tech TALK

A friend of mine recently checked his e-mail after an early morning struggle with a problem set. Although staying up into the wee hours of the morning to complete work is not unusual for Harvard students, his comment about the speedy response of the UNIX systems at the Science Center was startling.

"Man," he said. "I logged into fas nearly instantaneously. PINE loaded up right after I typed it on the command line. It's almost the way computers are supposed to work!"

I smiled pensively. It was my friend's final sentence that really got me thinking about the UNIX systems of the Harvard Arts and Sciences Computer Services (HASCS). He had expected a slow response, but was pleasantly surprised when he got a rapid one.

If HASCS Director Frank Steen implements the plan he now has in mind, many students should get rapid responses in the near future. After a disastrous introduction of the new SparcServer machine ("fas") in September, Steen is again considering revamping HASCS's UNIX systems.

If the words "revamping the UNIX systems" cause some uneasiness, relax. Steen appears to have found the solution to the problem that makes people's blood boil.

Right now, the "fas" SparcServer is the center of HASCS's Unix universe. Because fas is the physical location of the system's 14,000 user home directories, all 35 of the UNIX machines which students now use (including the workstations in the terminal room, the course machines and husc7) must be constantly connected to fas.

And therein lies the problem. According to Steen, a six processor SparcServer like fas can juggle about 350 users at a time. Although only about 300 users are logged directly into fas at a given moment, Steen estimates 400 other users are making home directory requests from alternate machines. Most users are all too familiar with how 700 users requesting service can quickly overload fas.

Steen's plan takes the onus of the home directories away from fas and puts it on a dedicated machine called a Network Appliance Box (NAB). The NAB, which Steen plans to purchase pending an initial trial, will handle one of the jobs that fas has now: acting as a server for 14,000 home directories. In fact, the NAB is designed specifically for that task.

The strategy is not without its pitfalls, however. The NAB's efficiency will depend on how fast it can process home directory requests and relay them back to the requester over the network.

Although it will have no effect on the speed of a user's network connection, thousands of requests and responses every second will bog down the local network segment between the UNIX machines and the NAB.

Steen and his technicians have foreseen the problem and are in the process of installing a new kind of network hub--called a switch--which should provide the requisite network bandwith.

While the NAB and the switch may seem to add needless complexity, they really don't. In fact, Steen is actually streamlining the UNIX configuration with these new devices. He is using the age-old computer science concept of abstraction--insulating different parts of the system from each other.

Moreover, Steen knows that people rely on the services he provides and that people do meaningful work on the UNIX systems daily. Any new solution will be more thoroughly tested than any other system has been to date.

If Steen has his way, the slowdowns and reliability problems that students have come to expect--the three minute waits for PINE or the routine crashes before a CS50 problem set is due--could finally be a thing of the past.

Free from the burden of serving as both a login and file server, fas should become a much faster machine. And with all of HASCS's UNIX machines relying on an extemely stable, dedicated file server, service should become much more reliable.

Everything from launching PINE to accessing home pages on the World Wide Web server fas-www should become a more pleasant experience. And if that is the case, students, faculty and administrators will be much happier users.

Matt Howitt '97 (mhowitt@fas), sports editor of The Crimson, co-founded Twisted Pair Consulting, a network and telecommunications consulting operation, in 1989. He has worked for Shiva Corporation, a maker of remote access networking products, as a quality assurance engineer the previous four summers.