivest Cipher 4 (RC4).
Restriction-free: No for RC4, but otherwise yes.
Open specification availability: No for RC4, but otherwise yes.
Open maintenance process: Yes.
Good technical design: No, WEP had an inappropriate use of RC4.
Even though WEP was not completely open and restriction free, and did not have a good technical design, it still became successful because it was incrementally deployable and it provided significant value when there was no alternative. This again shows that value and deployability are more significant success factors than technical design or openness, particularly when no alternatives exist.
Positive net value: Yes for both, and it only required changes by entities that got benefit.
Incremental deployability: Yes for both (just change clients and servers).
Open code availability: Yes for RADIUS; initially no for TACACS+, but eventually yes.
Restriction-free: Yes for RADIUS; unclear for TACACS+.
Open specification availability: Yes for RADIUS; initially no for TACACS+, but eventually yes.
Open maintenance process: Initially no for RADIUS, but eventually yes. No for TACACS+.
Good technical design: Fair for RADIUS (there was no confidentiality support, and no authentication of Access Requests; it had home grown ciphersuites based on MD5). Good for TACACS+.
No hard scalability bound: Excellent for RADIUS (UDP-based); good for TACACS+ (TCP-based).
Threats sufficiently mitigated: No for RADIUS (no support for confidentiality, existing implementations are vulnerable to dictionary attacks, use of MD5 now vulnerable to collisions). TACACS+ was better since it supported encryption.
Since both provided positive net value and were incrementally deployable, those factors were not significant. Even though TACACS+ had a better technical design in most respects, and eventually provided openly available specifications and source code, the fact that RADIUS had an open maintenance process as well as openly available specifications and source code early on was the determining factor. This again shows that having a better technical design is less important in determining success than other factors.
Although NAT is not, strictly speaking, a "protocol" per se, but rather a "mechanism" or "algorithm", we include a case study since there are many mechanisms that only require a single entity to change to reap the benefit (TCP congestion control algorithms are another example in this class), and it is important to include this class of mechanisms in the discussion since it exemplifies a key point in the discussion of incremental deployability.
Positive net value: Yes. NATs provided the ability to connect multiple devices when only a limited number of addresses were available, and also provided a (limited) security boundary as a side effect. Hence, it both relieved pain involved with acquiring multiple addresses, as well as enabled new scenarios. Finally, it only required deployment by the entity that got the benefit.
Incremental deployability: Yes. One could deploy a NAT without coordinating with anyone else, including a service provider.
Open code availability: Yes.
Restriction-free: Yes at first (patents subsequently surfaced).
Open specification availability: Yes.
Open maintenance process: Yes.
Good technical design: Fair. NAT functionality was underspecified, leading to unpredictable behavior in general. In addition, NATs caused problems for certain classes of applications.
The absence of an unambiguous specification was not a hindrance to initial success since the test cases weren't well defined; therefore, each implementation could decide for itself what scenarios it would handle correctly.
Even with its technical problems, NAT succeeded because of the value it provided. Again, this shows that the industry is willing to accept technically problematic solutions when there is no alternative and the technology is easy to deploy.
Indeed, NAT became wildly successful by being used for additional purposes [RFC4864], and to a large scale including multiple levels of NATs in places.
Appendix B. IAB Members at the Time of This Writing
Loa Andersson Leslie Daigle Elwyn Davies Kevin Fall Russ Housley Olaf Kolkman Barry Leiba Kurtis Lindqvist Danny McPherson David Oran Eric Rescorla Dave Thaler Lixia Zhang
Thaler & Aboba Informational [Page 26]
RFC 5218 Protocol Success July 2008
Authors' Addresses
Dave Thaler IAB One Microsoft Way Redmond, WA 98052 USA
This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights.
This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM 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.
Intellectual Property
The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-ipr@ietf.org.