Re: VMWare and Oracle IO

  • From: Jared Still <jkstill@xxxxxxxxx>
  • To: Stefan Koehler <contact@xxxxxxxx>
  • Date: Tue, 7 Oct 2014 17:01:40 -0700

comments inline:

On Tue, Oct 7, 2014 at 9:33 AM, Stefan Koehler <contact@xxxxxxxx> wrote:

> Hi Jared,
> a common example of the difference between service time (= I/O time that
> the SA usually measures) and host wait time (= I/O time from Oracle
> performance perspective). Referring to
> http://bartsjerps.wordpress.com/2011/03/04/io-bottleneck-linux/
>

That is a concise and informative article, thanks for that.


> What kind of I/O took 3.5 seconds - single block or multi Block I/O (in
> case of mb - what request size)? Any reason for using VMDKs?
>

As reported earlier, small multiblock read.  Even a big one should not take
this long.
Don't yet know about the VMDK question, but will find out.


> I would also cross-check the queue size of the ESX host HBAs and the used
> multipathing options. However i mostly recommend to use VMware RDM (for
> performance, management and troubleshooting reasons) for the Oracle VMware
> infrastructure. There are some specific reasons to use the storage
> virtualization layer (e.g. VM snapshots), but go for RDM if they do not
> get used or the (enterprise) storage provides the capability.
>
> You can do much more troubleshooting with direct SCSI (RDM throughput) in
> a VMware environment with Oracle databases.
>

Thanks, great points for discussion.


Jared Still
Certifiable Oracle DBA and Part Time Perl Evangelist
Principal Consultant at Pythian
Pythian Blog http://www.pythian.com/blog/author/still/
Oracle Blog: http://jkstill.blogspot.com
Home Page: http://jaredstill.com

Other related posts: