RE: log file sync waits / slow response time at log switches
Date: Fri, 15 Feb 2008 14:07:57 -0500
After you figure out the best way to handle the dataguard sychronization issues (several other comments in this thread seem productive, and I won't make a comment on that unless I deeply understand your remote site business continuation requirements and I've never seen that be a pro bono sized task), you probably also want to add redo log groups. Even if you completely remove the dataguard issues, you cannot start overwriting an online log that has been used before it has been archived (sorry about stating the obvious). But if you have plenty of online redo log groups to switch into to handle peaked insert/update/delete load you can defer jambing up wrapping around for a time equivalent to the extra space you invest. This of course dies under its own weight if your steady state rate of updates (insert/update/deletes) exceeds your steady state rate of archiving to your slowest destination, or if your burst rate of updates exceeds your slowest archiving rate for long enough. Fortunately Oracle doesn't die, it just waits.
It appears you only have 3 groups. I suggest an even multiple of the number of muliplexed independent stripe sets that make up your online redo log volumes. If 12 is a big enough number to handle the peak load in excess of archiving rate, that nicely handles 2, 3, and 4.
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org]
On Behalf Of Gerald Cunningham
Sent: Thursday, February 14, 2008 2:22 PM To: oracle-l_at_freelists.org
Subject: log file sync waits / slow response time at log switches
My 10046 tracing is showing a lot of "log file sync" waits. Also, users log files (processes are doing DML) are showing
spikes in response time that match the times of the log switches.
This is a physical Data Guard implementation with 1 primary / 1 standby. Oracle 188.8.131.52 64-bit on Solaris.
From the alert log, it appears that Oracle is waiting until the redo information has been written to the standby site
before it switches the primary's redo logs.
<snip>Received on Fri Feb 15 2008 - 13:07:57 CST