DB2 Linux settings (cache/buffers/IO)

[Jurgen Haan]

Hi. I'm running a DB2 8.2 install on a X86_64 SuSE install.
The tablespaces and logs are stored on a netapp filer through an NFS mount.

Now I came across this article:

http://www-128.ibm.com/developerworks/db2/library/techarticle/dm-0509wright/

In that document the possibility of Direct I/O is given to improve
performance. I was wondering if this would also apply to NFS shares (NFS
being a virtual disk etc).

Furthermore, what's the best strategy to pursue on buffer allocation?
How can I best minimize the use of linux buffers? (thus increasing the
use of DB2 allocated memory?)

-R-

[dunleav1]

Re: DB2 Linux settings (cache/buffers/IO)

On Mar 8, 9:12 am, Jurgen Haan <jur...@fake.do mwrote:

Hi. I'm running a DB2 8.2 install on a X86_64 SuSE install.
The tablespaces and logs are stored on a netapp filer through an NFS mount.
>
Now I came across this article:http://www-128.ibm.com/developerwork...rticle/dm-0509...
>
In that document the possibility of Direct I/O is given to improve

performance. I was wondering if this would also apply to NFS shares (NFS
being a virtual disk etc).
>
Furthermore, what's the best strategy to pursue on buffer allocation?
How can I best minimize the use of linux buffers? (thus increasing the
use of DB2 allocated memory?)
>
-R-

Set - db2set DB2LINUXAIO=tru e.

Turn off file system caching.

I don't think I would have database data files on a NFS mount.

Buffer allocation depends - But I have seen good results from 1
bufferpool for each of my database page sizes and 1 for large temp
page size.

[Mark A]

Re: DB2 Linux settings (cache/buffers/IO)

On Mar 8, 11:30 am, "dunleav1" <j...@dunleavye nterprises.comw rote:

Set - db2set DB2LINUXAIO=tru e.
>
Turn off file system caching.
>
I don't think I would have database data files on a NFS mount.
>
Buffer allocation depends - But I have seen good results from 1
bufferpool for each of my database page sizes and 1 for large temp
page size.

What about long tablespaces that contain LOB or LONGVARCHAR data that
do not use bufferpools, and are synchronously accessed from disk each
time?