Thursday night, I arrived at New York's Kennedy Airport and took a commuter flight to Ithaca. USAir decided that 15 cities with no luggage lost was tempting the fates and obliged me by sending my bags to Syracuse. My request for extra frequent flier miles, based on the route my bags took, was greeted with a blank stare.

While negotiating with the lost luggage functionary, H. David Lambert, my host in Ithaca, arrived to pick me up. He smiled knowingly when he saw that I was looking for my bags. Evidently, USAir uses some stochastic routing algorithm for luggage in upper New York state.

Dave brought me to his house, with a stop en route for massive quantities of Chinese take-out. There, David, his wife Carol, and I caught up on old times.

Dave Lambert operates on the front lines of computing: he manages academic computing centers. I got to know him when I was a graduate student at Indiana University and Dave was Assistant Director of the Academic Computing Services.

In those days, Academic Computing meant a CDC 6000, punch cards, and two terminals for the entire campus. My first word processor was on that machine. We would use punch cards and a homegrown text formatter written in FORTRAN to input our formatting directives, with the results sent to a line printer.

Dave brought me into the student consulting pools, rescuing me from yet another semester teaching a required class in economics to drooling undergraduates. My job was to sit in a cubicle and answer any and all questions. In the summer, I continued to do my consulting, but we had only a few students and most of them spent their time at the abandoned quarries swimming.

We did, however, have 6 VAX 11/780s that had just been delivered. My consulting cubicle was a designated receptacle for one of the sets of documentation, which filled up most of a wall. Having nothing else to do. I started at one end of the wall and read manuals. It was then that I learned the key to a successful consulting practice was to Read The FMS Manual (RTFM).

Under Dave's leadership, the Indiana University network grew from a few hundred to several tens of thousands of active users. He introduced a wide variety of programs that dramatically changed how computers were used on campus. An example was the so-called MBA Experiment.

The word "experiment" is a way of labeling some project to get over strenuous objections from people opposed to anything new (hence the use of the word "experiment" at the ITU). The experiment gave permanent accounts to all MBA students, instead of doling out resources as part of some specific class, as had been the case previously.

A single, over-worked VAX was loaded up with 1,200 students, along with tools to do e-mail, financial modeling, and even access external data sources such as the Dow Jones News Service. Non-credit seminars were held to familiarize students with the tools, and a consulting cubicle and terminal room were put into the business school.

The project met strong resistance from two places. First, the systems group in the computing center were aghast at this wild, uncontrolled use of resources. What if everybody started using the system at once? One system manager actually went so far as to say putting users onto the system would make it run inefficiently.

The second source of opposition was the business school faculty, who felt all this new-fangled nonsense would take away from a serious academic atmosphere. Behind the scenes, of course, was the unspoken fact that if the students learned the tools, the faculty would have to also.

Despite these objections, the experiment started up. The results were dramatic. Students started using sophisticated modeling tools to solve case studies, forcing the professors to learn the same tools in order to grade the assignments. Electronic mail became such a part of the culture that one professor even gave his exams by mail.

Eventually, the entire nature of the MBA Program changed. The obligatory Fortran class was dropped in favor of a computer skills seminar emphasizing tools like e-mail and Lotus. Many other classes in the curriculum were radically changed. Students left the program computer literate, with a noticeable change in placement success.

The MBA Experiment was one of dozens of projects that Dave Lambert started, usually over the strenuous objections of one or more factions. He helped wire dormitories, develop educational software in fields like English, obtain cheap PCs for students, and hooked the campus to the Internet.

What Dave started, Carol Lambert had to help finish. Carol specializes in setting up support structures for computing services. At Indiana University, this originally meant writing the reports on how to use the CDC mainframe. This matured over time to a highly trained, highly service-oriented operation, providing everything from a help desk (with people who had answers) to a problem tracking and escalation system, extensive user documentation and classes, and standardized procedures for obtaining resources.

The computing service that Dave and Carol helped build was so successful that it won the first INTEROP Achievement Award for excellence in education. By then, Dave and Carol had moved on to Ithaca, where they were part of a team trying to expand and develop the Cornell campus network.

Cornell is the largest Ivy League school, with a population of 17,000 undergraduates. It is also one of the key Internet hubs. The Cornell Theory Center was one of the original NSF supercomputer centers. Cornell is also the site of the Network Operations Center, supporting Sprint's role as international connection manager for the NSFNET. Cornell monitors international links to NORDUnet and INRIA in France and is part of both the T1 and T3 backbones.

Cornell is also one of the key sites in the Internet for work on routing protocols. In fact, the campus network is run over routers developed at Cornell and running on a PC/AT. These multiprotocol routers support, along with many other networks, over 284 separate AppleTalk networks. The Cornell routers are managed by a set of bootservers. When new software is developed, Trivial FTP is used to send the new boot images down to over 100 routers.

Some of the more interesting routing work at Cornell was done by the team of Scott Brim and Jeff Honig. Scott, quiet by nature, is a bit of an anomaly in the raucous world of routing. Meetings about routing protocols at places like the IETF tend to be full of histrionics and hysterics, routing being the original religious issue in the Internet, but Scott talks in a whisper.

Scott and Jeff are the developers of the Gateway Daemon (gated), a widely available piece of routing software for UNIX and other operating systems. The gated software has its origins in the period when regional networks started coming into being and hooking themselves up to the NSFNET.

TCP/IP has two kinds of routing protocols. Interior protocols are used to communicate reachability information within a routing domain. On UNIX systems, the Routing Information Protocol (RIP) was the quintessential interior protocol and was implemented in a software module called the Route Daemon (routed).

Reachability between routing domains, as in the case of a regional announcing its networks to the backbone NSFNET, uses a protocol such as the Exterior Gateway Protocol (EGP), or its more modern successor, the Border Gateway Protocol (BGP).

In the early days, both routed and EGP received reachability information. Each would take that information and update a master routing table in the UNIX kernel. As long as each routing protocol handles a separate set of target networks, having two modules update a single routing table sort of works.

What started happening, however, was that a target network might be reachable through two different routes. Information about one might arrive via RIP, the other via EGP. The update by one protocol would be walked all over by the other.

The gated software took the two separate modules and put them into a single daemon, allowing coordinated updates to the routing table in the kernel. Over time, gated was revised to make it very modular, allowing new routing protocols to be quickly added.

The gated software has become one of the standard platforms for prototyping new routing protocols. Current gated modules include the OSI IS-IS and its TCP/IP cousin OSPF, the Border Gateway Protocol (BGP), the old RIP-like Hello protocol, and the policy routing twins, IDPR and IDRP.

An interesting aspect of gated is the way it is distributed. Cornell maintains the reference implementation, which people can download from anyplace in the Internet. This makes it perfect for small organizations that have a UNIX box and want to connect to the Internet.

Vendors that incorporate gated in their products sign a redistribution agreement. This agreement requires the vendor to feed any enhancements to gated back into the reference implementation, making any improvements available to the entire community.

<<< • >>>