 | | | Adding CPU to a unix box--added capacity. | Adding CPU to a unix box--added capacity. 2007-11-03 - By Best, David
Back
A while back we doubled the number of cpu's in a system from 2 to 4.
I monitored CPU usage at 5 minute intervals for the week before and the
week after.
For the 4cpu week the average CPU usage was: 26.39815
For the 2cpu week the average CPU usage was: 54.78056
So, CPU usage pretty much halved which was what we expected. However,
that doesn't necessarily mean performance will increase by a factor of
2. It depends on the application. In this case the databases on this
server were a mix of OLTP and Datamarts. The OLTP databases did see an
improvement but the datamarts didn't. The reason is that OLTP
transactions are normally short and sweet, so more could be processed at
a given time. In the case of the datamart, the refresh job didn't take
advantage of multiple cpu's, so it still took the same amount of time to
complete.
__ ____ ____ ____ ____ ____ ____
From: oracle-l-bounce@(protected)
[mailto:oracle-l-bounce@(protected)] On Behalf Of Zhu,Chao
Sent: Friday, November 02, 2007 10:32 AM
To: oracle-l
Subject: Adding CPU to a unix box--added capacity.
hi,
Does anyone have experience on adding new CPU to a existing box,
like upgrade from 8CPU to 16 CPU for example.
I am curious about two things;
1. How many CPU usage% drop do we expect? Like previously it is running
at 50% CPU usage, typically will it drop to 25%? or something like 30%?
To ask the same question from another view, is , how much capacity we
add to the box? 16CPU box = 1.X(?) of a 8CPU box. Assuming same
archtecture/OS. (same cpu count server in different architecture may
have different processing power?)
Would be nice if you can share with your experience (like on
solaris/aix/hp etc). That's kind of typical server scalability.
2. Can we say, CPU_USED_BY_THIS_SESSION (stat in oracle) ~~
Unix_CPU_COUNT * USER_MODE_CPU% (vmstat usr column)? I did couple of
compare in my database, they look pretty close.
Does anyone ever thinking about have such a formular: it could be
very fun.
CPU_USED_BY_THIS_SESSION = X*logical_reads + Y*physical_reads + Z*sorts
+ A * executions + B * redo_size + C * Latch_Spins+ .......?
by the way, anyone like the book"forecasting oracle performance"?
--
Regards
Zhu Chao
www.cnoug.org
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META http-equiv=Content-Type content="text/html; charset=us-ascii">
<META content="MSHTML 6.00.2900.3199" name=GENERATOR></HEAD>
<BODY>
<DIV dir=ltr align=left><SPAN class=609305423-03112007><FONT face=Arial
color=#0000ff size=2>A while back we doubled the number of cpu's in a system
from 2 to 4. I monitored CPU usage at 5 minute intervals for
the week before and the week after.</FONT></SPAN></DIV>
<DIV dir=ltr align=left><SPAN class=609305423-03112007><FONT face=Arial
color=#0000ff size=2></FONT></SPAN> </DIV>
<DIV dir=ltr align=left><SPAN class=609305423-03112007><FONT size=2>For the
4cpu
week the average <SPAN class=hl>CPU</SPAN> usage was: 26.39815<BR><BR>For the
2cpu week the average <SPAN class=hl>CPU</SPAN> usage was:
54.78056</FONT></SPAN></DIV>
<DIV dir=ltr align=left><SPAN class=609305423-03112007><FONT
size=2></FONT></SPAN> </DIV>
<DIV dir=ltr align=left><SPAN class=609305423-03112007><FONT face=Arial
color=#0000ff size=2>So, CPU usage pretty much halved which was what we
expected. However, that doesn't necessarily mean performance will
increase
by a factor of 2. It depends on the application. In this case the
databases on this server were a mix of OLTP and Datamarts. The OLTP
databases did see an improvement but the datamarts didn't. The reason is
that OLTP transactions are normally short and sweet, so more could be processed
at a given time. In the case of the datamart, the refresh job didn
't
take advantage of multiple cpu's, so it still took the same amount of time to
complete.</FONT></SPAN></DIV><BR>
<DIV class=OutlookMessageHeader lang=en-us dir=ltr align=left>
<HR tabIndex=-1>
<FONT face=Tahoma size=2><B>From:</B> oracle-l-bounce@(protected)
[mailto:oracle-l-bounce@(protected)] <B>On Behalf Of
</B>Zhu,Chao<BR><B>Sent:</B> Friday, November 02, 2007 10:32 AM<BR><B>To:</B>
oracle-l<BR><B>Subject:</B> Adding CPU to a unix box--added
capacity.<BR></FONT><BR></DIV>
<DIV></DIV>hi,<BR> Does anyone have experience on adding new
CPU to a existing box, like upgrade from 8CPU to 16 CPU for example.
<BR> I am curious about two things;<BR>1. How many CPU usage%
drop do we expect? Like previously it is running at 50% CPU usage,
typically will it drop to 25%? or something like 30%? To ask the same question
from another view, is , how much capacity we add to the box? 16CPU box = 1.X(?)
of a 8CPU box. Assuming same archtecture/OS. (same cpu count server in
different
architecture may have different processing power?) <BR> Would be nice if
you can share with your experience (like on solaris/aix/hp etc). That's kind of
typical server scalability. <BR><BR>2. Can we say,
CPU_USED_BY_THIS_SESSION (stat in oracle) ~~ Unix_CPU_COUNT * USER_MODE_CPU%
(vmstat usr column)? I did couple of compare in my database, they look pretty
close. <BR><BR> Does anyone ever thinking about have such a
formular: it could be very fun. <BR>CPU_USED_BY_THIS_SESSION = X*logical_reads
+
Y*physical_reads + Z*sorts + A * executions + B * redo_size + C * Latch_Spins+
.......?<BR><BR clear=all>by the way, anyone like the book"forecasting oracle
performance"? <BR>-- <BR>Regards<BR>Zhu Chao<BR><A
href="http://www.cnoug.org">www.cnoug.org</A><BR></BODY></HTML>
|
|
|