previous next Up Title Contents Index

Shoot-out at the OK Corral

Jim Gray: The part I didn't understand is, there was something called QBE that came out?

Mike Blasgen: Oh, that's a whole other story. We have a whole session for query.

Jim Gray: No, it's before this.

Mike Blasgen: OK, let's go to that; that's a good point, you're right.

Jim Gray: There was QBE, and then VS/QUERY and the shoot-out at the OK Corral; all predates this ...

Mike Blasgen: OK, let's do that. Brad, can you do QBE?

Brad Wade: I'm going to have to have quite considerable bit of help on this, because in time-honored tradition of IBM, we had the System R group working on relational database systems at San Jose, and we had Moshé Zloof working on Query by Example[57] - another relational database system - at Yorktown. So a question arose as to why do we have two groups in IBM on opposite coasts who are doing exactly the same thing. And exactly where Moshé and his group came from, how they got started, I know not.

Mike Blasgen: Actually, I know the answer to that, and I'll talk about it only because it's interesting. There was a group in Yorktown that was working on something called Programming by Example.[58] Their idea was that somehow you could do business data processing, like payroll, by giving an example of what you wanted, right, it was sort of like the output back. If you just showed a fancy piece of software what you wanted as the output, it would figure out how to get it.

Jim Gray: The RPG model.

Mike Blasgen: So it was, "Show me what you want as the output, and I'll figure out how to get it to you from the data that I know is available." That never came to anything. Actually, do you know what programming by example is? It really actually happened; it's VisiCalc. That's the closest thing you have to programming by example, and it's very close, it really is what they were thinking about. But they never got that, they never got VisiCalc. But they got something kind of like it, which was Query By Example. So not programming by example, but at least query by example. And the idea was you could draw a picture of the answer, and say, "This is the answer I want" and then it would actually figure out how to get that and give you the answer. So that was Query By Example; Moshé Zloof was the key guy, and Peter de Jong was the manager.

Brad Wade: And so, Query By Example was an early graphical interface, if you will, because it would draw a picture of the table on your screen, and if you wanted "Salary = 10000," you would write "10000" in the Salary column. And I guess if you wanted people with Salary greater than 10000, you'd write ">10000" in the Salary column. In place of SQL's SELECT clause, you'd write a "P." in the columns that you wanted the query to emit. I can vaguely remember playing the Query Game with some of Moshé's stuff, but my reaction is that simple things in Query by Example were simple, and complicated things were impossible; at least they were possible in SQL.

But anyway, in the time-honored tradition of IBM, we had two groups doing the same thing. The performance issue also raised its head. The shoot-out at the OK Corral comes from a direct head-to-head performance comparison.

Jim Gray: Brad, I think that before this, somebody from the field fell in love with QBE, and they were actually shipping it.

Jim Mehl: As an RPQ or something.

Jim Gray: And there were people in the field, and they loved it. They had stories of tape librarians who'd automated their tape library with it, and Gene Trivett was going around and fixing some of the performance problems, and it was popping up all over the planet. So it had a very loyal following. It was obvious to everybody that this did something wonderful. That this was an end-user program. So then the question became, "So why don't we cancel System R?" or "Why don't we grow this thing?" I think Bob [Jolls] must have been confronted with this question a lot.

Bob Jolls: The VS/QUERY group was constantly trying to get the best of both SQL and QBE, and so that came later. But certainly when I managed that group, we had to settle all that. But I never was involved in anything where people looked at QBE as a production replacement for SQL.

Jim Gray: So what was the purpose of the shoot-out?

Pat Selinger: I think it was driven by research management.

Mike Blasgen: I'll tell you what the shoot-out was. It was a very unpleasant and interesting confrontation. Gomory was worried about the fact that he was investing heavily in two projects that seemed to be doing roughly the same thing. One of our arguments was that we had all these smart people like Franco and Jim that had built the RSS, and the RSS was really good stuff, really good code, I mean it's in SQL/DS even today; the same code.

C. Mohan: Around when was this?

Mike Blasgen: I don't have the exact date, but around 1978, right? When did the actual shoot-out occur? 1978? Gomory asked Dick Case to do a review of the work. Dick Case included Ashok Chandra, who currently runs the Computer Science Department - he's the latest version of Frank King - and one other person, who were all disinterested people, but were technically capable. They went to Yorktown and learned all about QBE, and then they came to San Jose to learn all about System R, and I gave them my long lecture about how the lock manager works and how Compare-and-Swap could do locking, and we did it all right, and we knew how to do Compare-and-Swap-Double. Dick Case was really impressed, because he's probably the architect of Compare-and-Swap. Ashok was there, and there was this whole issue of whether you could do QBE on SQL. So that was a different approach, that was, if you will, our approach, which we wanted VS/QUERY to adopt, which was that you wouldn't write QBE on the RSS. That's multiple personalities. We wanted QBE as a graphical interface, emphasizing its strengths - the graphical stuff - and de-emphasizing its weaknesses, which are XRM, no multi-user, poor storage management, and poor performance. Because they were not a compiler, they were an interpreter only. So Irv Traiger did an enormous amount of work, three months work, to show detail-by-detail how you could map QBE to SQL. There was a controversy about whether the language QBE was actually well-specified. There was an ambiguity in it and I have notes from a meeting in which Peter de Jong said, "Well I can't prove you wrong but I'm sure you're wrong. I don't accept your claim that we're ambiguous, even though you've documented it eight ways to hell that it was ambiguous." Anyway, you showed that it could all be done. And so these guys were out there to evaluate it, and one of the issues was performance, because our claim was you would want to do it that way among other reasons because you'd get far better performance. The people back east said, "No, that's not right because it is ad hoc query and interpretation's fine for ad hoc query." They wanted to program directly to the RSS.

