previous next Up Title Contents Index

SQL/DS

Mike Blasgen: OK, now into this picture comes System R, which as you point out is basically irrelevant except for the fact that you thought maybe you'd do an implementation of SQL on this code that you were writing. But something happened, and suddenly it was OK to talk about System R in Santa Teresa.

Franco Putzolu: Bob Jolls should talk about this. I really don't know the background of this transition. The only thing I know is that as far as the high-level components are concerned, there were problems in execution of the plan, but I don't know what prompted the transition to SQL and real DB2.

Bob Jolls: I think the best way to talk about this is to make it personal, because that's really what it is for each of us. We all personally got involved in this in one way or another. My way was that I came out to Santa Teresa from New York in August of 1977. I don't think I was part of any New York Mafia coming out here like some of the rest of you. I came out to be manager of Database Design at Santa Teresa. I had some technical background and some business background and I was quite comfortable being able to do this job. I was actually worried a little more about being in California. You know, all the people that I knew on the east coast had all kinds of things to say about what it was like to live in California, and raise your children in California, and so on. So I came in sort of thinking of California as the land of alternative life styles, and found out that in fact it was the land of alternative data models. [laughter]

I had this group of people working for me, and there was a data model per person. [laughter] So like we had the DBTG folks - that was a small group. We had people that were so enamored with the data dictionary; they kind of thought if you just put all the information in the dictionary, you wouldn't need to bother with those databases anymore, it would just all be there in the dictionary, right? Then we had UDL; Franco mentioned that. That was the idea that relational might be OK as long as you put it in the guise of really long-lasting products like PL/1, COBOL, and FORTRAN, so long as you didn't have to talk about it separately. So I was confused, to say the least, by all these people with all these arguments.

And then I started to learn about System R from some of my people and from coming down and meeting with some of you all. So I started using System R, frankly, as the sort of benchmark. I'd ask, "How does this compare to what they're doing with System R?" and always would learn about all the problems with System R, or all the problems, let's say at the beginning with SQL. Sometimes we're a little too tied up I think in this whole discussion today with what happened to the code. In my mind, what happened to the code is obviously important to those of you who wrote parts of the code, but the real thing that we all did as an entire group of people was to make relational real and to make SQL real, and there can be lots of different implementations of that. I would always hear what all the limitations of SQL were, and problems would be stated that couldn't possibly be solved with SQL. And then if you could write down the question and get a proper answer with SQL syntax and semantics, then here's all the performance reasons why this can never work, and etc., etc., etc. I guess this is, I'd say myself and some of the other people who were involved in this, at least for me, my experience as a programmer helped me here because I kept realizing, "Gee, here's something that is actually operating, that's working every day in different environments. All the people with the concept of how things might work are making up lists of reasons about what's wrong with the thing that is working." And I think this is sometimes the way that we operate in Programming, or was back then.

So I guess the biggest friend that System R or SQL had in this process at Santa Teresa was Eagle. You might think of Eagle as a tremendous resource drain, and so on, and so on, and it was, but the fact that Santa Teresa was completely preoccupied with how to completely replace IMS and do something much better than IMS and do many of the things that were in FS and all this stuff, sort of kept all the guns away from us, while we looked at what would be the best way to do what we then called DS/2. In other words, they sort of carved out everything but MVS and gave it to my group and said, "Well you guys go figure out what to do for database for the VM and DOS environment, and we'll worry about the really big `Production' problems," which was what Eagle was going to address. So I think it was the best friend of the whole process, in that we got to look at that in our own pace without an awful lot of help from White Plains or local management or Poughkeepsie or any of that. We were able to make a decision to go ahead and use System R as the basis for SQL/DS, and without that being a politically-incorrect decision. Which if we had tried to make that decision at that point in time to supplant a lot of the Eagle work with System R, I don't think it would have been possible for people to make that decision.

Mike Blasgen: My recollection is the midrange product that was in the field was DOS DL/1.

Bob Jolls: Yes.

Mike Blasgen: And fortunately for us DOS DL/1 was not considered to be a success. IMS, in contrast, was a big success.

Bob Jolls: Right.

Mike Blasgen: So it's very hard to take on IMS. But it was easy to take on DOS DL/1 because it was not thought of as being a good product. So when you got responsibility for the successor to DOS DL/1, much like Eagle was the successor to IMS, you didn't have as big a hurdle to get over.

Bob Jolls: Right, there wasn't a lot of baggage that came with it.

Jim Gray: I think ADABAS[53] was cleaning IBM's clock at that point.

Bob Jolls: Right.

