Glance (block on ) / glance (stop reason )

[David Fosgate_2]

I have been reviewing these two measurement tools and found one area that appears to be a bottleneck for client applications. Our application environment is HP UX running Sybase ASE. Our thin client application is written in Cognos Axiant.
When I receive calls indicating application performance problems I find application processes stop reason in SOCKT (Waiting for a socket operation to complete). Review of the performance tool help documentation is very limited. What is this metric reporting, a TCP/IP bottleneck ?, Unix Kernel bottleneck ?, Database Service bottleneck ?, etc. Any ideas on the tool to furhter define this wait state?

Thanks in advance for your thoughts

Dave

[Rhonda Thorne]

David:

Sockets and mbufs go hand in hand as kernel parameters handling memory for networking. I will look through my notes and see if I can find out more for ya. I will monitor this post and reply if I cna find it.

Rhonda

[David Fosgate_2]

Thanks Rhonda,

With the socket status being on of the most frequesnt application stop reasons it appears to be a bottleneck. I have not found any information regarding possible changes to effect this status. Network is running at
100 Mbps, it appears that the sockets are waiting for sybase database to take their request. Again I just want to know what HP is monitoring when displaying this block on state. I'm working from the sybase side to determine how it handles the many socket requests waiting for access.


Dave

[Jim Welch]

I believe that is just a wait reason read from the kernel. You may need to trace the process to see what it is doing...trace or tusc may be useful...

[Rhonda Thorne]

David:

Have you recently change kernel params of nproc, nfiles, niodes, maxuproc?

Rhonda

[David Fosgate_2]

Hi Rhonda,

The issue has been on going, system values have remained unchanged. As our customer base grows, performance issues have become more important. To continue to deliver service levels our customers expect we are reviewing any and all potential bottlenecks on the system.

Thanks,

Dave