1748243 Members
4301 Online
108760 Solutions
New Discussion юеВ

Re: Oracle Mount options

 
Jim Krol
Advisor

Oracle Mount options

I recently modified the mount options on the data and index file systems on my oracle packages. As suggested on this forum I used the following mount options:

-o rw,suid,
convosync=direct,mincache=direct,delaylog,nodatainlog,largefiles

We are now suffering performance problems. It appears that our physical io rates are much larger than the logical io rates.

The purpose of the cahnge was to disable double buffering of the oracle data and index IO.

Is there something we need to do on the oracle configuration to let him take advantage of this change. Or did I screw up the mount options.

Additional information. This is oracle 8.1.7 running unger a service guard cluster. The disk is all switched fiber channel attached XP256. The oracle group does backups using cpio on the data and index file systems. They claim that these backups are taking longer. Is cpio affected by these mount options?

Any help would be appreciated.
Thanks
Jim Krol
9 REPLIES 9
John Poff
Honored Contributor

Re: Oracle Mount options

Hi,

Are you running HP-UX 11.0 or 11i?

JP
John Poff
Honored Contributor

Re: Oracle Mount options

I asked about the OS version because Clay Stephenson has mentioned before that on 11i all the filesystems run faster when using the buffer cache. I haven't tried this yet with Oracle filesystems, but I have tested a filesystem using an 'awk' script that was in production and running slow. Without the buffer cache the awk script would take 5-6 minutes to complete. Using the buffer cache, the same awk script would complete in 25-30 seconds.

Take a look at this post:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x402436e69499d611abdb0090277a778c,00.html

JP
Sanjay_6
Honored Contributor

Re: Oracle Mount options

Hi,

We use the following options,

vxfs rw,suid,largefiles,delaylog,datainlog

The Filesystems is vxfs / largefiles.

Hope this helps.

Regds
Rammig Claus
Frequent Advisor

Re: Oracle Mount options

Hi Ted,

convosync=direct,mincache=direct meens, that there is no use of the unix-buffer-cache.
This could be good for Oracle-datafiles and redofiles because Oracle has its own datacache (SGA).
However you make an cpio there is no use of the SGA, and every write to disk had to be acknowleged by disk.
convosync=direct,mincache=direct are good for datafiles but you should mount your Oracle binaries only with delaylog.

Best regards ...
Claus
No risc no fun
Jim Krol
Advisor

Re: Oracle Mount options

The problem occurs on both 11i and 11.00. I realize that the cpio will probably take longer, that is not my primary concern. What I am really concerned with is that we are haveing oracle performance problems. Small transactions that used to take a second are now taking 5-10 seconds. I was under the impression that these mount options would improve oracle performance, or at least not decrease it. The most noticeable problem is on an 11.00 system. Prior to my attempt to improve performance the mount options were -o rw
which took the defaults. I guess there is some wisdom in the "Don't touch it you'll break it" philosophy.
Dietmar Konermann
Honored Contributor

Re: Oracle Mount options

Using the "direct" settings indeed switches to VX_DIRECT caching advisory, which means bypassing the buffercache and going directly to the disk.

The mincache=direct option indeed effects cpio. Especially because cpio uses really small reads (4K) during backup.

Oracle usually opens its files using O_SYNC/O_DSYNC. So you may try to use the mincache default setting with only convosync=direct.

Best regards...
Dietmar.
"Logic is the beginning of wisdom; not the end." -- Spock (Star Trek VI: The Undiscovered Country)
James Lakin
Occasional Advisor

Re: Oracle Mount options

Hi Jim,
I am not sure this will add value as I believe if you are running a clustered oracle instance you'd be running OPS with raw files as file systems cannot be shared across nodes.

Anyway, I approached the bypass of the UNIX buffer by utilizing the disk_asynch_io = true in my init.ora. You must also have the async driver running within the O/S kernal. The async driver in /dev is owned by oracle.

Again, this may not apply to your situation, but it is working well for me to do what you are looking to do. Just a thought.
Steven E. Protter
Exalted Contributor

Re: Oracle Mount options

lose the largefiles parameter if you're using 11.11(11i). Its not needed and if its in there the fs won't mount.

Here are are current mount settings for oracle data fs, we've made no performance improvements, perhaps this is a problem.

/dev/vg00/oracle2 /oracle2 vxfs delaylog 0 2

Steve
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Jim Krol
Advisor

Re: Oracle Mount options

The final resolution was to go back to the old mount options.
There were vague references in metalink about worse performance if you do long sequential access to the DBs. Some of our databases were fine with the new mount options but others had a lot of heartburn.

Thanks to all that responded, I think I got points assigned to all. I guess it comes down to "don't touch it you'll break it", "you touched it, you broke it"

Jim Krol