Re: lies damn lies and benchmarks

"Mark Townsend" <markbtownsend_at_attbi.com> wrote in message news:B90090F3.20F39%markbtownsend_at_attbi.com...
> in article YOzC8.717$eR3.46772_at_news.uswest.net, Pablo Sanchez at
> pablo_at_dev.null wrote on 5/9/02 12:08 PM:
>
> > For example, given a table with 1,000,000 rows, would it be better
to
> > issue a million row level locks or one table lock?
>
> It depends - what does it cost me to take out 1 million row locks ?
If it's
> not much, then I'd rather do that than risk the deadlocks due to
escalation.
>
> And if a different DB implementation only needed to take out 1/3 of
these
> locks, because it only issues write locks, then I'd really have to
start
> thinking that escalating to table level locks where a Bad Thing
(TM).

The cost is minimal because the structures are in-memory. Definitely not as costly has maintaining those locks on disk - as in other implementations. But since the other implementation uses read-consistency, that cost doesn't happen that often. So, it's Not Really That Big of a Deal [tm](*)

(*) Pat pending and not legal in Costa Rica.

--
Pablo Sanchez, High-Performance Database Engineering
mailto:pablo_at_hpdbe.com
http://www.hpdbe.com
Available for short-term and long-term contracts