fs_async

[Darren Gibbs]

Our Sybase team wishes to turn on fs_async for filesystems because their tempdb is on a mounted filesystem instead of raws. My concern with making this change is what effect it will have on all other filesystems on the server?

Can I limit the fs_async usage to specific filesystems via mount options? BTW - All of our filesystems, except /stand, are vxfs. We do not have On_Line JFS on this server.

[Dietmar Konermann]

fs_async is only for HFS... so don't worry about that for VxFS.

[Ted Ellis_2]

you sure they are not referring to creating the async device that can be used for asycronous transfer... we have that set for Sybase here... to create it:

cd /dev
mknod async c 101 0x000004

then change it to have sybase ownership and the DBA team can use that device for async work with the database... check with them and see if that is what they are after, as this will impact the database operation.

[Deshpande Prashant]

Hi
Check this thread out.

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x8995663ce855d511abcd0090277a778c,00.html

Thanks.
Prashant.

[Darren Gibbs]

Ted - I assume you're referring to moving the async driver into the kernel and then performing the mknod. Yes, we did that a couple of years back on our K-class servers and again on our N4000 servers when we implemented them in 2000.

The dba's use the async drivers for their raw devices. However, they've chosen to put their tempdb on a filesystem structure instead of raws and I'm wondering about fs_async on filesystems to gain performance of the tempdb for them.

I've heard that fs_asyn is only for HFS which to me is taking a step backwards.

On our DataWarehouse server, we run Oracle on filesystems and have On-line JFS utilizing the mincache and convosync options. Difference here is that we havn't purchased On-line JFS for the server in question.