intermittent very high waits in LGWR on Linux?

From: bugbear <bugbear_at_trim_papermule.co.uk_trim>
Date: Thu, 30 Jun 2005 11:49:06 +0100
Message-ID: <42c3ce22$0$41918$ed2619ec@ptn-nntp-reader03.plus.net>

Hello again;
thank to all the great advice I got in my initial request to the group, I am now better informed.

It appears that I am suffering from
some kind of LGWR process scheduling issue, possibly OS related.

Running TKPROF on a level 12 10046 event trace of a test of 5000 inserts
reveals 346 seconds waiting on
"log file sync", which googling
reveals to be the redo log.

Looking at the trc file in detail,
reveals that most (93%) of these
waits are under 10 millisecs, which is OK, if not great.

(as discussed in the other thread,
my application is still pretty
dumb in term of efficient use of Oracle)

But 7% of the waits (332 of them) are
almost exactly 1 second each.

It's almost as if these syncs are being
done on a clock, as opposed to "on demand".

Bizarrely, if a attempt to diagnose the LGWR process by attaching an strace(1) to it, it goes around 5x faster!!!

Has anyone seen anything like this?

BugBear Received on Thu Jun 30 2005 - 05:49:06 CDT