Performance problems

[Jani Kuituniemi]

Hey again,

Any tips for where to look what is the cause of causing very high load averages (>20-30)? I tried "sar", and it says that wio% is well under 5, so I think that means that the RAID discs are not the thing. Concurrent user count is around 100, most running pine or imapd from webmail machine (other machine) There's plenty of free discspace and if I read dmon's output correctly, there's lots of free memory too. What the machine runs is:
Oracle DB, but shutting it down doesn't affect the loads, so I suspect this is not the problem. Apache webserver, this causes some load, but cannot be the whole cause. NFS exports to few other machines, but this would show in 'sar' right? and it is a NIS master. Any thoughts?

//Jani

[Dan Hetzel]

Hi Jani,

Did you install the latest ONC/NF General Release/Performance patch?
This would help as your server is exporting some NFS filesystems

Here is the link for HP-UX 11.0:
ftp://ftp.itrc.hp.com/hp-ux_patches/s700_800/11.X/PHNE_22125

Best regards,

Dan

[Jani Kuituniemi]

Hey,

No, I don't think I have PHNE_22125 installed. I just fetched it and will install it shortly as soon as there's a good place to do it, as this is a 24/7 production server.

//Jani

[Wodisch]

Hello Jani,

have you increased the number of "NFSIOD" processes
in the "/etc/rc.config.d/nfsconf"? 4 ist by far to small
for any NFS server machine!

And use "nfsstat" and Hal Stern's book about NFS
(get it form O'Reilly).

HTH,
Wodisch

[Bill Hassell]

If you don't have Glance, run top and see what processes are consuming CPU time. Don't overlook the web server. Badly written web pages can clobber a web server and also generate massive LAN traffic.

The uptime load value (also shown by top) represents the number of processes in the run queue and not disk load or CPU load. By itself, a high run queue is not always a bad thing but could be a number of processes that are doing very fast polling which might not be a good idea. Rewriting the polling program(s) or changing the poll rate can help a lot.

[Ron Klerks]

For NFS look at the two following kernel parameters. 1. dbc_max_pct and 2. dbc_min_pct.
Don't set them to big, default is 50. To BIG!
On our machine we had problems with nfs too.
8 Gb memory and I put max on 5 and min to 2.
Check for further info the hp.docs site and search for dbc_max_pct and try to figure out what the best is in your situation.