File System Tuning

[Smucker]

I was doing some research on hp-ux tuning. I noticed that all our Service Guard Packages are mounted with log turned on (the default) according to what I am reading should they not be set to delaylog?

I also wanted to ask about "minicache=direct" on the oracle only mount points

Does anyone run the minicache option on mountpoints for oracle? Did you see any performance gains?

[Geoff Wild]

Have a look through:

http://h21007.www2.hp.com/dspp/files/unprotected/devresource/Docs/TechPapers/UXPerfCookBook.pdf

I have heard that mincache=direct should be done for Oracle file systems - as Oracle uses its own buffer cache...

Yes - they should be delaylog in SG - interesting - I'll have to change mine as well...

Rgds...Geoff

[A. Clay Stephenson · Accepted solution]

It greatly depends upon your data and OS. In the past, HP-UX 11.0 and below, I always ran convosync=direct,mincache=direct,nodatainlog,delaylog for Oracle datafiles and indices.

This almost always resulted in better performance (about 18% better in 10.20; about half that in 11.0) -- but that was my data; yours might well be different.

However, at 11.11 and Oracle 8.1 and above, I found that the best performance resulted from fully cooked files for everything. I now run with just nodatainlog,delaylog. Oracle now actually benefits from buffer cache in addition to its own caching in the SGA. Don't go nuts with the buffer cache; on systems with large amount of memory, somewhere between 800-1600MB seems to be the sweet spot. Again, this is with my data and applications; yours may differ. The only way to know is to measure. In any event, it's so easy to change the mount options that measurements are easy. You should divide your Oracle filesystems into at least four groups 1) datafiles 2) indices 3) redo logs 4) archive logs. You can then try the mount options on each group and measure for yourself.

[A. Clay Stephenson]

By the way, this is exactly the reason you have Test environments and Sandboxes. There is really no way to rely upon the advice of however many experts you choose. The performance will vary so much with your application and data that the ONLY way to know is to measure using your data. On the other hand, unless you do some absolutely terrible things with respect to mount options and filesystem layouts, it's really not going to matter all that much in comparison to how well the SQL is written.

[Smucker]

Clay, Thanks for the response. I was merly inquiring if anyone was using these options. I asked the dba's and other sys admins's and they never heard of the minicache option. I to find someone who acually used it and get thier opinions.

I plan on implimenting the delaylog ASAP. I know this shuld help things. I will then test the other options...thanks..

[Devender Khatana]

Hi,

Yes the same in enabled on our server for all file systems having oracle data files,redo logs and archive logs.

Alongwith this convosync=direct and delaylog is also enabled.

HTH,
Devender

[Ted Buis]

JFS Tuning white paper at:
http://docs.hp.com/en/5576/JFS_Tuning.pdf

[Smucker]

I tried the options usggested by the HP Oracle white paper that I found. It did not work very well the convosync=direct,mincache=direct, options slowed down database access by more than 20%.