CPU Utilization spikes

[Srikanth Arunachalam]

Hi,

We are in the process of merging multiple site database onto single larger database. We want to confirm few points on the % CPU utilization spikes that we have currently after the first site merger. We want to have a proper understanding before we merge 8 more sites.

The CPU% utilization after the first merger on average was 20%-30% and a single large spike of 90% at mid-night.

When I say CPU %utilization spike, I mean to say a sharp increase from 10% to 90% during the first second and a sharp dcrease from 90% to 10% during the second seconds.

On the second day after the first merger, the CPU% utilization is 10% on an average and 50% during 06:00, 08:00 and 10:00. It has been 75% during 02:00 (where it was 90% the previous day).

We receive peak automatic jobs during 02:00, 06:00, 08:00 and hence that shows the trend in the CPU % utilization graph.

We are using HP-UX 11.23 with 8 CPU processors and 64GB of physical memory.

(1) I would like to understand if such spikes is something to be analyzed or can we assume that spike for a few seconds can be safely ignored ?

(2) What performance analysis need to be done at this stage ?

Thanks,
Srikanth

[whiteknight · Accepted solution]

Hi Srikanth,

1)I would like to understand if such spikes is something to be analyzed or can we assume that spike for a few seconds can be safely ignored ?

a few spikes may not be a concern you can safely ignored, unless you are seeing constantly 90%, then you need to watch out.

My advise :- use Glance tool to monitor your CPU utilization.

2) What performance analysis need to be done at this stage ?
- Using Glance to monitor, also good to have your measureware running all the time to capture the CPU utilization before and after the merger.

[Yogeeraj_1]

Hi Srikanth,

If you are running an Oracle Database, you can run the Database control (Oracle Enterprise Manager 10g) and verify the Host Performance (or even run the ADDM).

There could also maintenance tasks (database level) than also impacting on this.

please check

hope this helps!

kind regards
yogeeraj

[Srikanth Arunachalam]

Hi ,

Thanks everybody for your reply. We do have monitoring tool for database and OS. I was trying to find out what decision needs to be taken further. As confirmed spikes for a few seconds need not be of a concern.

Thanks,
Srikanth

[Bill Hassell]

The spikes are meaningless until you attribute the increase usage to a specific process or application. The following 4 line script will create 100% CPU usage for every invocation:

while :
do
:
done

Is this bad? Not really. The shell is doing what it was told to do. Same with the processes that consume a lot of CPU. As a system administrator, you can report the change in CPU usage to the people responsible for the application but there isn't anything else for a system administrator to do.

Note that if you run 8 copies of the above script, you will have 100% CPU usage but it will have little effect on your activities. The reason is that HP-UX (like most flavors of Unix) will see long-term compute-bound processes as not very important and will degrade their priority rapidly until they only run when there are some free processing cycles.

A database application is very different. Every time I/O is requested, that process is bumped up to a higher priority. So even when a CPU hog seems to take all the compute cycles, the applications with some I/O will still get top priority.

Now a well-designed database will trade CPU cycles for disk I/O meaning that disk is very slow compared to memory. So if a lot of records are being cached and a large query or report is run, rather than waiting for a long time to read all the records, a cached set of records can be queried instantly from memory and the report is completed very quickly. So for a short time, the application will have high CPU usage -- which is a very good thing.

[Srikanth Arunachalam]

Thanks for all those replied.