From:  Josh Hewitt

To: Editor

Date: 17 Apr 2005

 

Just to get the facts about the TRM discussion right:

 

1.      The TRM was being advertised by you (and C.J. Date) as a revolutionary technique for implementing relational databases. If I believe the claims then it must be the Holy Grail of relational technology.

 

2.      So I decided to analyze it.

 

3.      It turned out to be useless in its current, published form, to implement disk-based databases.

 

4.      Then you, and the TRM guys said, "Yeah, of course.  Because it is for in memory/read-only usage only."

 

5.      After a couple of months one of the guys behind the TRM came forward and wrote an article that claimed that there was a disk-based version of the TRM as well, but, unfortunately, he cannot say anything about it, because of some ambiguous legal situation.  Practically that's all he said.

 

So to summarize:

 

·   The published form of TRM, the one that can be objectively analyzed, is for "in memory/read-only" usage.

 

·   The disk-based form cannot be analyzed.  It might or might not have the claimed superlative performance characteristics.

 

I couldn't, even if I wanted to, analyze the disk-based one, because there's no information available on it.

 

I must admit that I did overlook the passage in C.J. Date's book where he mentioned that the published form of TRM was for "in memory/read-only" situations only.  At least now we know why :)

 

I'm still waiting for the disk-based TRM.

 

 

Fabian Pascal Responds: You have a knack for either not paying good attention to what we say, or misinterpret it, or you fail to understand it.

 

1.      It is simply a theoretically sound model that when used to implement the relational model, yields far better performance, that's all. Holy grail is your interpretation (an influence of the industry, I would guess). There is no reason to believe anything without evidence. Unfortunately that is delayed by factors that have absolutely nothing to do with the technology itself, and over which we do not have control.

 

2.      You did not analyze it. You analyzed a limited aspect of it, which was explicitly used for expository purposes, so as to not divulge valuable trade secrets. I advised against publishing anything based on that, but you did not listen. You were more intent on “making a splash” than the truth.

 

3.      No, because Date explicitly warned not to draw conclusions about the disk-based algorithms from the memory-based ones. Even though we don’t know how you could have missed that, you now claim you did. Yet you still refuse to admit that drawing conclusions about disk-based aspects from the memory-based aspects was wrong.

 

4.      We did not say it is for "in memory/read only usage only". We explicitly said that the disk-based algorithms are different, but cannot be disclosed.

 

5.      He's not "behind the TRM", just privy to it. And yes, there are legal constraints, which are perfectly justifiable, whether you like them or not. We don't like them either, but we understand them. It's someone's asset and livelihood. Perhaps you should investigate who benefited financially from the relational model. It wasn’t Codd, for whom the industry did zilch.

 

So to summarize:

 

·   You should have (a) stated upfront that the analysis was of the in-memory algorithms only, and (b) waited for me to get back to you and inform you that the disk-based ones are different. You didn’t and ended up with the wrong end of the stick.

 

·   Correct. But you nevertheless drew your conclusions about it even though you did not know it (this is like looking for something you lost where there is light, not in the dark place you lost it in). And you either ignored the fact that both Date and I said there is a prototype implementation, or you are implying we were not telling the truth.

 

No, you still don't get the why. There is a different disk-based solution, which cannot be disclosed for the time being. And because it cannot be, you are questioning its existence. That is what you should have said then, but that’s not what you did say.

 

Let me close with as strong a statement as I can muster: the fact that you want to analyze an invention does not mean that the inventor is obligated to disclose it, and if he does not, it does not mean there is no invention. By all means, remain skeptical until he chooses to reveal it, and you are free to state your skepticism. but drawing conclusions about something you do not know does not advance knowledge (incidentally, this is what happens when you insist on making inferences from the unknown, something which is missing (pun intended) from your understanding of the missing data problem, as exhibited in your comments on it).

 

Caveat emptor.

 

A Note on the TransRelational™ Model

Defending the TransRelational™ Model

 

 

Posted 6/10/05