Computer science research can easily become "irrelevant" without connections to the world of industry, database developer Mike Carey told a Stanford audience. "It's really important, I think, to be plugged into current technology, what's going on the world—what are the costs of things? What are the current issues in the industry?...It's really easy to do sort of irrelevant research if you don't understand how the system is structured." Carey saw many papers, he said, that were based on ten-year-old assumptions about costs and presumed access to information that would actually be impossible to know.
After finishing his PhD at Berkeley, Carey served for twelve years on the faculty of the UW-Madison Computer Sciences Department. "Some of the problems we were stating and solving weren't connected well enough to problems that people actually had," Carey said. "I was sort of tired of making up my own problems...I wanted access to real problems and systems." So Carey went to IBM Almaden, working for five years adding object relational extensions to SQL and DB2 Universal Database—what he called "development in the safety of a research lab."
The world of testing, code reviews, documentation, system catalogs, and migration paths taught Carey first and foremost that "Products are hard to build": "Nobody that I know of understands the whole thing," he said. "The other thing that I discovered, much to my disappointment, is that you have to do the other eighty or ninety percent of the work. So in addition to thinking about how cool it would be if we had authorization work this way...there's also a lot of work that is actually sort of mundane and can cause you a lot of pain."
Carey's corollary: "Adding to a language is hard." In SQL, he said, the original authors made a number of assumptions that later became hindrances. For instance, subqueries were designed to find scalar values. When Carey's team was attempting to add collection types, they found that the representation they wanted was already taken. And the language can't be changed—only added to—because any change could break all the existing SQL databases. This pragmatic fact, Carey said, was also lost on a number of academic researchers. When the Almaden designers looked through papers for inspiration, they found a lot of "very nice work" that required a change in the language. So "all those papers were irrelevant."
Last year, Carey completed his transition from research to industry by joining the startup Propel, where he found that commercialization of technology imposes its own roadblocks to development. "There's definitely a tension between standards and innovation," he said. "What customers want are standards-based solutions." But Carey finds that standards are adhered to blindly, perhaps at the expense of quality. "Three and four-letter words strung together are really important when you talk to customers...It's not about the technology, it's really about the standards. They didn't care what it was: it could have been, I don't know, COBOL...they didn't care what the standard was or what it could do." To a researcher, databases are just databases. In the industry world, "Oracle and Sun and Solaris are the answer; what was the question again?...If you've got some new paradigm for building the top layer of a web system, there's going to be some real reverse traction about that."
While Carey encourages academics to be mindful of real-world constraints, he still considers their most valuable "product" to be students. "Probably the most important thing you're doing...is you're basically training students to think," he said. "Well-trained students are really valuable."
