Under Pressure, University Upgrades E-Mail Capabilities

This fall, the Faculty of Arts and Sciences' (FAS) 15,000 e-mail users dealt with multiple system crashes and daily system slowdowns, and Dean of the FAS Jeremy R. Knowles received a lesson in e-mail's importance from his students, faculty and staff.

One crash kept the system down three days. Another wiped out the e-mail of 3,000 people, some of which was permanently lost. And almost daily, logging on to the system at 5 p.m.--a process that should take 30 seconds--would regularly take 30 minutes.

The situation became so dire that Professor of Classics and History Christopher P. Jones felt obliged to interrupt the normal repose of a November Faculty meeting to protest the problems.

"Many of us who use e-mail regularly...know that the situation is not well. I would like to know...what amelioration we can hope for?" he said to Knowles.

Knowles annual budget letter to the Faculty reflected e-mail's increased importance. "Until recently, e-mail and the World Wide Web were viewed as convenient tools for many faculty members and students. Now, we all look upon them as necessities," he wrote.

Though Harvard Arts and Sciences Computer Services (HASCS), responsible for the upkeep of the network, had upgraded Harvard's e-mail system in August to cope with the increased usage, HASCS undertook a complete redesign of the system in January, taking until March to implement all the changes.

Problems

Harvard's e-mail problems rested at the intersection of two separate causes. The first problem was technical: Harvard's e-mail system was not designed to handle the high usage it constantly received. The other was that turnover in HASCS staff had prevented them from crafting and implementing a long-term solution to the problems.

In 1994, Harvard's e-mail system was processing about 40,000 to 50,000 messages a day. This fall, the system was processing closer to 300,000 messages a day and at peak times would have more than 1,000 users logged on at once.

"We're swimming upstream with a torrent coming down," said Dean of the Division of Engineering and Applied Sciences Paul C. Martin '51, chair of the FAS Committee on Information Technology.

Though the new machines Harvard installed in September to handle the problem had the processing power to handle the University's load and much more, technical problems resulted in lengthy login times, crashes and general instability in the system.

Staff turnover among HASCS's UNIX system managers, the people charged with managing and upgrading the network, compounded the technical problems.

"Turnover makes anything difficult," said Coordinator of Residential Support Rick Osterberg '96. "You lose your internal consistency when you bring new people in and they have to catch-up and we have to get used to their needs and our environment."

Since 1994, six people have left HASCS's UNIX group, although one later returned. This turnover prevented HASCS from implementing system architecture changes it had been planning since last spring, according to Director of FAS Computer Services Franklin M. Steen.

Restructuring

Before being reengineered this winter, Harvard's e-mail system relied mainly upon two powerful computers, one for logging in and the other for distributing e-mail.

With the new system, four machines will handle logins and three machines will distribute the mail. The new system is faster and will provide "redundancy," meaning one machine's crashing will not bring down the entire system.

"When designing a central system, you effectively have two options: one big box or a bunch of little boxes," Osterberg said. "We had used the big box option, but we hit the point where one big box just could not keep up."

Now that HASCS has solved its immediate problems, its staff plans to work on a series of projects which had previously gone undone because of constant e-mail problems. These projects include improving network security and creating new network tools so the College can provide services such as students' grades on-line.