Originally posted by: SystemAdmin

Dear all
I had collected stats report from nmon.From that cpu summary graph i was not able to tell wether the cpu or the io disk are used properly .What are the avg utilization that should be for IO and user%+sys%. Below are the stats from nmon

Server 1
CPU: User% Sys% Wait% Idle% CPU%
Avg 3.2 2.2 10.8 83.7 5.4
Max 25.6 13.0 72.7 99.5 28.6
Peak:Avg8.0 5.8 6.7 1.2 5.3

Avg tps during an interval: 123
Max tps during an interval: 866
Max tps interval time: 18:51:40
Total number of Mbytes read: 5,167
Total number of Mbytes written: 6,763
Read/Write Ratio: 0.8

Server 2

CPU: User% Sys% Wait% Idle% CPU%
Avg 9.5 3.6 0.3 86.7 13.1
Max 23.5 5.7 1.1 93.8 28.4
Peak:Avg2.5 1.6 4.2 1.1 2.2

Total System I/O Statistics
Avg tps during an interval: 5
Max tps during an interval: 22
Max tps interval time: 9:32:04
Total number of Mbytes read: 146
Total number of Mbytes written: 62
Read/Write Ratio: 2.3

Thnx in advance.
Waiting prompt answers.

Regds
Harmeet

#AIX-Forum

Originally posted by: nagger

Please put nmon questions on the Performance Tools Forum
at http://www-128.ibm.com/developerworks/forums/dw_forum.jsp?forum=749&cat=56
and not here in future.
#AIX-Forum

Originally posted by: nagger

I do not understand the urgency here!
If this is urgent then you should immediately contact AIX Support and not a User Forum.
As both of your machines are less than 15% CPU Busy (User + Sys), this does not look like particularily overworked systems.

If you are asking: What is a typical number for IO and User+Sys?

• Then the answer is: Every machine is different.

If you are saying these machines are running identical transactions and why are the numbers different?

• Then the answer is: They are not, in fact, identical either the workload is different or the setup is different and there is not enough information here to decide.

Your question implies that you think there should be an amount of CPU time (which you call CPU IO but I would call CPU Utilisation Wait for I/O) to drive the I/O per second. This would highlight that you do not understand what CPU Utilisation Wait for I/O means. And you are not alone in this. The CPU does not spend any real time "doing the I/O". The CPU runs the device driver which requests the adapter to do the I/O and immediately returns. Device drivers are very small in code terms and very quick to run. Then the AIX (same is true for any UNIX) kernel will use the CPU for other useful work until the adapter signals that is is finished the I/O with an interrupt. At this point the kernel runs the device driver again to deal with the data or request the next I/O. So Wait for I/O is not time on the CPU running the I/O - it simply means the CPU is actually idle (or it would have been reported as User or System time) but it is not reported as idle because the kernel knows there is outstanding disk I/O. This is used to highlight that the CPU is idle but waiting for the I/O to return. This can help in small system to highlight poor disk setup but on larger system high Wiat for I/O is often just the result of the workload - if you process the data quicker on the CPU than it takes for the data to arrive you get high Wait for I/O times - the SAP RDBMS is the classic example.

All the above is basic UNIX performance tuning stuff that can be found in Performance Tuning courses, the AIX Performance tuning manual and the IBM Redbooks.

I hope this helps, N
#AIX-Forum