[Ocfs2-users] OCFS2 Block / Clustersize with Oracle 10gR2

[Sunil Mushran]

If your aim is to compare db performance on raw v ocfs2, look
at the actual thruput. As in, transactions per sec. Use whatever tool
you want or that corresponds best to your workload.

The detail statspack data provided by enterprise manager is useful
in tuning the db itself. Say, for sizing the buffer cache. But there are 
some
numbers useful for getting the ave io time.... sequential read, parallel 
read,
etc.

iostat will provide raw io thruput numbers like MB/s. Do dd against the raw
device and you will get a baseline for comparing io performance.

ios/sec, as a standalone stat, is not that useful. Look at the thruput.
mb/s and/or tps.

Brian Long wrote:
> When measuring performance between raw and OCFS2 using Enterprise
> Manager and taking half-hour snapshots, are physical reads/writes per
> second directly comparable or should we only be looking at "db block
> changes", "session logical reads", etc?
>
> I've got iostat data and graphs of the storage when doing raw and OCFS2
> IO tests.  I've also got Enterprise Manager screen shots if they provide
> any value.
>
> Besides switching to Swingbench, how can we verify we are truly getting
> the same "IO" between raw and OCFS2 if the physical reads/writes are not
> in the same ball-park.
>
> In fact, when the DBA disabled async IO on the database on OCFS2, the
> IO/sec did not decrease at all.
>
> Thanks for helping.
>
> /Brian/
>