Jim Gray: So the Europeans said, "We need a low-end database system." And the fascinating thing is that Jim Frame, who I think was the manager of Santa Teresa, did the standard thing: he put low-end database as "out plan." IBM Fall planning was built around the notion of in plan (funded) and out plan (unaffordable). You put your pet projects in plan, and put their pet projects out plan. You know that they'll give you more money to make you do what they want you to do. You put everything they don't want you to do in plan, and the stuff they want you to do out plan, and then you get more money. This was how the funding game was played. Endicott - maybe we're getting ahead of the story, but - Endicott just had its operating system effort canceled. It was going to do a unified VM and DOS. So there were seventy zombies wandering around without anything to do, and they said, "Gosh, there's this thing that Santa Teresa can't do; they're just too busy: this low-end database stuff. Now we don't know anything about low-end databases, but maybe we could learn, and it would be a job." So they bid on this low-end database stuff. Is that a ...?

Bob Jolls: Yes, that's right, and I have a personal reaction to that one, too. It's very painful for me and my little cadre of people, because we had finally gotten on the bandwagon with System R, and we were saying, "Hey, we can make an interesting, successful product out of this." And because of all these management machinations that Jim's talking about, even though we had a lab of 1200 people, and this was - I don't know - a thirty-person effort, or something like this, it was "Out-plan". So, yes, it got all of a sudden migrated to [IBM] Endicott.[54]

That's when we came up with what we called Plan B. [laughter] Since our mission went away, we said, "OK, we don't have that mission anymore. We'll do the database part of Eagle. We came up with Plan B as saying if something perchance should happen to all the rest of Eagle, here's what we would do. Somehow again, talking about politically correct and incorrect, it wasn't OK to say that would happen, but it was OK to say, "Well if it did happen, here's what we'd do," and that's Plan B. So my team and I began working on Plan B, which was to essentially build DB2. There were a lot of debates within the group, and I think John Nauman can talk about this too, about how much code we should import from System R versus what we should develop ourselves, and there were a lot of good arguments on both sides of that, as there always are[55]. The surprise in the reverse direction happened with this one, frankly from my point of view. The surprise of SQL/DS was that it went out from under me, or us, I should say. The surprise of the MVS project was that it happened faster than I thought it would. In other words, Plan A collapsed, all right? Eagle collapsed, and all of a sudden, everyone turned to us and said, "OK, when can you ship this database product?" [laughter] And that's when we had to make some fairly hasty, difficult decisions on ...

Franco Putzolu: When was it realized that Eagle wasn't going anywhere? I mean, is there a date that you can ...

Bob Jolls: When was it realized that? Well, I don't know. I found the whole process very mysterious. I don't know how you - well, I guess I do know how you felt. [laughter] As a former programmer and a manager of development, I sort of knew how to ask questions of groups to find out whether they were on track or not. It was very obvious to me that that group was never on track. We had the same phenomenon at Santa Teresa with Andy [Heller] that you described, in that he was kind of the great guru of this whole project. In this case, it wasn't just some people trying to write down some algorithms; there was like one hundred fifty - two hundred - three hundred people programming based on these meetings that were sort of fly-ins. Andy would fly in; they'd have a meeting about how something was supposed to be designed. He'd leave and the ripple effect - three weeks later somebody's writing a module where there's really no overall design.

C. Mohan: Was he in Watson at that time?

Bob Jolls: He was in Austin a lot.

C. Mohan: No, was he working in Yorktown Heights, or where was he at that time?

Franco Putzolu: He was everywhere.

C. Mohan: Probably had a home location somewhere.

Mike Blasgen: He lived in New York.

Bob Jolls: Yes; he was all over the place. But he was at Santa Teresa I'd say one day a month - maybe two days a month - something like that. There'd be all these big meetings, and all these decisions made ...

Franco Putzolu: Lots of shouting.

Bob Jolls: Yes, lots of shouting, screaming, and all this stuff.

Josephine Cheng: That's how we found out he was there.

Bob Jolls: Right; if you were anywhere near that wing, you could tell he was here. So I think if you asked anybody who was an experienced, professional programmer, they would have told you this thing was going to crater. But I think there were so many layers of management, and frankly there was so much politics around - you know, "What does Poughkeepsie think? What does Harrison think?" and all that stuff, that people just kind of blindly went out on - management people did. I think the one thing that, you know, Mike when he asked me to come and speak at this meeting, I told him and I told Jim, that I think the biggest thing I had going for me in this whole process was that I was politically naive. I wasn't from either the Poughkeepsie establishment or the Harrison establishment, or any of the old Santa Teresa establishment, so I could decide that we should make System R into a product based on what I thought was best for the company, the customers, et cetera, and it was actually pretty easy.

C. Mohan: Where were you before you came to STL? Which organization of IBM were you in?

Bob Jolls: I had worked for five years on a very large internal transaction processing application, as a programmer and manager. And then I'd worked for five years or so in Business Planning in Harrison, so I'd done product forecasting and that kind of thing.

