Re: performance comparision
Date: Thu, 28 Jan 2010 09:22:21 -0800 (PST)
On Jan 28, 5:16 am, UXDBA <unixdb..._at_googlemail.com> wrote:
> q1) what for phase-4 waiting ? [ as it took time 4x longer than
> phase-1 for the same number of
> rows & get per execution are almost double in phase-4 compared to
> phase-1 with poor buffer hit ratio ]
Well, sequential read waits probably means you are using an index to figure out which blocks to update. Of course the buffer hit ratio would be poor, you are jumping all over the place to grab random blocks. How big is the table (size and rows)? How long does it take to do a full scan of the table? Sometimes it is faster to use full scans because multiblock reads can help, but we have no clue here of your total sizes. Hit ratio tuning is silly, anyways, but here it may be a clue you may be doing it wrong. But you may have no choice. Don't be afraid of large undo, in any case.
> q2) is "delayed block cleanout" playing some role here ?
Before and in between phases, select some field that is not indexed to force a full table scan and cleanup of the blocks. If that makes the times quicker and more stable, it was likely delayed block cleanout. If I'm not totally confused.
-- _at_home.com is bogus. "Pick any combination of four numbers plus a percentage point, use it as your story's stardate. For example, 1313.5 is twelve o'clock noon of one day and 1314.5 would be noon the next day. Each percentage point is roughly equivalent to one-tenth of a day. The progression of stardates in your script should remain constant but don't worry about whether or not there is a progression from other scripts. Stardates are a mathematical formula which varies depending on location in the galaxy, velocity of travel, and other factors, can vary widely from episode to episode." - Star Trek writer's bibleReceived on Thu Jan 28 2010 - 11:22:21 CST