Afterword: The Future of Link State Protocols

OSPF and IS-IS are both almost 20 years old, and are based on a mathematical algorithm that is almost 50 years old; in Internet years, this is nearly prehistoric. They were designed in the days when processors were slow, memory was expensive, and each router added significant latency to the network. As a result both protocols have features and characteristics that no longer matter, meant to solve problems that no longer exist. The 32-bit alignment of all OSPF data units for orderly processing and the original 6-bit IS-IS metric for faster processing are just two examples. Although the operational experience gained from the original link state protocol used in the ARPANET led to some key design differences in OSPF and IS-ISsuch as the use of linear rather than circular sequence number spacessome of the problems the protocols were designed to address were speculative rather than tied to practical experience. And a few features originally intended for one purpose have since been utilized for another purpose. The IS-IS overload (OL) capability is a good example: The original intent was to signal memory overload in a router. Today, memory limits are not a serious issue, but the OL function is used to prevent the blackholing of packets during BGP convergence. This practice has become so common that a capability similar to the IS-IS overload has been added to OSPF.

As the change in application of overloading would indicate, modern changes to OSPF and IS-IS are no longer motivated by speculative problems but instead driven by practical experience. Throughout the 1990s, the challenges were to enhance the scalability and reliability of the protocols to keep pace with the enormous increase in the size of IP networks. This growth continues, and beginning in the late 1990s the increasing interconnectedness of IP networks added security to the list of issues OSPF and IS-IS must address. Also beginning in the late 1990s and continuing to the present day are new capabilities in IP networks requiring extensions to OSPF and IS-IS such as IP multicast, IPv6, and MPLS.

Some of the newest networking capabilities being developed spring from experience with MPLS. MPLS separates the control and signaling functions of networks from the switching and forwarding functions, so that control and signaling can be isolated to the network access nodes. That is, MPLS keeps the "intelligence" of the network at the edges. The benefit of MPLS, as discussed in Chapter 11, is that you can use it to build virtual circuits across IP networks, thereby providing what appears to be a connection-oriented service over a connectionless network. These virtual circuits, called label switched paths (LSPs), can be used for traffic engineering, QoS, and for building various kinds of virtual private networks (VPNs). The concept has been successful enough that MPLS is being extended to create LSPs across not only packet switched networks but also time domain networks such as SONET/SDH, wave domain networks such as lambda switching, and special domain networks such as fiber switches. A version of MPLS called Generalized MPLS (GMPLS) extends a common control plane not only to label switching routers and ATM switches but also to the lower-level elements of a network such as DWDM systems, optical cross-connects, and add-drop multiplexors (ADMs). OSPF and IS-IS are being extended to support traffic engineering capabilities in GMPLS as they presently do for MPLS.

Given the age of link state protocols in general, and OSPF and IS-IS in particular, one must ask whether link state protocols will be replaced in the future by some newer, better algorithm. It is true that new algorithms are proposed in academic papers from time to time. But no algorithm has stimulated widespread interest either among the operators of large-scale networks or among the router vendors that supply these operators. So although some new, improved algorithm might eventually supplant link state protocols, this will not happen in the foreseeable future.

And although growing and evolving IP networks produce new network capabilities and applications, engineers usually prefer to hold on to what is familiar and proven whenever possible. Hence, OSPF and IS-IS continue to be extended to support these new network technologies. The arc of engineering and operational experience invested in these two protocols over the past 20 years, and their ongoing extensibility, indicate that OSPF and IS-IS will continue to be the preferred routing protocols for large IP networks for many years to come.