Re: Method R and CPU Time

From: <robson_at_medidata.com.br>
Date: Fri, 2 Jul 2004 10:37:11 -0300
Message-ID: <OF481C1F8A.330AD2B8-ON03256EC5.00482DE7@medidata.com.br>

Recently I was involved in analyzing the performance of a big Oracle 8.1.7.4 and got Statspack
reports from an all day long with snaps at every half hour. To use meth= od
R, I picked the following
data from the Statspack reports:

1 - Total CPU time from the line: CPU used when call started 2 - Total Wait Time from the sum of the wait time column of the top 5 w= ait
events

Then I noticed that the CPU time sometimes were in excess of the elapse= d
time (30 min).
As the system is a Sun SF10K running with 44 CPU I calculated the avera= ge
CPU rate
scaled to the number of CPUs. That is CPUavg=3D (CPU time/elapsed time)= / 44.
The numbers I found for each interval were in turn very reasonable and=

pretty followed the
transactions per second rate.

Anyway on some periods I found the relation between the original Total = CPU
time and the Total
wait time very odd with values like 0.02 % of total time(CPU/(CPU+WAIT)= .

I used also the Anjo Kolk oraperf analyzer but there the wait times are=

being multiplied by 100
on the response time calculation giving very small figures for the CPU = rate
results.

Does anyone have seem this behavior when analyzing Oracle running on massive multiprocessor systems ?

Thanks for any help

Robson Gomes

Medidata Inform=E1tica S.A

(21) 2546 3734

P Antes de imprimir pense em sua responsabilidade e compromisso com o M= EIO
AMBIENTE! O conte=FAdo desta mensagem (e o de seus eventuais anexos) =E9 de exclu= sivo
interesse do destinat=E1rio acima indicado e pode conter informa=E7=E3o=

confidencial e/ou propriet=E1ria, que n=E3o deve ser copiada ou divulga= da sem o
consentimento expresso da Medidata Informatica S/A.. Em caso de recep=E7= =E3o
por pessoa diversa do destinat=E1rio, a mensagem e seus respectivos ane= xos
dever=E3o ser destru=EDdos imediatamente sob pena de responsabiliza=E7=E3= o por uso
indevido.
The contents of this message (and of its eventual enclosures) are inten= ded
to the exclusive benefit of the addressee indicated above and may conta= in
confidential and/or proprietary information and may not be copied or disseminated without express consent of Medidata Informatica S/A. In ca= se
of unduly reception thereof by any other party than the addressee above= ,
this message and its enclosures shall be immediately destroyed. The use= or
disclosure of the contents of this message and/or its enclosures will become such other party liable for such action.=

Please see the official ORACLE-L FAQ: http://www.orafaq.com

To unsubscribe send email to: oracle-l-request_at_freelists.org put 'unsubscribe' in the subject line.
--

Archives are at http://www.freelists.org/archives/oracle-l/ FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html

Received on Fri Jul 02 2004 - 08:35:16 CDT