| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Usenet -> c.d.o.server -> CPU used by this session - abnormalities
When is "CPU used by this session" accurate and when is it inaccurate?
The biggest problem with this stat that I've dealt with is that plsql blocks don't seem to report the CPU until they have completed, and then there is a huge spike. Other problems seem to be that background process CPU is not reported and network and datagram work is not reported. I've heard that there are problems with parallel query slaves and cpu reporting.
The reason I'm thinking about this is that CPU is such a good metric to use in figuring out if wait event times are important. I've written a graphic monitoring tool ( http://oraperf.sourceforge.net based on Quest's http://www.quest.com/spotlight_oracle/index-new.asp) where the main graph shows a CPU usage line verses time being spent waiting on non-idle wait events. For me, the annomolies haven't posed more than an annoyance, but I'm used to the spikes at the end of a plsql batch job etc. I was trying to think of more dependable ways of reporting a CPU type metric, but before I did that I thought I'd ask other peoples observations on the CPU stats. Currently, if spikes were the only problem, then in the graphic monitor I could take out the spikes cutting them off at 100% CPU and retro fitting the CPU line with the amount of CPU that got lopped off - a bit of a hack, but seems reasonable. I was just wondering if people knew of other annomolies that could be misleading. I think the worst case would be CPU not getting reported at all like with the background processes. (I haven't checked but I'd hope parallel query slaves actually report their CPU)
Other ideas I'm thinking about are getting CPU from the OS. THis would be my last resort as it gets into all sorts of port specifics. The other choice would be calculating a simliar type of metric from several other stats like user calls and lio, although lio seems to be reported for backgrounds, again as with CPU calls don't.
Kyle Hailey
http://oraperf.sourceforge.net
oraperf.sourceforge.net
 |
 |