Summary: CPU utilization comparison.

From: Monappallil, George <george.monappallil_at_rpfl.com>
Date: Wed Oct 24 2007 - 12:45:42 EDT
I found the problem, which was a resource pool configured with 1 cpu
assigned to the pool. Hence, the one cpu being idle all the time and
also hence the vmstat output was showing 25% idle time because one (of
the four cores) couldn't be used.

I disabled the pools and the output on the SF490 is similar to the E2900

Thanks to Richard Skelton and Vikas Sharma for prodding me in the right
direction.


-George


________________________________

From: Monappallil, George
Sent: Wednesday, October 24, 2007 9:58 AM
To: 'sunmanagers@sunmanagers.org'
Subject: CPU utilization comparison.


Guys:
I was using the below script to stress test the CPUs on an SF490 and
E2900.
The SF490 has 2 CPUs and the E2900 has 4CPUs. Both machine have 1.5MHz
UltraSparc IV CPUs.

On the SF490, I start off with 2 threads (because 2 physical cpus)

perl -e 'while (--$ARGV[0] and fork) {}; while () {}' 2
The behaviour is
*****
vmstat 1 10
 kthr      memory            page            disk          faults
cpu
 r b w   swap  free  re  mf pi po fr de sr m0 m1 m5 m1   in   sy   cs us
sy id
 0 0 0 12930032 3493952 30 6 181 2 1  0  0  5  0  0  4  697 1604 1080  1
1 99
 0 0 0 11784400 2091176 0 30 0  0  0  0  0  0  0  0  0  521 1623  930 51
0 49
 0 0 0 11784400 2091176 0 2  0  0  0  0  0  0  0  0  0  488  823  867 50
0 49
 0 0 0 11784400 2091176 0 0  0  0  0  0  0  0  0  0  0  466  365  885 50
0 50
 0 0 0 11784400 2091176 0 2  0  0  0  0  0  1  0  1  1  476 6591 1196 51
6 43
 0 0 0 11784400 2091176 0 0  0  0  0  0  0  5  0  1  5  501  524  905 50
0 49

I see roughly 50% utilization. Now when I increase the number of threads
to 4 (because these are dual core cpus) the behavour is below
******
vmstat 1 10
 kthr      memory            page            disk          faults
cpu
 r b w   swap  free  re  mf pi po fr de sr m0 m1 m5 m1   in   sy   cs us
sy id
 0 0 0 12930024 3493944 30 6 181 2 1  0  0  5  0  0  4  697 1604 1080  1
1 99
 1 0 0 11783656 2090904 0 30 0  0  0  0  0  0  0  0  0  588  477  923 74
1 25
 1 0 0 11783656 2090904 0 0  0  0  0  0  0  1  0  0  1  589  396  910 75
0 25
 1 0 0 11783656 2090904 0 0  0  0  0  0  0  0  0  0  0  572 1907  878 75
0 25
 1 0 0 11783656 2090904 0 0  0  0  0  0  0  0  0  7  0  658  399  937 75
0 25
 1 0 0 11783656 2090904 0 0  0  0  0  0  0  0  0  0  0  642 2120  916 74
1 25

I see 75% CPU utilization.
**************************

For the E2900, I start off with 4 thread and then increase it to 8
(again because it is a dual core)
With 4 threads:
**********
#vmstat 1 6
 kthr      memory            page            disk          faults
cpu
 r b w   swap  free  re  mf pi po fr de sr m0 m1 m5 m1   in   sy   cs us
sy id
 0 0 0 35573824 27958752 34 4 75 1 1  0  0  4  0  0  3  384 4503  355  1
1 99
 0 0 0 33317024 26119200 0 36 0 0  0  0  0  0  0  0  0  352  781  197 50
0 50
 0 0 0 33316952 26119128 0 2 0  0  0  0  0  0  0  0  0  342  643  148 50
0 50
 0 0 0 33316952 26119128 0 0 0  0  0  0  0  0  0  0  0  357 2723  187 50
0 49
 0 0 0 33316952 26119128 0 0 0  0  0  0  0  0  0  0  0  354  668  173 50
1 49
 0 0 0 33316952 26119128 0 0 0  0  0  0  0  0  0  0  0  344  679  172 50
0 50

I see roughly 50% utilization
With 8 threads:
******
#vmstat 1 6
 kthr      memory            page            disk          faults
cpu
 r b w   swap  free  re  mf pi po fr de sr m0 m1 m5 m1   in   sy   cs us
sy id
 0 0 0 35573816 27958744 34 4 75 1 1  0  0  4  0  0  3  384 4503  355  1
1 99
 0 0 0 33315656 26118792 0 36 0 0  0  0  0  0  0  0  0  448  753  174
100 0  0
 0 0 0 33315584 26118720 0 2 0  0  0  0  0  0  0  0  0  469  691  189
100 0  0
 0 0 0 33315584 26118720 0 0 0  0  0  0  0  0  0  0  0  461  153  184
100 0  0
 0 0 0 33315584 26118720 0 0 0  0  0  0  0  0  0  0  0  445 1184  159
100 0  0
 0 0 0 33315584 26118720 0 0 0  0  0  0  0  0  0  0  0  441  675  168
100 0  0

I see 100% CPU utilization. Why is that the E2900 has higher CPU usage
than the SF490.

Both servers are running Solaris 10 with the same patch level. Both
servers aren't running anyother processes and baseline cpu utilization
before these tests were roughly 100% idle time.

You help is appreciated.

-George
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Wed Oct 24 12:46:01 2007

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:44:07 EST