Oracle Logs in VA7400

[Thomas Maddox]

We are running Oracle 8i on some of our database servers. The data resides in a VA7400. We have had performance problems that are indirectly related to when we create several 200MB archived logs in an hour.
Currently we have our Oracle data on 1 controller and the indexes on another controller. The log files are primarily on the controller the Oracle data resides.
Does anyone have any suggestions on arrange the logs in the array to get the best performance?

[Steven E. Protter]

You might want to collect performance data on your system to see where the bottleneck actually is.

I'm attaching a background sar script that you can configure to measure and collect data over time. It is production in my shop.

SEP

[Eugeny Brychkov]

Perfomance problems with VAs are usually seen when VA or SAN are configured improperly, something happened and VA's optimizing (data is in RAID5DP mode and VA should move data from RAID5DP to RAID0/1 to improve perfomance) or there's VA background process running (rebuild).
If configuration and VA data allocation are optimized, then if you have 3-5 LUNs I do not think system will be able to load VA badly. Too small access concurrency. Of course load balancing LUNs between controllers is good idea, but I think it's significant when there're 5+ LUNs are accessed simultaneously.
And in addition, as we're talking about LUN 'balancing' between VA controllers, this balancing should be planned at the LUN creation stage (and even at VA installattion stage). Why? Because for all LUNs created within redundancy group 1 (RG1, odd slots) perfomance path is VA controller 1 (C1), but for LUNs within RG2 (even slots) is C2.
And finally, 'sar' is command to measure disk perfomance from host's 'point of view', I would also use 'armperf' command to measure perfomance from VA's 'point of view'. Output of this command can be easily imported into excel spreadsheet
Eugeny

[Stuart Abramson_2]

Well think about this: What the redo and archive logs are, are copies of the data tables before and after "updates" while the system is running. One should really seperate the data FOUR ways:
1. Data tables.
2. Data indices.
3. Redo logs
4. Archive logs.
Every time you update a row, you write redo log entries. Every so often the redo logs are copied to the archive logs. The redo logs and archive are much smaller than the data files.

So, really you have 4 streams to balance. If you had 4 controllers, that would be good. But if you only have 2 controllers, you will have to balance 4 streams across 2 controllers.

You might want to put the archive logs on the data path, and the redo logs on the index path. Or switch them around.

[Tim Sanko]

Performance is tied to the logwriter. If you need more speed, you could consider kicking off more logwriters.

The separation of indices and datafiles helps, but check ot see how quickly you are writing out log files. If you are transaction based, you may want to kick off multiple logwriters. And as the load reducces let them expire.

Tim

[Tim Sanko]

Thomas,

I want you to know that I have seen several logwriters improve performance by better than 50% in transaction based services.

Tim