The Model Is ChangeIn a conversation with Microsoft's Jim Gray, it's clear that exciting times are ahead for database developersJim Gray is a legendary figure. In 1998, the Association for Computing Machinery honored him with the A.M. Turing Award for his "seminal contributions to database and transaction processing research and technical leadership in system implementation from research prototypes to commercial products." That's a pretty good generalized description of a career that has spanned the earliest days of database technology and relational systems at IBM, through the development of OLTP and Tandem Computers' NonStopSQL, to his current work as senior researcher and distinguished engineer with Microsoft's Scalable Servers Research Group. Gray is manager of Microsoft's Bay Area Research Center, which is located in San Francisco. Jim had planned to join Intelligent Enterprise's roundtable discussion featured in this issue (see "What's Next for the Database?"). However, due to unforeseen circumstances, we had to change plans. Instead, a few days later, I had the pleasure of talking with Jim in his San Francisco office. I also got a chance to see firsthand one of Jim's more amazing projects: SkyServer, which provides access over the Internet to the Sloan Digital Sky Survey. SkyServer is applying the new power of commodity platforms, database software, and data mining algorithms to tasks that would otherwise have taken astronomers many, many years to synthesize into information. Ultimately, Gray sees the work providing astronomers, educators, and the general public with online access and search capabilities for a virtual observatory: a "union of all the telescopes in the world." Mixing spatial, temporal, and multispectral aspects of massive quantities of data coming from a range of telescopes presents a "wonderful database challenge," Gray said. A discussion with Jim Gray always startles my understanding of what's become possible with low-cost, commodity platforms. This time, exhibit A was a March 12 Transaction Processing Performance Council (TPC)-C benchmark, which reported the total cost of a Microsoft SQL Server 2000 system running in a client/server configuration with a Dell Computer Corp. Dell PowerEdge 2500 as the server. Capable of serving 9,480 users with strong throughput and massive data memory and storage available, the system cost $42,451, according to the TPC benchmark. Clearly, the scalability picture is changing fast. Still, Gray was understated: "Tandem and Teradata offered a nice scale-out story, which I thought we would have been able to commoditize years ago. In fairness, things are a little better now." As our database roundtable members noted, the bigger issue is manageability: "Users won't scale up to 100 nodes if they can't manage it. If you want to sell a lot of licenses and nodes, you want to make this work," Gray said. Like IBM, Oracle, and other major software vendors, Microsoft is engaged in intense research to enable self-managing systems that can learn and adjust without human intervention. Ultimately, this will become a decisive issue in IT spending as organizations confront the challenges of deploying distributed, commoditized systems. Web services and hosting providers will also come to depend on intelligence living deeper inside software and hardware, giving systems the ability to optimize performance for unexpected conditions. Web Services Meet DatabasesWhat will Web services really mean to database technology? How will decades of careful work by database professionals fuse elegantly with the frontier world of distributed components and object-oriented applications? "Many of us have been chasing the object model for a long time," Gray said. "Depending on how old you are, we could trace it back to the late 1960s, with things like Simula67. More recently, we've had the CORBA model, the ODMG [Object Database Management Group] model, and the things that Microsoft has pursued, such as OLE, ActiveX, and so on. All of these have been attempts to serve objects rather than bytes." Benefiting from this long view, Gray offered some insightful comments about how database technology and Web services come together, which I will quote at length here. "Klaus Wirth once wrote, 'programs are algorithms plus data.' The whole goal of XML is to give people more of the semantics associated with the data — but we're still just talking about data. Web services allow you to have programs, or algorithms, plus data. They allow you to publish a set of methods; and when you invoke a method, get back 'the answer.' XML contributes by being fundamentally a good data representation language. "We're starting to see some interesting extensions. One that I'm excited about is the notion of a data set, which would be familiar to anyone who knows IBM's IMS or any hierarchical or network model database. With a data (or record) set notion, you can have a set of relations; relations can have foreign keys, so, of course, you can do good old relational operations, if you want to. A fundamental aspect of IMS is that you get back a tree of records. One of the things we learned with IMS hierarchical or network databases is that you want one message in and one message out. Well, the XML data set notion lets you bring back all the things you'd want from a database packaged in a data set. You get back the entire answer, which you can present in a pretty interactive way. You can let people update the data; the updates are packaged and sent back to the database, where the old, or base values, are transformed into the new values. It's degree-one consistency, or what we might call optimistic concurrency control. "The relational camp resisted the object database challenge by extending and building object/relational technology. But the relational guys never really had a very strong object model. Now, Web services and XML give us an object model with a set of methods, a way of invoking methods, and an extension mechanism. All the database vendors are busy trying to create a world where database systems are 'consumers' of Web services. On the producer side, existing stored procedures can be published as Web services; in the most recent version of [Microsoft's] XML stuff, you can just push a button on any stored procedure and make it a Web service. Databases have been 'data joiners': well, on the consumer side, databases can just 'join' anybody who is willing to produce data sets." Hold On to Your Hats"These are exciting times: Once, there was an inside and an outside to the database system. The database guys still know where the outside is, but there's no 'inside' anymore. The system has some local data, but it's being called upon more and more to pull data from other places, including email systems, text archives, and other stuff. The database model is changing fast." With the long evolution of database systems and its battles between different models in mind, Gray offered a surprising conclusion: "You know, I think the war is over, and the network guys won. If you're a relational guy, you can pretend that these data sets I've been speaking of are a relational thing, with foreign keys and so on. But the Internet and the tree-style document object model have simply tuned people to following pointers and working with parent/child relationships. I keep talking to people who say, 'boy, X-path is wonderful, I can just wander up and down the tree.' People from the old school might say: 'Where's the control, where's the global schema? What's going to happen when the designer leaves?' Well, the flip side is that it's getting so easy to build applications that if the application isn't satisfactory, you just rewrite it. Astonishing, but true: We can now do things in weeks that would have taken us years to do in the past." David Stodder [dstodder@cmp.com] is editorial director of Intelligent Enterprise.
|
 |
Most Popular This Week
IE Weekly Newsletter
Subscribe to the newsletter
| |
|
|