Question about CPU Switchover

[Ed Hudak]

I'm trying to find out about a problem we had at work. We had a old remote startx process (root) tying up 1 processor (97%) of a 2 CPU PA-RISC older Rp5340 11.11 class server(2 gigs ram)without any additional software for CPU management. A user was submitting a query into a database that generates a Lawson report to his screen (from a PC using a Terminal session). The report was seen in the Lawson Job Queue was stuck in a wait/processing mode. My question is with one processor being tied up by the startx process did it cause the problem with the Lawson query? Or how much can a runaway process actually tie up a CPU and does it allow switchover to the other CPU?

I'm not sure if 1 problem process can cause this much trouble??? Other users were trying to generate reports but were blocked the way this job queue was setup (jobs waiting on queue).

[Johnson Punniyalingam]

very pity hard to find "that because of that said process" which you are mentioning above post .<<< causing the CPU bottle neck..


You run some commands to check real load of your server.

glance
iostat -xtc 5 2
vmstat 5
uptime

it also depends , kernel parameter settings

[Ed Hudak]

I guess my question is if a process causes 1 cpu(only 1) to have 97% utilization (when I killed the process the cpu utilization went down to a low percentage) how does it affect the other cpu? Does it see the bottleneck and switch over to the other cpu (totally), and still allow process's to perform at a reduced performance??

[Rita C Workman]

I doubt that the one processor process (startx) had anything to do with your Lawson query.

More than likely your Lawson query had it's own issue and hung. Depending on the Lawson application, the queue may act in a FIFO methodology (first in first out), so if the guy at the front of the line is hung....everyone behind him sits and waits.

Talk to the Lawson Administrator and see if that person, or a DBA can kill the hung/stuck query. If the others begin to run...then that's the answer.
Someone should however identify the query/report request that caused the issue and fix it, prior to killing it. Or you're just going to hit this again.

Just a thought,
Rita

[Ed Hudak]

What happened was that moving the job to another queue allowed all jobs to finish normally(you were right about this FIFO queue). After deleting the job the user performed the same query/report and it finished without problems(without the CPU consuming process in the picture). Some IT people here think that the problem process caused the Lawson problem. I'm trying to find out more about how a 2 CPU HP-UX handles problems like this. I searched online for HP documentation about CPU switchover without any luck. Also how critical is a 97% 1 CPU utilization with the other CPU idling?

[Dennis Handly]

>how critical is a 97% 1 CPU utilization with the other CPU idling?

This isn't a problem. Why are you worried?
Unless you use threads, a process can only use one CPU at a time.