Mike Blasgen: So I'll just tell a personal thing, because this is all very consistent with my view of what happened. My job was to make the sale, close the deal, right. System R, with all its warts, take it, love it all; love me, love my daughter. We met a lot with Bob, personally, not just Bob's group, I mean Bob as a single person. And he looked at System R and I guess he took it away and maybe we installed it, or you installed it; I don't remember exactly what happened. Anyway, it was for this DOS DL/1, this midrange database opportunity that Bob had been given responsibility for. And he came in one day to my office, or called me, or something, and said, "Well, we decided to use System R." I couldn't even think of that; the most I was hoping for was we'd have a meeting to discuss it again. [laughter] We hadn't been rejected. At that time, not being rejected was good; meets minimum. But this was way more than I ever expected, and he was serious. I thought, well, he doesn't even know what he's saying, but he did. He was actually going to take System R and ship it out of Santa Teresa as a midrange database product. I'm sure there were lots of things were going to be redone and modified.

Anyway, then there was this game you talked about - out-plan, and Endicott appeared on the scene. The thing that was so important about that is that Endicott didn't re-decide. They didn't say, "Oh, we have the midrange database mission; now let's go out and do a bunch of research to see what the market requires," because that would have been two more years of no progress. They actually said, "OK, now we've got to ship this code. What do we have to do to make this code ready to go?" That was pretty much the attitude of the Endicott team. Bob's team actually helped in this transition a lot, and we got very heavily involved. We were talking about sending off people to live in Endicott, and that was eventually not considered to be a good idea. [laughter] Not considered salable. But those guys did not re-decide, and that saved two years.

Bob Yost: They basically sent two or three people here to live with us for a period of six weeks or so, and they took it all back.

Mike Blasgen: They took the RSS with almost no change; they took the RDS with almost no change, but they completely rewrote it - they transliterated it from PL/1 to PL/S.

Bob Yost: In fact they were under instructions not to do anything more than that.

Mike Blasgen: When I was living in Washington, I went up to Endicott to help them with some problems, and I remember that they were concerned that the bug curve wasn't dropping at the rate that they wanted it to be, and they were having performance problems and working set problems and all this stuff. It was touch-and-go near the end. And then there was a big announcement, which was at some conference in Atlanta or some place like that, I went to as part of the announcement roll-out. In fact a lot of these slides that you saw earlier were actually a part of a talk that I gave as one of the front-men for the announcement of that. I have a copy of the booklet that Ted Codd wrote, called The Significance of the SQL/DS Announcement[56], which he published in 1981. So I think once that transition to Endicott occurred, and Endicott didn't have any new ideas, it wasn't the data model of the week, the data language of the week, that they were just going to do it, then all that debate ended. And then we were able to take advantage of the situation that Bob just described, about what happened to Eagle, which was the cratering of Eagle. Would you like to be called on?

[53] ADABAS is Software AG's database system.

[54] For details on the evolution of System R to SQL/DS, see:

Donald D. Chamberlin, Arthur M. Gilbert, and Robert A. Yost. "A History of System R and SQL/Data System" Proc. VLDB, Cannes, France (September 1981) pages 456-464.

[55] Bob Jolls notes: "Irv Traiger, since he was on assignment to Mike Saranga, had the opportunity to view these arguments from both the System R and the STL perspective. Perhaps he'll want to add his perspective to this discussion." Irv Traiger responds: "It was a tricky situation. I was obviously a Research Division guy, on assignment at STL for a year (which later turned into two years). So anything I said or did favoring System R, or feeding "intelligence"-type information back to Research, could obviously hurt my usefulness at STL. So I had made a decision early on to spend the year as a Saranga advocate, helping any way I could, and building credibility. There were several things I got involved in, but the biggest one was to help him understand how Eagle was doing, how serious some of the weaknesses were, how or if they could be improved, etc. So I'd work with some of those folks. And at the same time, I had gotten to know Bob quite well. We had a lot of discussions about Eagle, and also about the database work. Saranga made the decision to stop Eagle, which took a lot of courage. Once that decision was made, STL was focused on a couple of tough questions that I was able to help on. One was to help figure out what should be salvaged from all the system services part of Eagle. And of course the other was to get the database work moving quickly. As Bob mentioned, there were good arguments on both sides, about whether to use the RDS from System R, or the so-called LSS that was being designed at STL. But it was pretty clear that the LSS had quite a ways to go, and that they'd have the usual problems along the way when you build any complex system. And the RDS, with all of its warts, at least already existed, and could be improved over time. So I tried to help Bob and Mike from that perspective."

[56] E.F. Codd. "The Significance of the SQL/Data System Announcement" Computerworld (February 16, 1981).


previous next Up Title Contents Index