Vmstat showing very high blocked processes on HP-UX server

[amreek bansal]

Hi Guys,

Can someone kindly lend some advice. On our HP-UX database server, users were reporting slow DB connections/ slow performance. Mmeory ise is has enough, its not shortage, nor cpu is geting maxed out. So I checked the output of vmstat which is showing a very high number of blocked processes i.e consistently over 30. So theres seems some I/O problem, can any one tell me how to investigate this further and to resolve this ?

[Turgay Cavdar]

Hi,
If you are sure there is no cpu/ram issue then check IO usage with sar or glance.
With sar especially check the service time (avserv) and queue(avque) values, then you can see if there is any IO contention.

[amreek bansal]

ok, with sar, what are the switches to use to find out this information ?

Thanks

[R.K. #]

Hi Amreek..

Your system is responding slow..is this only for some specific file sys or across the whole system.

Following will help us find if there are any disk I/O related issues. take this when you find the server responding slow:
sar -d 10 5

When average wait (avwait) is greater than average service time (avserv) it indicates the disk can't keep up with the load during that sample. When the average queue length exceeds the norm of .50 it is an indication of jobs stacking up.

[amreek bansal]

Hi Guys, After some days of monitoring my HP_UX database server, I have a few queries regarding thresholds. From the sar output's. can someone please tell me what should be the normal threshold level for %wio ? In busy times, it averages around 30%, is this over normal ?

Thanks.

[R.K. #]

Hello Amreek,

%wio - idle with some process waiting for I/O

Typically what I think is %wio > 15 means there is a bottleneck somewhere, may be disk related.

Have you checked "sar -d" to see if there is no disk bottleneck.

Disk bottleneck is indicated by:
%busy > 50 (also 30 for some systems)
avwait > avserv

[Turgay Cavdar]

Hi,
just lookıng at %wio, you cannot simply say that "there is an IO bottleneck".If you suspect IO bottleneck, please check disk statistics with sar -d and glance. Tell us about disk's service time and queues and also your storage config...

[Pinhead]

We are seeing something similar. We are intermittently seeing slow DB connections (via the listener only). When we see this condition, we also see many (80-150) blocked OS processes via vmstat. Did you ever resolve this problem?

[Michael Steele_2]

Hi

The %wio is the percentage of time the CPU spent blocked on IO. This only includes block IO, raw IO, or Virtual Memory operations like pagein/swapin. It does not include time spent on tape IO or terminal IO. If %wio is high, it could be an indication of a disk bottleneck or it could simply mean that the CPU has nothing else to do while it waits for a process to complete its IOs.


'iostat' is not a reliable utility and hasn't been since 9.0. Use UNIX95 or glance.