[Orca-users] Determining Disk I/O Bottlenecks with Orca

Charles Dennett charles.dennett at kodak.com
Fri Mar 24 08:50:43 PST 2000

From: Charles Dennett <charles.dennett at kodak.com>

I've installed Orca on a whole bunch of systems here.  Some of
us met yesterday to discuss Orca and compare it to something
else we had been using.  This other stuff was home grown and
uses simple UNIX commands (df, vmstat, iostat, ps swap, netstat)
to collect data and produce some rudimentary graphs.   Orca
is far superior.  

However, there was one item we collected that Orca seems not to
have.  That is the disk service time as collected by iostat -x.
As a side note, I read the docs for the SE script siostat.se and
saw the reference that what iostat -x puts out is really not the
disk service time and read Adrian's (God bless him) article
that was mentioned there.  I have not fully studied it yet but
I get the gist of what he is saying.  We used this (erroneous)
data to help figure out where disk I/O bottlenecks were.
If the service times went too high we started looking at distributing
the loads across new or other disks.  We based our rules on the
ones Adrian specified on page 228 of the first edition of
his book.  I also have the second edition of Adrian's book and
can see the same article in Chapter 8.  I'm studying that 
chapter now.

The way orca is now, the only measurement of disk activity I see
is the Disk Run Percent graph (which I believe is showing the
% busy of the disk, right?)  So, using orca, how would one
determine if an I/O bottleneck existed?  Would it be useful to
extend orcallator.se and orca to collect and display data similar
to what the siostat.se script collects? Is it a simple case of
keeping the Disk Run Percent below a certain value?  Something

Thankls for any words of wisdom anyone can provide.

Charlie Dennett
Eastman Kodak Company, Rochester, NY
Sitemaster http://www.kodak.com

Free Cool Black Virgin T-shirt with ANY order at virginmega.com!

More information about the Orca-users mailing list