Itanium 2 ( Hpux 11.22 ) process vhand ...

[Andreas Siedler]

Hi there,
we have the problem ( on 21 machines ), if Oracle runs, the 'find' - command produces a high performance problem ( vhand runs with 20% CPU-Load and more ).
swapinfo shows memeyory PCT used over 98%, when the problem occurs.
Patch PHKL_28670 memory pressure with JFS3.3 is already installed.
Does anybody know what to do?

Andy Siedler

[Stefan Farrelly]

Take a look in the patch database for PHKL_21678. There is no equivalent for 11.22 but either HP may be working on one or you dont need a patch - you just need to tune disksort_seconds. Read the description for the patch - its to fix a find problem which causes the symptoms you see.

[Kent Ostby]

I verified that the "disksort_seconds" feature made it into the 11i code path in 11.11 so it should be in 11.22 as well.

As the other poster noted, you would need to set the tunable parameter disksort_seconds to a value of :

1 2 4 8 16 32 64 128 or 256 .

This is the number of seconds it will wait before it goes and handles all requests that have come in during that timeframe.

I would set it HIGH (256) and move it down until you get the performance you are looking for.

By setting it to 256 seconds, you are saying to do best disk access for up to 256 seconds then take care of the old requests.

Hope this helps,

Best regards,

Kent M. Ostby

[Andreas Siedler]

many thanks for your hints, but in didn't really help. It looks and feels a little better than before, bitte vhand still need form 7 to 11 % of CPU-time when starting 'find'

But: vhand comes to end, when I setdisksort_seconds to 128 .

THANK YOU
Andy

[James Murtagh]

Hi Andy,

Can you provide a bit more info please? You say when Oracle is running, a find command will cause the system to start paging? What are the swapinfo and free pages (vmstat's free column) before and after the find command? The find command will cause the OS to populate the inode and buffer caches so this may push your memory usage higher over the paging threshold, hence vhand coming into play. When this happens, as you no doubt well know, system performance is very poor is most cases. To find the buffer cache usage:

# echo bufpages/D|adb -k /stand/vmunix /dev/kmem

Again. before and after totals would be useful.

cheers,

James.

[Andreas Siedler]

Hi,
here some more info:
RX2600 99MHz, 1GB Memory

before 'find'

swapinfo:
Kb Kb Kb PCT
TYPE AVAIL USED FREE USED
dev 2097152 142020 1955132 7%
reserve - 866000 -866000
memory 667512 637052 30460 95%

vmstat free 1403

during 'find':

swapinfo did not change
vmstat free 777

after 'find' the system returns to state "before". I would like to test dbc_max_pct = 10 and disksort_seconds=128 on machine with appr. 10 oracle-users the next days. I will give a reslult message.

Greetings
Andy

[Bernhard Füß]

Hi Admins,

is the problem solved with setting "disksort_seconds" to 128?

maybe i have the same problem...

Regads