Veritas Quick IO & fs_async setting

[Greg Wooden]

Hello,

We're preparing to use Veritas Quick IO for async Oracle IO to EMC storage. I've installed the appropriate drivers (asyncdsk & asyncdsk_included) & setup special file /dev/async, but haven't found any documented guidance on whether to set kernel param fs_async to "1".

Anyone have any ideas?

[Rita C Workman]

Nope..don't know.

But I posted something asking about Veritas and got no responses either. I did however find a newsgroup for Veritas you might want to check into...

veritas.cluster_server.english

Hope that at least helps put you in contact with someone who might know...

Rgrds,
Rita

[Rammig Claus]

Hello Greg,

the fs_async kernelparameter is a parameter only for HFS and not for VXFS or rawdevices.

Claus

[Dietmar Konermann]

Yes, Claus is correct. Forget about fs_async.

Take care to use the correct minor number for the /dev/async device special!

See Doc KBRC00009582:
http://www4.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000062910205

I saw several sites where an improper setting of bit 0x000004 caused severe data corruption.


Best regards...
Dietmar.

[Robert Thorneycroft]

I don't know where the information fs_async is for HFS comes from I have certainly never heard of this before, it does not mention it in help on the parameter within SAM or within this rather useful link:
http://docs.hp.com/hpux/onlinedocs/939/KCParms/KCparam.FSasync.html

What I would say on this matter is that you should really contact technical support at Veritas for a definitive answer as if for any reason the two conflict with each other then you could potentially corrupt data which is probably not the desired effect.

If Veritas say that it has no impact then I would simply perform a few benchmarks to find out whether there is any noticable performance advantage to be had. If you do not notice any difference I would be tempted to remove the parameter as it can cause problems in certain failure instances.

Kind regards,

Robert Thorneycroft

[Robert Thorneycroft]

My mistake I was forgetting the mount options for vxfs specified the cache behaviour.

Doh! I will just shut up now ;)