Brad Wade: Thank you, Mike. My recollections don't go anywhere near as deep. I do remember it came down to performance. We had System R running on one of the 3270's in the terminal room. We installed QBE on another user-id, and had it loaded up on another terminal there in the terminal room. We primed it; we typed in a SQL query to do something or other; I don't even remember what. We set up the QBE thing to do the same query, and, sophisticated processes of the era, we pressed the ENTER keys at the same time and held our stop-watches up to it to see which would come back first. I even forget whether it was thirty seconds, or a minute and thirty seconds - System R was back with the answer, so we looked over at the other terminal and - it's crashed. System R's star ascended fairly nicely after that.

Mike Blasgen: We let them make up some of the queries, because they had come ...

Bruce Lindsay: You mean they didn't finish any queries?

Mike Blasgen: No, they did. There was a big performance difference. Even though we were compiling ad hoc queries ...

Jim Gray: The most striking performance difference was that Brad could type, and I forget who the person for QBE was, but he typed with two fingers and lots of mistakes. [laughter] So when they'd say, "Next query," Brad would go "Whirrr." [laughter]

Mike Blasgen: Well, anyway, it turned out to be a dramatic performance difference, and you could hear the cheers in your brain, even if you couldn't hear them with your ears. Everybody was very elated, because the performance difference was so dramatic, even on the queries that de Jong had given to Ashok and Case to give, because he was hoping those were our weak underbelly, even on those we were dramatically faster, like a factor of ten I think in some cases.

???: This was on UFI[59]?

Mike Blasgen: Yes, we were using UFI against the QBE front-end.

Jim Gray: What was the OK Corral?

Mike Blasgen: The terminal room, where the RSS was developed.

Brad Wade: After the event, and our visitors had gone, someone - I do not remember who - made up a sign. I think the sign just said, "The OK Corral." And somebody stuck it on the door. There was some movie that I remember seeing when I was ten years old; some famous western gun-slinging shoot-out.

Irv Traiger: Wyatt Earp and Doc Holiday and Hirsh Cohen. [laughter]

Jim Gray: High Noon.

Mike Blasgen: So that's exactly what happened, and it was great, and we called it the OK Corral after that, always the shoot-out. What happened was QBE continued as an IUP[60], but it never really went anywhere; it was never significantly enhanced. Santa Teresa didn't want to take it. DP, which was the sales division that supported the IUPs, continued selling it but didn't invest in it.

Jim Gray: Didn't it get reimplemented as part of DB2?

various: QMF[61].

Mike Blasgen: Sorry. Yes.

Roger Miller: It didn't become a very popular usage even in QMF.

Roger Bamford: We cloned it.

Mike Blasgen: That's interesting.

Paul McJones: It's now called Microsoft Access. [laughter]

Mike Blasgen: Some version of that will probably win.

C. Mohan: But other companies are implementing it, right? Paradox and all sorts of guys are implementing it.

Roger Bamford: I think the original objections are true: anything complicated is impossible. Our customers are using SQL statements that go on for pages. In QBE that would put you against the wall??? ... take all the joins and everything.

Mike Blasgen: So there was a report. What I remember about the report was a ringing endorsement of the RSS.

Jim Gray: The index component in particular. [laughter]

Mike Blasgen: I think they especially liked the locking; I think they loved the use of Compare-And-Swap-Double. [laughter] And then there was some ambiguous recommendation, I don't know what ever happened to it, but as far as I know, Zloof never got the RSS. Peter de Jong went to MIT and is still there.

Brad Wade: And Moshé; where is Moshé these days?

C. Mohan: HP Labs. He started his own company, that didn't fly, then he joined Ashton-Tate; now he's with HP Labs.

Jim Gray: And Peter is at Apollo doing transactions inside of CORBA; HP Apollo.

Mike Blasgen: What happened to Moshé Zloof is he went on to something called Office by Example. In fact that talk that I gave about the SQL/DS; the next speaker was Moshé about Office by Example - OBE. He got a whole group of about twenty-five young, real programmers - not designers - he was real careful about to hire programmers. So he had a bunch of people and he got them to give him his own building - the Bernen House at Yorktown, and he took it over, and he was going to build a PC product. Gomory supported it completely. And then one day, the management changed - I think it had to do with Birnbaum's departure and the reannointment of Herb Schorr, and then maybe Abe Peled on staff. Somehow that set of people ganged up and said, "We're not going to support that at the level it's being supported," which was thirty people times a hundred thousand - three million dollars a year. And Moshé didn't like this, and eventually left IBM. And so he formed his own company as a startup to do it - I guess it didn't work - went to Ashton-Tate; Ashton-Tate was bought by Borland, then he went to HP.

C. Mohan: He's now doing Rendering by Example[62] - easy-to-do programming by example. So maybe he's going back to the System for Business Automation kind of stuff.

Afternoon break

[57] M.M. Zloof. "Query by Example" Proc. NCC 44 (1975) pages 431-438.

[58] M.M. Zloof and S.P. de Jong. "The System for Business Automation (SBA): Programming Language" CACM 20, 6 (June 1977) pages 385-396.

[59] UFI stands for User-Friendly Interface.

[60] IUP stands for Installed User Program.

[61] QMF stands for Query Management Facility.

[62] R. Krishnamurthy and M.M. Zloof. "RBE: Rendering By Example" Proc. Eleventh International Conference on Data Engineering (March 1995) pages 288-297.

previous next Up Title Contents Index