Thousands Remember Missing Microsoft Database Pioneer Jim Gray

Before Jim Gray addressed the problem of defining a transaction, isolating it from other processes and supplying rules to govern its execution, an online transaction cost a bank $5. After his definition and rules had been widely accepted, an online transaction cost 0.5 cents.

He tended to have that kind of impact on whatever problem he addressed, or at least that seemed to be the consensus of the 1,000-plus people who gathered at Zellerbach Hall at UC Berkeley to honor him Saturday. Gray disappeared on his sailboat, "Tenacious," Jan. 28, 2007, and despite an intensive search, no trace of his boat was ever found. The event at his alma mater was organized by his friends and family to remember his accomplishments.

While frequently described as highly approachable, his "biblical knowledge" of computer science left many of his peers feeling they were dealing with a giant intellect. Gray obtained the first PhD issued in computer science by Berkeley. "He had an enormous intellect," said Ed Lazowska, holder of the Bill and Melinda Gates chair of computer science at the University of Washington.

"There are a lot of smart people at Microsoft. He was the one person who everybody thought was smarter than they were," said David Vaskevitch, senior vice president at Microsoft.

Pat Helland, one of the authors of Transaction Server/COM at Microsoft, told of hearing a Gray technical presentation as a young Silicon Valley worker, then sharing his team's ideas on "shadow page recovery," an approach to rebuilding lost data. In a brief exchange Gray was able to advise him he needed to rethink his plan. He described the experience as a "four to five minutes of being simultaneously crushed and uplifted."

"Jim Gray was one of the few people I found intimidating," said Michael Stonebraker, a leading author of the Ingres and Postgres systems at Berkeley where he formerly taught computer science.

Paula Hawthorn, a now-retired VP of research and development at Informix, remembers a Berkeley symposium where a graduate student delivered his research results, only to have Gray, wearing sandals, stride out of the audience and demand to know who had reviewed his paper.

Gray said the research results had been discovered years before. "He said, 'I don't blame you, young man, but I want to know who reviewed this paper,'" she recalled at the Zellerbach Hall event. She said he glared at the audience, expecting some culprit to emerge. No one raised their hand. Hawthorn asked her companion, who the interloper was. "That's Jim Gray," was all the explanation needed.

It seems obvious now that a transaction is a discrete unit of work, that the data it's using must not change before the transaction is finished, and the whole unit of work must be completed or the must be process thrown away and started over. But at the time he undertook it, the process of conducting hundreds or thousands of transactions a second by computer was bewildering in its complexity. He even defined what data needed to be locked up during a transaction and what chances might be taken with other data at minimum risk to system integrity.

"Jim had the talent of making complex things seem obvious. If we were talking about a complicated issue around a table, he could summarize everything we'd said in three sentences. Afterward, we'd say, 'that's pretty obvious,'" only it hadn't been until Gray said it, said Eric Allman, who worked with Gray on Ingres, now CTO of SendMail, a firm that supplies trusted email systems. He is the author of Sendmail, the open source mail transfer agent that still delivers 65% of email on the Internet.

After Gray defined transactions at part of the System R relational database team at IBM, he moved to Tandem Computers where he helped implement NonStop SQL, the Tandem fault tolerant operating system. His work there demonstrated how transactions could both be implemented correctly and be executed with blinding speed by a system that didn't fail.

Tandem had been trying to hire him for a long time when he called Jerry Held, CTO, to say he'd take the job. Held asked him did he want to know about his salary and Gray said, "No, I'll find out next week when I come in." While working on NonStop SQL, he wrote a paper, "Why Do Computers Stop And What Can We Do About It," a title that showed his willingness to test the limits of managerial patience.

He received the Turing award in 1998 from the Association For Computing Machinery, the computer industry's equivalent of the Nobel prize with a similar cash award, for his work on transactions.

Gray was working as lead researcher in the Microsoft Bay Area Research Lab at the time he disappeared.

"Jim leaves an astonishing wake behind him and I miss him," concluded Helland.