Re: Transactions: good or bad?

From: Todd Bandrowsky <anakin_at_unitedsoftworks.com>
Date: 27 May 2003 07:22:34 -0700
Message-ID: <af3d9224.0305270622.1e381ca6_at_posting.google.com>


Paul,

Time for me to eat major crow.

While arguing this logical point with you, I've also been working on transactions for my commodity server. It's a nasty problem and no answer is a good one.

It turns out that if I flip to something like your approach, and allow for atomic assignment and some sort of a serialization primitive in the script, then, I can get much better concurrency, superior deadlock avoidance, simpler code, and overall a better result. While my language is not SQL, it is relational in the sense that for the domain space I do implement, I have select, update and project operations although they are not called as such. I expect to have the actual modifications for atomic assignment finished by the end of the week and probably go on sale with it in a month. If you like, I can post a free download for you to look at and a brief paper on the details. Received on Tue May 27 2003 - 16:22:34 CEST

Original text of this message