Reading Sar Output

[Marc Bohnert]

I have a K Class machine with 6 processors. I collected performance data using sar -A in 5 minute intervals for 1 day. I am now trying to understand the collected data to determine if I have an i/o bottleneck. Does anyone know if documentation exists that will tell me what the data means and how to relate it when there are multiple processors? The sar man page does not.

For example the data

%usr %sys %wio %idle
14 8 70 9

How do I relate these values based on a 6 processor machine.

Any help would be kindly appreciated.


Thank you

[Philip Chan_1]

In your example,

%usr - percentage of cpu time spent on user process
%sys - percentage of cpu time spent by the kernel system calls
%wio - percentage of cpu time that spent on polling devices (high percentage imply inefficiency)
%idle - percentage of idled cpu time

Your %wio is 70%, that seem too high to me. This indicated certain disks may be overloaded. Take a look at the "sar -d" output to see the disk activities, one or more of your disks might have reached near 100% utilization. If that is the case then try balancing the disk activities if you could. Also, take a look at "sar -w" for the swap rate, make sure these disk activities werent' coming from swapping.

See if you could get hold of the book "System Performance Tuning" by Mike Loukides, which has got some good description on sar. The man page is useful too. But you must read through it with patient.

Rgds,
Philip

P.S. turning on file system asychronous I/O may help reducing wio percentage, any thoughts from you guys?

[Philip Chan_1]

Marc,

Wouldn't it be easier to view historical performance data via perfview? may be you should look into such tool.

Rgds,
Philip

[Andreas Garbe]

Marc,

for collect sar infos 1 entry on crontab for root user

0,10,20,30,40,50 * * * * /usr/lbin/sa/sa1

the files are save /var/adm/sa

for multiprozessor systems you can see all cpu

sar -M

18:10:00 0 0 0 1 99
1 0 0 0 100
2 0 0 0 100
3 0 0 0 100
system 0 0 0 99
18:20:00 0 0 0 1 99
1 0 0 0 100
2 0 0 0 100
3 0 1 0 99
system 0 0 0 99

Average 0 0 0 1 99
Average 1 0 0 0 100
Average 2 0 0 0 100
Average 3 0 0 0 100
Average system 0 0 0 99

rgds
Andreas

[Bill McNAMARA_1]

regarding wio problems, you'll probably can get that down by tuning lvm correctly.
send the output of your strings /etc/lvmtab and ioscan -fnk and we can take a look.

Later,
Bill

[CHRIS_ANORUO]

The document is- Doc id:S3100002312A-C
Description: Sys Adm: determining the cause of system performance problems