Re: computational model of transactions

paul c wrote:
> ...
> You may be tired of me harping on about this, but locks (logical ones,
> not physical ones) in the first place were merely an implementation
> device to give the effect of the single-threaded user work. In the
> context of many tables with many columns they can end up being quite a
> crude, heavy-handed effect of that.
> ...

Also, in the interest of preserving Marshall's sanity, I must mention that somewhere I think Gray mentioned that lock managers can be a bottomless pit. I once hired an assistant professor to implement more thorough predicate locks. Before he finished he had a nervous breakdown. The last time I saw him the secret police were tailing him.   Then I got ordered to hire a professor. The walk-throughs were okay but he prompted only code that wouldn't compile. I thought it best to get out with my own sanity. After that the whole thing was handed over to a system programmer who knew nothing about db. He ignored predicates but being an expert in system resource monitoring for accounting purposes and having heard that lock managers could be performance hotspots, he added system calls to record things like cpu utilization which I thought was a self-fulfilling prophecy and proved Gray's point.

p Received on Tue Aug 01 2006 - 21:45:58 CEST