This should have been a "reply-all" ---------- Forwarded message ---------- From: Rich <richa03@xxxxxxxxx> Date: Tue, Apr 7, 2009 at 10:25 PM Subject: Re: Mirroring redo log groups or not ? To: "Crisler, Jon" <Jon.Crisler@xxxxxxx> How much control do you have over the application and what it does? Can you "fix" the app to commit less or batch commits? Failing that, I think RAID10 for the redo logs might be a decent fix at the database/storage level, however, be cognizant of recovery in the event of potential issues with non-grouped redo... I would recommend separate redo LUNS - something like redoa1 & redoc2 on LUN1, redoa2 and redob1 on LUN2, redob2 and redoc1 on LUN3 with the letter being the group and and the number being the member. While I understand you may not have the LUNs to do that, seems it might be prudent. On Tue, Apr 7, 2009 at 9:35 PM, Crisler, Jon <Jon.Crisler@xxxxxxx> wrote: > Its one of the top issues measured in AWR / ADDM. The problem > corresponds to poor response time for disk as measured at the host, and at > the SAN level. > > > ------------------------------ > > *From:* Rich [mailto:richa03@xxxxxxxxx] > *Sent:* Wednesday, April 08, 2009 12:33 AM > *To:* cicciuxdba@xxxxxxxxx > *Cc:* mwf@xxxxxxxx; kennaim@xxxxxxxxx; david.barbour1@xxxxxxxxx; Crisler, > Jon; Rajeev Prabhakar; Oracle-L Freelists > *Subject:* Re: Mirroring redo log groups or not ? > > > > Back to the OP's question. > > > I think you mean "log file sync", right? > > This means a session waits for LGWR to write redo to disk after commit or > rollback. > It is usually due to too many commits or short transactions; you should > commit in batch or use faster redo log disk IO. > > However, we need more information. This wait has numerous steps with even > more implications. > > First off, what makes you think this is THE issue? > > What are the following: > "redo write time" statistic > "log file parallel write" waits > > Is the system otherwise busy during high numbers of this event? > > It may not be as simple as just moving the redo logs to faster disks... >