Re: Poll: How fast is YOUR database in rows per second, on a full bore table scan???

From: Frank van Bortel <frank.van.bortel_at_gmail.com>
Date: Tue, 15 Jan 2008 14:45:12 +0100
Message-ID: <8c6c$478cb8e9$524b5c40$6681@cache4.tilbu1.nb.home.nl>


Some observations:
- same machine, using RH ES 5 (CentOS 5.1+updates), with LVM/stripe set
  does not make a difference in timing (differences are   too small to measure...)
- playing around with the LIMIT clause actually *does* make a
  difference;
  from 1000 to 500: none
  from 1000 to 100: -10% (.30 secs to .33/.34)   from 1000 to 10000: +10% (.30 secs to .27/.26 sec)   from 1000 to 100000: -10% (.30 sec to .32/.33)

So it may pay off to test with different LIMIT values, as far as memory consumption allows. Thomas Kyte mentions 500 as a reasonable value, Daniel uses 1000, on my configuration, the optimum is even higher, but I do not have to consider other processes here.

High volume ETL processing may well be worth checking for the extra percent.

-- 

Regards,
Frank van Bortel

Top-posting in UseNet newsgroups is one way to shut me up
Received on Tue Jan 15 2008 - 07:45:12 CST

Original text of this message