vxfsd uses 90% CPU

[martino]

Hello HP guru's!

I noticed that on our RP7410 HP11.11 server, with 12GB of ram and 4 CPU's, the vxfsd is using almost 90% of CPU with nothing running at all. (an average of 32% CPU usage!!)

On the server we have a tiny Oracle 9.2 database installed but there aren't many users connecting, and the database is not havily being used.

I have the December 2003 patch bundle installed on the server and have tuned dbc_max_pct to 7 and dbc_min_pct to 2.

All the other kernel settings seem to be correct to.

I checked with sar, glance, top and perfview to get a good overview.

Does anyone have an idea to what to look at/adjust/modify?

Any specific patches to install or kernel settings to adjust? I phoned HP, but they came up with the answer to install the december 2003 patch bundles....hmm...duh?

Thanks in advance!

[Massimo Bianchi]

Hi,
maybe ou are facing a problem with the inode cache.


Please check ninode and vx_ninode

If they are 0, meaning that they are controlled by OS, and you can allow a reboot, change these accordingly:

ninode: 30% of nfile
vx_ninode: 80% of ninode

both these parameters are only CACHE, and from time to times they lead to high cpu usage and memory leak.

Havng nindo at 100% in sar is aboslutel normal. I always try to keep them low.

Above formulas are my own rules of thumb.


Also: be sure to have the pastest vx patches, they help a lot!

HTH,
Massimo

[RAC_1]

I agree with Massimo to some extent. ninode applies only to hfs file systems. And on most of the hp-ux systems the only fs that is hfs is /stand. So setting ninode should be easy.

Real time usage can be seen with
glance -t
sar -v 3 3

About vx_ninode, I prefer to set it to 0. So that OS takes care of it. And if I remember correctly, in case you want to tune it, you must have a patch.(I do not recall which patch was this)

What is your setting for vx_ninode???

Anil

[martino]

I have the following kernel settings:

ninode :72864 - (8*NPROC+2048)
vx_ninode : 0 - 0
nproc : 8852 - (20+8*MAXUSERS)
nfile : 21987 - (16*(NPROC+16+MAXUSERS)/10+32+2*(NPTY+NSTRPTY+NSTRTEL))

[Rgomes · Accepted solution]

Hi,

Problem should be fixed by the following patch:

http://www2.itrc.hp.com/service/patch/patchDetail.do?BC=patch.breadcrumb.main|patch.breadcrumb.search|&patchid=PHKL_29688&context=hpux:800:11:11

****. VxFS may create new threads unnecessarily
even when enough flushing threads are already
present. It results in increased VxFS flushing
threads which may block creation of
fdd/DMAPI threads and
increased CPU/memory consumption.
****

Regards,
Richard