system is slow

[Rushank]

ENV: N-class (rp7400), HPUX11.00, Oracle 8.1.7. 8GB RAM

I down loaded all the recomended patches from CPM (Custom Patch Manager) and installed them all without any problem , and then I changed the mount option in fstab with vxfs rw,nosuid,delaylog,mincache=direct,convosync=direct,nodatainlog 0 5 , option.
Since then system is running very slow and load on system shows more then 2.50 , previously it never reached 1.00 UNIX95=ps shows all the oracle process are consuming system resources
Is the new mount option giving problem or the patches ??

Thanks.

[Ted Ellis_2]

most likely it is your mount options.. do you fully understand all of these? The rw, nosuid, delaylog and datainlog are pretty standard, but mincache and convosync are not. I take it these are Oracle recommendations? You may be missing something in bewteen to make these work properly.... the convosync especially

check man pages on mount_vxfs

[A. Clay Stephenson]

Firstly, the =direct mount options should only be used on the datafiles and indices. All other files (including redo and archive logs, ORACLE_HOME, executables) should use the conventional mount options.

It should be very easy to determine if the mount options are your problem or the patches. Remount the filesystems with the old options and if your performance problems persist then the patches are the culprit, - iff these are all the changes that you have made.

Non-Oracle operations on the =direct filesystems (e.g. cp, backup, ftp) will be slow; the idea is that you have bypassed the UNIX buffer cache so that Oracle is doing the caching in the SGA. You can then decrease buffer cache and increase SGA size.

Under 11.11, Oracle usually performs best using cooked files.

[BFA6]

Hi,

I would suggest backing out the mount options, and run the system with just the patches as the change & see how it goes.

Regards,

Hilary

[Ted Ellis_2]

from the man page of vxfsio:

VX_DIRECT
Indicates that data associated with read and write operations is to be transferred directly to or from the user supplied buffer, without being cached. When this options is enabled, all I/O operations must begin on block boundaries and must be a multiple of the block size in length. The buffer supplied with the I/O operations must be aligned to a page boundary.

Ted

[Darren Prior]

Hi,

This is one of those situations where in hindsight it would have been better to have made just one change, then tested the system before making the next change.

I reckon that putting your mount options back to their previous settings and then testing is the easiest way forward for you.

regards,

Darren.

[Rushank]

Well, I used the new mount options only for datafiles , and used conventional mount options for redo, oracle_home executables and archive logs.

My backup is also taking considerably long time now.

I've some other file system ( different database instance) is using conventional way mountpoint. Those are also running damn slow. Tonight I'm planing unmount and remount every thing as it was before and see if it makes any difference. But If the patches are culprit then how do I know which patch is doing difference? There are more than 200 pathces were installed on that day.

[Wodisch]

Hi Rushank,

since you seem to have the OnlineJFS installed, you could use "mount -o remount,..." to switch to the original values for "convosync" and "mincache" - no need to wait for an Oracle shutdown...

Just my $0.02,
Wodisch

[Rushank]

Woodich, I didnot use mincache and convosync in my original fstab. It was just vxfs delaylog 0 4.
You mean to say I can use mount -o option to remount these filesystem without shutdown database.?

[Sandip Ghosh]

It may happen that after changing the mount point the total load is going to the Oracle SGA and It is not big enough to handle the load. So, it is better if you once consult with the DBA about it.

Sandip

[Rushank]

Any other clue from anybody else..?
I'm worried if this is problem of patches I installed then I have swremove more than 200 patches.

[Rushank]

Here is the update.
Well,I mounted the filesystems as it was before. No, convosync or mincache etc.
System is performing much better now. But my whole purpose of mounting file system in a new better way is failed completly.
I was under impression mounting data base file system in this new method would increase performance level , but It din't.
Any idea why??

[A. Clay Stephenson]

Because you only did half the job; the idea is that you take the memory that was being used as UNIX buffer cache and use it to increase Oracle cache inside the SGA.

To fully implement the new mount options, you need to add the options to only the datafiles and indices and then reduce your system buffer cache size. You have, in effect, converted your datafiles and indices access to raw i/o. Next, you increase the SGA.

On most 10.20 and 11.0 Oracle implementations, this combination results in higher throughput but in 11.11 I have measured better performance using the conventional mount options for everything.