Help for disk activity

[Juan Francisco Esteve_1]

I'm monitoring with sar disk activity (among other indicators).

I obtain an "extrange" situation:

Some disks (that use O.S. cache), have got the following perf indicators:

% Busy: 3-4%
Average number of requests in Queue: 30-40
Average Service Time: 1 msec
Average Wait Time: 1-2 msec.

Also, to complete the analysis, the file access system routines is huge (> 90000) during this period. May be the system is working (how?) with directories that contains a huge number of files.

According to another message in this forum, the percentage of time that queue is inspected, and it has requests enqueued is just 3-4%. But for the cases that there are enqueued requests, it seems that there is an average number of enqueued requests of 30 - 40.
Finally, the service time is good in abosolute terms: 1 msec, but as the wait time is greather than service time (1-2 msec), this is not good.

Please, any suggestion to understant these data??

Thanks in advance.

[Steven E. Protter]

Shalom Juan,

I'd ask what kind of data this is. If its a database on raid 5 storage and writes are heavy, your situation is actually expected.

Database writes to raid 5 multiplies the actual number of disk writes that occur. I would imagine such a situation if an OLTP system were running heavy writes on storage that is not appropriate for that scenario.

Questions that would help analysis:
1) Is it a database?
2) What kind of storage layout is there?
3) Server type.
4) OS version
5) Patch level especially what is the last quarterly patch install. I/O hang patches would also be helpful to know if they have recently been gathered or installed.
6) More complete statistical analysis with some background scripts: http://www.hpux.ws/system.perf.sh

I look forward to the answers.

SEP

[Juan Francisco Esteve_1]

Gracias amigo Stephen (I'm Spanish and by the momment it remains catholic and not muslim :) )

these data is not a DB. The box has got also DB devices, but their behaviour are totally diferent (avoid O.S. cache, they're in RAID 5, etc..)

I think that the disks I'm tr to see activity has being writed to files, and for any reason, the directories contains a huge number of files.

Anyway, what was a litle bit unusual for me is that %busy is not high, but avque is high and avwait is bigger than avservice.

Any other has experienced this?

Juan