previous next Up Title Contents Index

Teradata; Ingres family: Relational Technology, Britton-Lee, Sybase, Microsoft

Jim Gray: So, let's see. Teradata. So here's just some background. There was a guy by the name of Phil Neches at UCLA, and he said, "We ought to do parallelism on commodity hardware." He fell in with some people at a startup, and they started this company, and I guess in about 1984 shipped the first parallel SQL engine. It's very similar to the Tandem story: it's a sort of one-trick pony. It doesn't have referential integrity; it doesn't have all the fanciness. You give it SQL; it gives you back the answers, very fast, and presumably cheaply, because it's running on Intel processors and on commodity disks. Teradata got bought by NCR; NCR got bought by AT&T; and AT&T last I heard ...

Along those lines, there was this whole other development, which was the INGRES project at U.C. Berkeley. The INGRES project had a language called QUEL. They started a company that implemented QUEL. QUEL fought SQL tooth-and-nail, and explained how QUEL was better than SQL in many different ways, and in fact it is better at doing aggregates. There are lots of areas where QUEL is better. Some people at Ingres now feel that the reason that they were less than successful is because they fought SQL rather than embraced it, so this gave Oracle a chance to differentiate themselves. The fact is that ...

Mike Blasgen: Just as a point of time: I had a conversation on the phone with Stonebraker while I was living in Washington, and I left Washington in June of 1983. So it's obviously prior to that. I said, "I think Oracle is going to do well." He said, "Why is that?" I said, "Because they are one of the few who support SQL besides IBM." He said, "Well that status won't last more than a few weeks. Everybody's on that; that's done." So by, I would say, the end of 1982 or the beginning of 1983, they were far over that; they had made that decision. I don't know when they shipped their first code.

Tom Price: Although the first code they shipped was SQL on top of QUEL ...

Mike Blasgen: It was see-QUEL. [laughter] That's right.

Tom Price: Which made me nervous about buying it.

Jim Gray: And so there was that thread. And spun off from the INGRES project was a Britton-Lee group. And the Britton-Lee group included Paula Hawthorn and Bob Epstein and Mike Ubell and probably a lot of other people. And they built a database machine[81]. In that era, there was this whole notion that you could really do much better by building a special-purpose piece of hardware and a special-purpose operating system and then a database system. Build up from the bare metal and it's going to run a lot faster. I think Roger mentioned that that was part of the Esvel concept as well. Louise Madrid was another ...

Roger Bamford: I think we really believed you could get the revenue for it; I don't think we really believed that it was cheaper to build; it was just easier to sell.

Jim Gray: And the performance would be better. I think that was one of the arguments, that you couldn't get good performance on general-purpose; the special-purpose would beat ...

Don Slutz: Lot of special hardware.

Jim Gray: They had an accelerator, which was their hook ... The Britton-Lee guys in turn spun off Sybase, and Sybase came out. The key thing about Sybase as far as I can tell was they ran on UNIX; they didn't use any of the UNIX services except a single process. They used raw disks; they took a single process and multithreaded it, and ran SQL inside of that, or actually ran DB-Library. They were not very SQL enthusiast or compliant. They had the QUEL tradition; they were from INGRES. So the key thing that allowed them to be successful is they had great performance. You would send one request in; it would work all inside of this process; no operating system dispatches, no operating system I/Os, just raw disk I/Os. So they were like a factor of three better than everybody else in terms of performance. They managed to establish themselves as the client/server, open, database thing.

Tom Price: They did a deal with Microsoft.

Jim Gray: And Microsoft took their code and sold it on OS/2. The reason for that was that about 1986, IBM was trying to take over the PC market, and they had their own operating system - OS/2 - they had their own hardware. Microsoft said that they had to somehow protect themselves against something called OS/2 Extended Edition. There was going to be this thing called OS/2, which was basic OS/2, and then Extended Edition, which was going to cost hardly anything more, was going to have a database system in it, and compilers, and query - QBE was going to be built into it, and all sorts of stuff. So Microsoft felt they had to have something like that. So they went to Sybase and said, "We'll get our SQL engine from the Sybase guys, and that will be our Microsoft Extended Edition." And Microsoft remarketed Sybase in the OS/2 world. The relations between Microsoft and Sybase were not warm or cordial. When it came time to port Sybase to NT, Sybase let Microsoft do the job. And then there was a divorce at some point, similar to the IBM divorce about OS/2, that IBM would do OS/2, and Microsoft would go its own way. There was a similar divorce vis-à-vis Microsoft, where Microsoft now owns the Sybase code, so the Microsoft SQL Server now is going its own way, and they've made it more SQL-compliant, and they're adding GUIs to it, and so on. It's now a major force in this whole database world. And the thing that's driving everybody crazy I believe in the database world is, this thing is very cheap. It's, order, five thousand dollars for a server, as opposed to a hundred thousand dollars for a server. This server is capable of doing hundreds of transactions a second. Scary. Pat, did I ... ?

[81] R. Epstein and P. Hawthorn. "Design Decisions for the Intelligent Data Base Machine" Proc. National Computer Conference, Anaheim, California (May 1980).


previous next Up Title Contents Index