Performance issue - Blocked on "System"

[Janne_5]

We are investigating complaints on performance in a distributed system (Oracle 9 application servers, Oracle 9 database servers all running on PA-RISC hp-ux 11.11).

I can't find anything wrong on the operating system/server level besides one thing which I don't know how to interpret.

When looking in glance/gpm on a process wait states, I can see that quite many processes are blocked on system ~80% of time. On global level I got 40%/600 threads.

See
gpm -> Process List -> Reports -> Process Wait States
and
gpm -> Reports -> Wait States

Glance help tells that SYSTM is LVM, VFS, UFS, JFS and Disk Quota subsystems.

Does anyone know if this affects performace negatively in a significant way? How can I tune it?

[Shahul]

Hi,

In any busy system, if you go to glance, you can see processes blocked on many things, eg. PIPE, NFS, STRMS, CACHE, MESG, daid, etc. You need to look at each process waiting for what? it is not an easy task, if they waiting for Cache, then it could be something to do with Data Buffer cache, if it is waiting for NFS, then there could be a Network bottleneck. So, you need to investigate the reason why they are waiting.

Good luck
Shahul

[Emil Velez]

Blocked on system means that the process is waiting on updating or accessing some system resource. It might be allocating space on a disk, paging, LVM subsystem. Something not covered by priority (other processes)

[Wim Rombauts]

What have you been using to conclude that you don't see anything at the system level ?
If you can tell us what tools you used, and what numbers you have seen, some of us may fill the gaps or point you at something you missed.
Sometimes, as I learned, you just have to look harder.
Have you tried "sar" for instance ? With "sar -u 60 120" you will have a 2 hour log with 60 second intervall showing you CPU activity, including the amount of time the system is waiting for IO to complete.
It could be that your CPU usage is low, but that your system is continuously waiting for IO.
But of course, maybe you have already looked at this.