Re: interpreting wait/stat time per second
- From: "Alex Gorbachev" <ag@xxxxxxxxxxxx>
- To: Laimutis.Nedzinskas@xxxxxxxxxxxxx
- Date: Wed, 1 Aug 2007 12:32:06 -0400
It depends where you are looking.
If you look at instance wide like Statspack (or V$SYSSTAT
V$SYSTEM_EVENT) then you are right about CPU. Note that for waits it
can be much more than that - it's possible that 100 sessions were
waiting on IO and that would give you 100 seconds per second easily.
If you are looking at session level (like V$SESSTAT and
V$SESSION_EVENT) than you should see 1 CPU second per second and not
more than 1 seconds of waits.
However, it's all valid only when you have very short waits and calls
because of the following.
However, waits are only posted at the end of the waiting period so if
you were waiting for long time - it will be attributed to the last
second. CPU - it's reporting at the end of the call (well, there are
actually two metrics and I reckon they are a little different in this
regard). I don't think that it has changed in the latest releases
either but I could be wrong.
On 8/1/07, Laimutis Nedzinskas <Laimutis.Nedzinskas@xxxxxxxxxxxxx> wrote:
>
>
> Is is true that time metrics per second can be more than 1 second(per
> second) when server has more than one CPU? Meaning for example that on 8 CPU
> machine maximum wait time should not exceed 8 seconds unless we have some
> rounding errors or bug.
>
>
> Fyrirvari/Disclaimer
> http://www.landsbanki.is/disclaimer
--
Alex Gorbachev, Oracle DBA Brewer, The Pythian Group
http://www.pythian.com/blogs/author/alex http://www.oracloid.com
BAAG party - www.BattleAgainstAnyGuess.com
--
//www.freelists.org/webpage/oracle-l
Other related posts:
