Home > Articles > Networking > Network Design & Architecture

  • Print
  • + Share This
This chapter is from the book


As we have seen, addressing is a subtle problem, fraught with traps. Early in the development of networks, simple solutions that ignored the major issues were more than sufficient. But as networks grew, the addressing problems should have been investigated. With the exception of two seminal pieces of work, however, they were largely ignored. However, the very long incubation period as an R&D effort (more than 20 years) removed from the pressures of business and used primarily by experts allowed people's ideas to calcify. The effect of Moore's law, increasing power, and decreasing cost of equipment made it possible to ignore the problems until long past the point when they should have been resolved (making it very painful to fix them). Early on (in CYCLADES), it was understood that it was necessary to make a transition from physical to logical address at least once (and even better if more than once). This was supported by Shoch's and then Saltzer's view that applications, nodes, points of attachment, and routes were the fundamental elements of addressing that had to be distinguished. From this and early distributed computing experiments, we recognized that application names were location independent, whereas nodes were location dependent but not route dependent. Although nodes seemed to be synonymous to hosts most of the time, there were counter-examples that showed that this was another false intuition. Oddly enough, it turns out that the only requirement to name a host or a system occurs in network management. Naming hosts is irrelevant to communications.15

This was later refined as topologically dependent. It was still unclear how these properties should manifest themselves. Given how network topologies can change, it was often unclear how this could be accomplished without being too tightly coupled to the physical topology of the network. It even took some time to realize (and is still unlearned by many protocol designers) that the limited scope of some layers meant that not all addresses had to be globally unambiguous. It is a sorry state of affairs that there has been almost no progress in understanding addressing in the past 25 years.

It should also be pointed out that although one can point to these facts in the literature, they were generally not understood by 99% of the engineers involved in networking. Very few, if any, textbooks in the field teach general principles of networking; they generally only teach current practice.16 By the 1990s, current practice was the only general theory most engineers knew. There had always been a tendency to concentrate on research directly applicable to the Internet, instead of understanding the field of networking as a whole. Such general research had always been a fraction of the total, as one would expect, but by the mid-1980s it had pretty much died out entirely. Has the field begun to more resemble an artisan guild than an engineering discipline? This was compounded by no new applications to drive new requirements. The three applications that existed were all special cases that did not expose the full structure. This was not helped by the fact that addressing is a hard problem. Saltzer gave us the basics of what needed to be named, but finding a meaningful interpretation to location dependence was a major stumbling block. Both IP and CLNP made attempts, but both were rooted in the past. Now with all of this background, we are ready to consider how to assemble larger architectural structures.

  • + Share This
  • 🔖 Save To Your Account