Disk performance - how to find bottleneck

[Rajasekhar Raman]

This a continuation of an earlier question I had asked.

I have two systems, both are primarily Sybase servers (HP-UX 11.00, patched to same level, one has JFS 3.3 (machine A) and the other JFS 3.1 (machine B), same amount of RAM and similar disk configuration)

I have the sar -b output for loading a 21 GB database (same database) which is stored locally on each machine.

The performance is markedly different. On machine A, it takes about 75 minutes to load the database, and on machine B about 20 minutes. Can anyone tell me how to diagnose this problem?? It will be huge help!!

I have attached the averaged "sar -b" output for the the two machines.

[Uday_S_Ankolekar]

Use Glance to anlalyze this. ( if you don;t have glance installed, download 60 days trial or install itfrom the application CD)
From Glance you can check realtime status of individual server for any bottlenecks ,either disks or, network, or memory or any process.

-USA..

[Rajasekhar Raman]

What should I be looking for after running Glance ? The difference in performance exists only for database load. There is not much difference in the database performance, atleast not noticeable, so I was wondering if it may have to do only with filesystem.

Also, vmstat shows abnormally high %wio (greater that 60%) on machine A ( the slower machine)

-Shekar

[Uday_S_Ankolekar]

Check for possible memory leaks, process that growing or using lots of CPU etc..

USE GPM ( gui version on Glance) That will be more helpful..

-USA..

[V.Tamilvanan]

Hi,
The %wio is the amount of time cpu waits idle with some process waiting for I/O. Normally it shouldn't be that high (60%).This situation occurs if there is too much swapping or there is a problem with one of your disk. Check by dmesg and the syslog files for any I/O error.
Try by Applying the latest JFS patch.