Oracle and cache

[William Pribble]

I have a mount point and I have it set up so it won't use HP cache, but instead use Oracle's.

From /etc/fstab:
/dev/vg10/lvol1 /db01_new vxfs mincache=direct,convosync=direct,nodatainlog 0 2

However, when I check it in sam, under filesystems, and vxfs options the following is checked

Full Integrity for All Metadata and User Data (Lowest Performance)

Is this right?

Thanks,
Jeff

[James R. Ferguson · Accepted solution]

Hi Jeff:

Yes, from the standpoint that fetching from a cache (memory) is faster than actually doing a phyiscal I/O. Here, you have bypassed the in-memory, Unix buffer cache.

The buy-back here is that Oracle is managing the buffers for its filesystems; not *both* Oracle and the Unix syncer daemon (which is responsible for buffer cache flushes).

Regards!

...JRF...

[William Pribble]

James,

Let me see if I understand this correctly?

Because I am not using HP's cache, the server thinks that it will be doing physical IO, even though it is really using Oracle cache.

And that is why I am seeing the option, with the lowest perfromance selected in SAM?

Thanks,
Jeff

[James R. Ferguson]

Hi Jeff:

Yes, that is correct. The label "lowest" is of course, in the abstract. This depends on how well-tuned Oracle is; the type of I/O you do (large sequential vs. small random); whether or not you doing multiple logical I/O's to the same physical block; etc.

Again, by bypassing the Unix buffer cache, you save memory (since you don't keep as many file buffers around). This, in turn, means that the 'syncer' daemon has less work to do, which means then that its I/O is reduced.

All this to say, "your milage may vary".

Regards!

...JRF...

[Steven E. Protter]

You have told teh OS to do nothing with regards to buffer cache and therefore it thinks(for lack of a better term) that performance will be degraded.

What you actually have done is stopped double buffering the oracle data filesystem and let oracle handle the cache. Oracle is probably better at managing performance of its files than the OS is.

I would say the risk of data loss goes down because its less likely write data will be in the OS buffer cache waiting for write when its actually not been written to disk.

Whether this is a factor depends on how busy the filesystem is from the i/o standpoint.

If its a busy fs, then you probably have improved data integrity and the ability to recover from sudden system faults.

Both your question and susbsequent post seem in essence correct to me.

SEP

[William Pribble]

Thanks for the answers.