Direct I/O and NFS

[Jurgen Haan]

Hi everybody.

If I would create tablespaces that do not use filesystem caching on an
NFS share... What would this change in performance? Is this of any use?
And is it safe? I've read some applications have trouble in using direct
I/O on NFS shares (though none of which were DB2).

In short: is bypassing filesystem caching on a AMD64 Linux system with
tablespaces on NFSv3 shares advisable?

Ty.

-R-

[Knut Stolze]

Jurgen Haan wrote:

Hi everybody.

If I would create tablespaces that do not use filesystem caching on an
NFS share... What would this change in performance? Is this of any use?
And is it safe? I've read some applications have trouble in using direct
I/O on NFS shares (though none of which were DB2).

Different question: why do you want to place your containers on an NFS share
in the first place?

--
Knut Stolze
DB2 z/OS Admin Enablement
IBM Germany

[Jurgen Haan]

Knut Stolze wrote:

Jurgen Haan wrote:

>Hi everybody.

If I would create tablespaces that do not use filesystem caching on an
NFS share... What would this change in performance? Is this of any use?
And is it safe? I've read some applications have trouble in using direct
I/O on NFS shares (though none of which were DB2).

Different question: why do you want to place your containers on an NFS share
in the first place?

Well.. The server has limited local disks, plus failover is not too easy
that way. So we're using a NAS storage array.

It's a netapp FAS250 filer storage array.
(http://www.b2net.co.uk/netapp/networ...app_fas250.htm)
We're using the filer for the tablespace container storage array.
Although the NetApp supports iSCSI, the use of snapshot technology is
crippled by this, and SAN, well... Not too big a fan of the performance
vs price, plus it can not scale too much further anymore.

http://www.redbooks.ibm.com/redbooks/pdfs/sg246538.pdf
Redbook db2 on NAS arrays.