HPUX Patch for NFS Causes Db Issues on NetApp Filer
- From: "Post, Ethan" <Ethan.Post@xxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Wed, 27 Oct 2004 14:37:10 -0500
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
After applying some patches to an HPUX box the problems below began to
occur. This is an HP box attached to NetApp filer, anyone ever seen
this?
Applied HPUX patches:
1) PHNE_30398
address slow NFS performance
Changed kernel parameter "nfs_new_lock_code" to activate bug fix
JAGad15675=20
Changed kernel parameter "async_read_avoidance_enabled" to activate bug
fix JAGad72416=20
2) PHKL_30378
Put on because PHNE_30398 depended on it
No other changes=20
Hardware Info
Class: HP L3000
Model: HP 9000/800
OS: HPUX 11.11 v1
Oracle: 8.1.7
Netapp filer: FAS940 (connected to the database server using a Gb
network)
This happened...(see screenshots http://poststop.net/temp.htm)
Abnormal buffer busy and enqueue waits.
DBA noted...
I did notice that our waits were mostly "undo header" waits instead of
our normal "data block" waits. This points to issues with the rollback
segments. We also had a number of checkpoints that did not complete. It
was every interesting, to watch the graphs of wait events via Spotlight.
Over a 3-5 minute period, the wait events (mostly free buffer and buffer
cache) would balloon from 5000 ms to 140,000 ms, and then all of a
sudden drop to 5000 ms in an instant. This would be repeated about every
7-10 minutes.=20
It was almost as if the db writer couldn't flush dirty block fast enough
to keep up with demand for free blocks in the buffer cache, the waits
would build and then all of a sudden, a hugh flush would occur.=20
--
//www.freelists.org/webpage/oracle-l
Other related posts:
- » HPUX Patch for NFS Causes Db Issues on NetApp Filer
