Re: Determining what is causing high wio%

Neil Greene wrote:

> > Well....I determined the cause of the high percentage of wio% was coming
> > from the databases running on the system. After starting one instance,
> > the system averaged around 26% wio, after starting the second instance,
> > 46-50% wio, and the % increased as additional databases were started.
> >
> > Has anyone else noticed this same problem or cpu usage??
>
> Here I go answering my own questions again. You would think this was an
> Oracle tar. Anyway, looking at my cpu and system statistics from
> vmstat....I see what I expect to see. CPU utilization is around 20-30
> user, 10-15 sys and 40-50 idle. That is how the system "feels" like it
> is running and what I would expect. And yet, iostat, mpstat and sar
> report different numbers with high wio%.
>
> Has anyone else experienced these problems on Solaris 2.5.1 (no patches
> on this particular system) and other programs?
>
> --
> Neil Greene
> Senior System Engineer / Oracle DBA "Tyra, you left your toobrush
> MCI Systemhouse, Inc. at my house last night."
> email: ngreene_at_laoc.SHL.com

Neil,

	This is a very common question on Unix.  Vmstat reports CPU
	usage as %usr, %sys and %idle.  The other commands (sar, iostat,
	mpstat, etc.) report CPU usage as %usr, %sys, %idle and %wio.
	Here is the difference:

	%idle and %wio are BOTH idle CPU time.  If you add up %idle and
	%wio from sar it should equal %idle from vmstat.  Vmstat just
	doesn't differentiate between wio idle time and non-wio idle
	time.

	wio means that the CPU is idle (doing absolutely nothing) but
	it has an outstanding IO request it is waiting on.  The cpu is
	available for additional work.  In fact, if you increase the
	%usr and %sys load on the system, %wio will decrease because
	the cpu's aren't idle anymore.  So the only difference between
	%idle and %wio is that for wio the cpu is waiting on IO (hence
	the name) and for %idle it is not.  

	High wio on an oracle database machine is not unusual - it
	reads and writes a lot of data.  It can be an indication of
	overloaded or imbalanced disks, so you should check the disk
	statistics to see if you have a problem.  If that looks ok and
	response time is good, then you're OK.  If it bothers people
	tell them to use vmstat and they won't see it!

	Hope this clears up the confusion.

-- 
---
Allen Kirby			AT&T ITS Production Services
akirby_at_att.com			Alpharetta, GA.