How does the Oracle DBA get redo -> archive to alternate archive log filesystems?

[Michael Schulte zur Sur]

Hi Stuart,

what Volker does, is establshing an alternating logging by adding group by group and forcing Oracle to use it by switch log. Since you would want to use at least two members per group for data safety you need four disks for logging. Alternating the archive destination is in my opinion not necessary, since you have only one log to archive and it is sufficient to separate log writing and log archival.

I hope, that makes sense,

Michael

[SteveKirby]

Howdy All,

The two disk rotating Redologs are something I do at two client sites.

I am afraid I was laxy when I set them up and just use Oracle Enterprise Manager.

I can give you details if you need them, but off the top of my head the way I did it (with no down time) is:

Create 4 NEW redo logs. When you create them have set to different locations (e.g. Disk group 5 = Disk array 1, Disk group 6 = disk array 2, Data group 7 = Disk group 1, etc.)

Then force log swithces until up are on Disk group 5. You can then DROP the previous 4 log groups. And .. ta-da ... you now have alternating redo groups.

If you REALLY want to clean it up you can now RECREATE the first 4 log groups on alternatin drives and DROP data groups 5-8.

There is some performance improvement. I am curious how you discovered that you redo log bound, this is not a usual condition unless you have redo logs that are too small or you have an OBSCENE amount of transaction ALL the time. The redo logs are typically switched as needed and are normally a low priority so they are not normally going to hold the system up ... plus you have BUFFERS in RAM to hold stuff...

[Volker Borowski]

Hi,

Michael has got it right, that is the reason for the switch.
Oracle does not care about the LOG# to decide which Group is next.
Been there, done that (and learned it by doing mistakes :-)
- Created group 11 / 13 / 15 / 17 on "A"s
- Created group 12 / 14 / 16 / 18 on "B"s
- Switched all the logfiles to make sure it works.
- Checked alertlog, and found it was switching
"A","A","A","A","B","B","B","B"

Had to do it again :-)
Volker

[Jakes Louw]

An alternative that we have for this would be to assign the filesystems to alternate disks:
-create your HP volume group as normal
-do the lvcreate with a null (ZERO) size
-lvextend the lvol onto the appropriate disk
-create your redo-log space as normal in Oracle DB.

In this way, you can make sure that log_1a and log_2a are on seperate disks.
Repeat for the process for the "B" redo log group.

[Stuart Abramson_2]

We're not going to have multiple archive areas.

We decided that it's not necessary to have more than one Archive Log area. You probably only write to one archive log time at a time, so there is no problem with disk contention. You do want to have multiple redo log disks, and ensure that when you are reading from redo 1 to Archive, that you write redo 2 on a different disk, etc.

[Jean-Luc Oudart]

Hi,

May be the question is related to the performance of dumping the redo onto the archive destination with no or little impact on the rest of the application.
1) How often do you have a redo switch ?
2) how long does it take ?
3) do you use same controler(s)/set of disks

Any clue from the perfstat report ?

Regards,
Jean-Luc