Was there any noticeable improvement in performance / waits (e.g. significant decline in "log file sync wait"s) after cutting down to 1 member per group ?
If your "log file sync waits" were truly because of redo log write waits you would also have seen "log file parallel write" waits. The former are waits reported by foreground processes while the "log file parallel write" waits are reported by LGWR exclusively.
In some environments "log file sync waits" are not necessarily all because of LGWR write performance. See MetaLink Note#34592.1.
Interleaving your Online Redo Logs between LUNs helps reduce contention between LGWR and ARCH. It may not necessarily help LGWR alone if LGWR is unable to write. Then, again, if 1 LUN is faster than the other, you will see some Redo Log writes faster than others !
You will always get the advice that you have to mirror your redo logs. Few DBAs would want to stick their necks out on this one. I have run production databases with only 1 member per group. After all, if I have only 1 filesystem available, all groups (whether comprising 1 or 2 members each) are on the same LUN !
Hemant K Chitale At 12:16 AM Wednesday, Crisler, Jon wrote:
I am dealing with a possible performance issue with "redo log sync wait". This is for 10g 10.2.0.4 on Linux 64. In troubleshooting this problem, one of the things I did was to cut down the redo logs so that there is only 1 member per group rather than 2 (i.e. redo log mirroring). Each member goes to a different LUN, and the LUN's are Raid-5.Here is the layout- LUN 1 is 1 TB approx, Raid-5 used for database files LUN 2 is 500 gb Raid 5 used for RMAN backups to disk, LUN 3 is 500 gb Raid 5 used for archive log destination All are SAN connected via 2 - 4gb Qlogic HBA's with multipath.Now, currently online redo logs are going to Lun 2 for the first member and Lun 3 for the second. So there is going to be a little bit of contention when archiving, but whichever LUN I point to is going to encountered contention. It would be very difficult to add more LUN's due to SAN restrictions, but I have the option of converting either LUN 2 or 3 to Raid 1 / 0. If I converted to Raid 1/0 I could point one or both members to this LUN. There appears to be some performance difference with our hardware between Raid 1 and Raid 5, but not a great deal in actual testing.So, if the LUN that holds the online redo logs is already protected (raid 1, 5 etc) do I really need to mirror the online redo logs ? Should I convert the LUN to Raid 1 / 0 ? if I convert to Raid 1/ 0 should I keep redo log mirroring and point to the same LUN ?