Hi,

I am in midst of benchmarking one of our product on HPUX 1.1.11
(9000/800), 5G RAM, 60% empty disk space. Unfortunately, I am getting
quite 'random' readings - using timex.

real 17.21
real 9.36
real 8.28
>real 30.72

real 9.18
>real 25.61

real 7.37
real 7.32
real 8.97
real 9.14
real 7.52
real 7.31
real 7.28
real 8.60
>real 23.96


Here is few characteristics of the application that I am benchmarking:
1) Disk IO Intensive - read/write vxfs v3.3 (The file to read is just
about 530MB).
2) No network or other dependencies.

As for the machine, there are plenty of free memory for system file
caching:
# swapinfo -t
Kb Kb Kb PCT START/ Kb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 8192000 0 8192000 0% 0 - 1
/dev/vg00/lvol2
reserve - 116648 -116648
memory 3971500 1470752 2500748 37%
total 12163500 1587400 10576100 13% - 0 -

And I made sure the server is idle while running this test.

Do you have any pointers on why the machine exhibits such behaviours?
(there are few subsequent runs that about 7-9 secs - which I believe
good proof its able to cached the file IO, but why does it suddenly
goes up to 20+ secs?, infact even 7-9 secs not seem to be consistent).
Is there any fix for this? (Basically some tuning to make it for
consistent readings)

Thanks and regards,
Ravi Nallappan