Operating System - HP-UX
1835143 Members
2805 Online
110077 Solutions
New Discussion

VxFS Performance Question

 
Bryan Lenihan_1
New Member

VxFS Performance Question

Hello!!

I am at a customer site they are currently running VxFS on a 0+1 software strip. We are getting a very heavy lag time from the time Oracle sends the write request to the time the disks are acutally written to.

The disks are hitting about 15-30%, while the sys on the CPU is increase between 30-50%. I am thinking the request is getting lost in the buffer cache or the RAID management.

I am looking at setting the mincache and convosync to direct according to the man pages and some documentation I have found.

The SA at the customer site said that this was not available for the version on VxFS we have. What version do we need to be on, and why are they listed in the man pages?

They are running V10.20 on a T600.

Thanks,
Bryan Lenihan
Oracle Certified Professional
5 REPLIES 5
Patrick Wallek
Honored Contributor

Re: VxFS Performance Question

I believe the mincache=direct and convosync=direct are available only with Advanced / Online JFS. If they have that, then you should be able to use those options.
James R. Ferguson
Acclaimed Contributor

Re: VxFS Performance Question

Hi:

The options you want require Online JFS. Sorry.

...JRF...
A. Clay Stephenson
Acclaimed Contributor

Re: VxFS Performance Question

Hi:
You need B3929BA OnlineJFS (or Advanced VxFS).
I would also add the nodatainlog mount option.

One way that you could test if that is the problem is to go to raw i/o. It's really not as bad as it sounds. All you really need to do
is set up a logical volume that is as big (plus a bit more) as you current data file.
You then copy via dd from the current data file to the rlvol partition. Lastly you symbolically link the data file to the rlvol partition. All of this done with the database down, of course.

e.g. ln -s /dev/vg05/rlvol1 /u01/data/oradata/data01.dbf

You could do this for 1 or two of the 'hot' data files to see if the situation improves. If so, order and install OnlineJFS and you get almost all the benefits of raw i/o with the convenience of cooked files. To take full advantage of the raw/Online JFS i/o you should reduce the unix buffers and increase the SGA buffering.

Food for thought, Clay
If it ain't broke, I can fix that.
Mark Mitchell
Trusted Contributor

Re: VxFS Performance Question

Careful with these Oracle practises with Raw devices and Async I/O. They trade redundency for speed. There are some good posts here about that. I found a JSF product from Veritas calles Database Edition for Oracle on HP-UX. It seems to submit file systems as raw devices, you get the performance and get to keep the advantages of having an Online JFS file system. Check it out.
A. Clay Stephenson
Acclaimed Contributor

Re: VxFS Performance Question

Hi,

Actually, raw i/o and asynch i/o are two very different issues. Raw i/o does not compromise integrity in any way and it actually enhances it because a write to raw disk goes to the disk before returning to the process while a write to a 'cooked' file merely goes to the unix buffers before returning. The major reason to use raw i/o is speed but managing and monitoring raw i/o is a pain. The best of both worlds is OnlineJFS.

Clay
If it ain't broke, I can fix that.