Section 1.7. Separate But Equal
1.7. Separate But EqualOSI appealed to governments, telcos, and other organizations that were much more comfortable with established standards bodies than with the freewheeling style of the IETF. And because such organizations represented major customer bases, the computer and networking vendors of the time set to work developing OSI-compliant protocol suites. Novell (NetWare), Banyan (VINES), General Motors (MAP and TOP), Apple (AppleTalk), and others scrambled to show how their network operating systems fit into the OSI reference model (sometimes a tight squeeze). But the company that made the most progress on OSI, and whose protocols became synonymous with OSI protocols, was Digital Equipment Corporation (DEC). DEC had already established their Digital Network Architecture (DNA) in the mid-1970s and had developed four versions of their DECnet software as implementations of DNA. They called each version a phase, specifically DECnet Phases I through IV. In 1987, DEC introduced DECnet Phase V, and began selling products supporting Phase V in 1991.[31] This newest version moved sharply away from the earlier versions of DECnet to comply with the OSI model. In 1987, the ISO adopted the work done by DEC, and so DECnet Phase V is mostly indistinguishable from what we now call the OSI protocol suite.
The network routing protocol used by DECnet Phase V, developed at DEC by Radia Perlman, Mike Shand, Dave Oran, and others, was adopted in its entirety by the ISO as ISIS. "ISO's standard for routing CLNP is known as IS-IS," Perlman writes, "because all the other names (e.g., 10589: 'Intermediate system to Intermediate system Intra-Domain routeing information exchange protocol for use in Conjunction...') are worse."[32]
Around the same time in 1987 that the ISO adopted IS-IS, the IETF recognized the need for a link state interior gateway protocol. The NSFNET backbones and the regional networks at the time were either using static routes or, where dynamic routing was needed, RIP. Static routes did not scale for simple management reasons, and RIP was showing many of the vulnerabilities at scale that the early ARPANET Bellman-Ford protocol showed. With the experience gained from developing and operating the ARPANET's SPF protocol, developing a link state IGP for use at larger scales seemed logical. But this move created two camps in the IETF. One camp looked at IS-IS and decided that it made little sense to develop a new link state protocol when one was already at hand. Why not just extend IS-IS to support TCP/IP? The other camp did not want a protocol that was controlled by an outside body, particularly one as rigidly bureaucratic as the ISO. The IETF approach was proven and familiar, so why not develop an open, nonproprietary version of the ARPANET's SPFOSPFto better coexist with the open TCP/IP? A visceral resentment of the ISO's arrogance in dismissing TCP/IP also fed the second camp; IS-IS was unacceptable simply because it was an ISO protocol. Rather than choose between the competing camps, the IETF decided to compromise by accepting both an extended IS-IS and a home-grown OSPF as separate but equal protocols. An IS-IS Working Group and an OSPF Working Group were formed. The IS-IS Working Group finished extending IS-IS to support IP in 1990 and called the extended version "Integrated" or "Dual" IS-IS. The IP extensions were published in RFC 1195, authored by Ross Callon, a Digital Equipment Corporation engineer who had previously worked at BBN.[33]
The OSPF Working Group produced their first version of OSPF in October 1989. However, this first version (OSPFv1) revealed several operational problems and some areas where the protocol could be optimized; it was never deployed. The working group revised the protocol, and OSPFv2 was published in RFC 1247, authored by John Moy, in July 1991.[34] Moy worked at Proteon, Inc., an early router manufacturer, and like Callon was an ex-BBN engineer.
OSPF was deployed successfully in several regional networks in 1990, and was demonstrated successfully at INTEROP in October 1991. "It is difficult," Moy wrote, "to create a flashy demonstration of a routing protocol. When the routing protocol is working, one does not really notice that it is there at all."[35]
Throughout the development process, the two working groups borrowed from each other. The concept of a designated router on broadcast networks, for instance, which is used by both OSPF and IS-IS, was first developed for IS-IS. They also both used lessons learned from the ARPANET's SPF protocol. For example, the surprising meltdown of the ARPANET on October 27, 1980, described in RFC 789, was partially due to the protocol's circular sequence number space. Both IS-IS and OSPF adopted a linear sequence number space (described in Chapter 2). The working groups also recognized the complexity of adaptive metrics, and specified configurable, nonadaptive metrics. By the mid-1990s Cisco Systems was well on its way to becoming the dominant commercial supplier of routers for the Internet backbone and regional networks. Cisco began supporting both OSPF and the OSI version of IS-IS in 1991; the following year, they released an implementation of Integrated IS-IS for IP support. But the significant event for IS-IS, leading to its use in service provider networks worldwide, happened in 1994 when Cisco began supporting NLSP. Novell Networks had some years earlier begun developing a link state protocol for its NetWare network operating system. Under the direction of Neil Castagnoli,[36] Novell released their NetWare Link Services Protocol (NLSP). NLSP was essentially IS-IS adapted for routing Novell IPX. (Radia Perlman joined Novell shortly after NLSP was released, leading to the myth that she created NLSP from IS-IS.)
When Cisco Systems implemented NLSP, it decided to rewrite its IS-IS code to merge the two very similar protocols internally in IOS wherever possible. Dave Katz conducted the rewrite, and the result was a reliable, robust IS-IS implementation. The Cisco OSPF implementation at the time was less refined, and service providers were becoming dissatisfied with it. Partly spurred by the OSI mania of the time, many switched to IS-IS and have been convinced IS-IS users ever since. The Cisco OSPF code has of course become just as reliable, but those experiences in 1994 through 1996 led many to argue even today that IS-IS is a more reliable protocol. |
EAN: 2147483647
Pages: 111