Re: I/O wait question/scenario

bordenm_at_methodisthealth.org (bordenm_at_methodisthealth.org) wrote in news:db50b9a0.0402240410.492a1604_at_posting.google.com:

> After the vendor has researched this a
> little, they have said the the I/O wait will go down once we add
> more users to the application.
>
> [ snipped ]
>
> My question is, could this theory be correct?
>

Using the high-level metrics of CPU Busy, CPU Idle, I/O Wait, etc are typically not very useful. I/O wait is, in essence, another form of idle. The CPU isn't doing anything at the moment but waiting for I/O to complete. That's about the extent you can gather from this metric.

Use sar -d with three second samples and monitor your disks. See if you have long queues, service times, etc. It's a bit more lengthy to tune a box than a simple Usenet posting. Your next question should be, what's the definition of 'long X' and that will have to come from experience on the hardware and from the hardware vendor. Typically when using SAN, you'll have a cache and your I/O's will typically be on the order of 2 to 3ms.

--
Pablo Sanchez - Blueoak Database Engineering, Inc
http://www.blueoakdb.com