RHEL3: iowait and idle output swapped(?) in top

[Lord of Gore]

Hy Mike,


yes I did. It was not that specific setup, in my case it concerned a
mail server - sendmail - where I had 10% cpu time and the load average
went to 15 ...
You cand find bottlenecks in the network. Clearly if this is the
situation you have enough processing power but the data is coming too
slow for the processes.
This could be due to faulty or under capacitated hardware but also
misconfigured software.
Of course given the complexity of the problem and the lack of info on my
side it is hard to pinpoint exactly where the problem is ...
You could start if it is not a very big problem to stop services one by
one and see if the load average drops. If not then you should focus on
the SAN ...

Bogdan Dumitrache
http://www.logsoftgrup.ro

On Mon, 2009-10-19 at 09:53 -0400, Mike Burger wrote:

> I've recently inherited a small set of RHEL installations, which were run
> through P2V and are now running as VMWare guests.
> 
> The users were complaining that one of them, running as an oracle server,
> appeared to be running slowly.
> 
> Of course, my first instinct was to run top, which showed a pretty high
> iowait number (around 134% for a 2 CPU instance).
> 
> As it turned out that we were running a deduplication on the storage that
> housed the VM, our VM admin had the storage team stop the dedupe...our
> thought was that the iowait might be in the storage area, and that turning
> off the dedupe might speed things up.
> 
> The opposite effect was shown in the iowait number...it actually went up,
> and has hovered around 170-190% since stopping the dedupe, while the
> number in the idle column hovers between 0 and 10%.
> 
> vCenter and our monitoring tools (CA's eHealth) show the opposite...they
> show the CPU's idle state between 80 and 95, during the period since
> stopping the dedupe.
> 
> My question...has anyone else seen a situation where top, running on a
> multi-cpu config, in (or not) a VM, might have the iowait and and idle
> numbers reversed in the top output (procps 2.0.17)?
> 
> Thank you.
> 
> 
> -- 
> Mike Burger
> http://www.bubbanfriends.org
> 
> Visit the Dog Pound II BBS
> telnet://dogpound2.citadel.org or http://dogpound2.citadel.org
> 
> To be notified of updates to the web site, visit:
> 
> https://www.bubbanfriends.org/mailman/listinfo/site-update
> 
> or send a blank email message to:
> 
> site-update-subscribe at bubbanfriends.org
>