Re: I/O wait question/scenario

bordenm_at_methodisthealth.org wrote:

> Ok, here is the scenario. We have an application, clinical, that the
> db server is having pretty high I/O wait. It can be as high as 50 to
> 60%, but it is not consistently like this. It is more consistently
> around 30 to 40 during busy times. This is running on a p690 regatta
> attached to a shark. The partition is set up with 7procs and 16gig of
> ram. 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.
> They are saying that the I/O wait is high because the processesors are
> so fast that it processes the transactions faster than what it can
> handle. Like I said, they say that the I/O wait will go down once we
> make the runqueues go up by adding more users to the application. To
> me, this doesn not sound correct.
>
> My question is, could this theory be correct?
Hi,

it depends...

If your not IO-bound, then the theory is right, cause wio is (from the cpus point of view) a kind of idle. It simply means, that the cpu is idle while a IO is underway - an IO, not necessarily a foreground oracle read. I might be an dbwr idle write, it might be a IO from a non-oracle process!
Increasing the cpu load will remove the wio in this case.

But, if your sessions are already spending most of the response time waiting for IO (check v$session_event), the increasing the load will most probably increase your IO-bound problem.

Stephan Received on Wed Feb 25 2004 - 07:48:41 CST