Hi all,
After my summary some more answers came in from fellow Sun Managers.
Attached is their explanations and worth to look at and read.
I also found the culprit of all the traffic to the disks. Our Oracle DBA
was (and still is) busy downloading data from his Oracle Database on
the mainframe to the SUN (Decision Support Server) into the Oracle
tables which is striped over all the disks in the SSA-214. This
explaines why my two RSM-2000's is not hogged with wio. The problem is
that the usually nightly jobs fell behind so far that they are running
it during the day. They promissed that the downloaded programs is in the
process of rewritting and will only run for a couple of minutes every
night.
Thank again for all the help.
Gert Marais
>From Alun.Gwynne@capgemini.co.uk -
Gert,
Just saw your summary, missed the original.
Without checking your stats in detail, I think Casper is right with this
- if you check out Adrian Cockcroft's book/papers he explains how the
wait for io is calculated on multiprocessor systems - basically, if
there are any processors at all in wait state (vmstat "b" column > 1)
then ALL the processors on that system are measured as being in wait for
io. It's just a "feature" which he's bugged to Sun, and I think is fixed
in Solaris 2.7.
Check out his papers in SunWorld. This very subject is in October's
issue in fact.
Cheers,
Alun.
>From Brion Leary [brion@dia.state.ma.us]
Marais,
Some more info for you to ponder ...
(1) Check out
http://www.sunworld.com/sunworldonline/swol-10-1998/swol-10-perf.html
A. Cockcroft explains in this article that wait time on SMP
servers is
charged against all processors, not just the processor that is
waiting.
That would mean that your sar 'Average' line should really look
like
Average 9 1 3
88
(2) Usually when I see a number under w in vmstat, and the system is
not
swapping, it means that ...
(1) a memory shortfall occurred
(2) to satisfy the demand for active memory idle processes where
swapped out
(3) the idle processes remain swapped out (listed under w) until
they run again, if ever. Often it will be processes
such as
getty on an unused port, daemon processes for unused
services,
shells that were used to start a currently running job,
etc.
and from Patrick Shannon <pshannon@macromedia.com>
Could it be network lag? I believe that can show as wio.
Patrick
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:12:50 CDT