From: Holger.VanKoll@SWISSCOM.COM
Date: Mon Oct 07 2002 - 08:55:51 EDT
Hello,
I got a S80, 6 CPU, 6GB with 95-100% cpuload, 4.3.3 ML10.
There is no wait, sometimes 3-5% idle. Topas shows no disk busy more
than 5%.
According to topas system load is approx. 40-50% kernel (vmstat: sy) and
50-60% user. 5 to 15 procs are in r column of vmstat, 2-4 in b.
There is no paging.
The user-load is distributed among 60 oracle-processes, those with the
highest cpu-time have 4-5%. There is rarely a oracle-process that uses
10% or more; and only a few seconds if there is one.
If I sum up all aioservers, I get about 30%. maxservers is 480, thats
reached 20min after reboot.
Any suggestions what is wrong here?
I assume the cpu-time of the aioservers is completely counted to kernel
and nothing goes to user, right?
If yes, it might be worth a try to schedule more cpu-time to oracle.
I assume I wont gain anything be decreasing maxservers; fewer aioservers
can user the same amount of cpu (not sure, however).
I think giving them the same priority (setting kprocprio to 40) as
user-processes would be more effective.
What do you think?
The problem with testing aio is as always the need for a reboot. chdev
-T does not work.
So if you ever had a system with same load-characteristics, I would
appreciate any experiences.
Thank you.
Regards,
Holger
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 22:16:15 EDT