system performance question

[mike worrell]

I have a box hpux 11.0 running network node manager 6.3.1. This is all that is on it and when i do a top i get the following info:
load user nice SYS idle
4.01 64 0 42 5

and no processes exceed 10% ... where is all my power going so i'm only left with 5% idle.

Anyone know anything I can check or parameter I can change to make things better.

Thanks,
Mike

[mike worrell]

also i just did a dmesg and got this message
"file table is full"
What is that coming from and how do I fix?
Not in syslog.log

[Mel Burslan · Accepted solution]

From my past experience, the NNM application, despite how demanding it is, usually gets stuck on a single cpu, low memory, old architecture system. In my case, when T500s were the top of the line, my whole IT/O architecture was stuck on a 715/100 workstation including but not limited to NNM.

Looking at your numbers, I am almost going to say the same thing as you did not indicat what class of CPU NNM is running on. According to the numbers 64% of your CPU cycles are hogged by user processes, which are, again according to what you said, is HP OpenView related processes as there is nothing else significant running on the system.

Also looking at the system process percentage, which is 42%, I can say that your cpu is cracking under pressure, unless this is a severe misconfiguration case.

My number from an L3000-5x system are like
CPU LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS
0 0.20 0.6% 0.0% 0.2% 99.2% 0.0% 0.0% 0.0% 0.0%
1 0.11 0.2% 0.0% 0.2% 99.6% 0.0% 0.0% 0.0% 0.0%
2 0.13 0.4% 0.0% 3.6% 96.0% 0.0% 0.0% 0.0% 0.0%
3 0.12 0.6% 0.0% 2.8% 96.6% 0.0% 0.0% 0.0% 0.0%
--- ---- ----- ----- ----- ----- ----- ----- ----- -----
avg 0.14 0.4% 0.0% 1.6% 98.0% 0.0% 0.0% 0.0% 0.0%

and this server is running an oracle db, an apache web server and actively used by sysadmins as the main management node.
But it has 4G of memory installed.

without that model, memory etc. info, it is hard to say what the reason is.

[Mel Burslan]

file table is full in dmesg if it was seen once and your server is up for a long time is not something to worry about too much but the next time you need to schedule down time for the machine, you may want to bump up the kernel parm nfile about 1/3 of its current value.
if your machine and kernel is in a pretty recently installed state, nfile must be a function of maxusers parameter, so instead of breaking the formulas, you may want to bump up the maxusers by 1/3 maybe.

[Steven E. Protter]

Please read this:
http://www1.itrc.hp.com/service/cki/search.do?category=c0&docType=Security&docType=Patch&docType=EngineerNotes&docType=BugReports&docType=Hardware&docType=ReferenceMaterials&docType=ThirdParty&searchString=UPERFKBAN00000726&search.y=8&search.x=28&mode=id&admit=-682735245+1086815236399+28353475&searchCrit=allwords

I suggest collecting some perf data. These scripts are based on the ones hp uses. See attachment.

SEP

[John Dvorchak]

I looks to me like you an unusually high system number of 42 percent. System should normally be down around 1 to 4 percent not 40 percent. Does this system have a lot of NFS exports or a lot of people logging in as the same userid?

Do you have glance installed? If so what does glance -t look like? See if you can post a screen shot of glance -t.

[rick jones]

NNM - probably means a non-trivial quantity of networking happening, which could result in user CPU time being lower than kernel/system CPU time. Particularly as the number of managed systems increases.

With something like glance or measureware or lanadmin check the packet rates.

If the NNM system is polling other systems for stats, you might increase the time between polls.

[KapilRaj]

CPU idle time "0" is not a problem always. If CPU is free processes would use it. Do you have any performance issues with it ?. Something like vmstat shows more processes in runq >?.

I would like to have a look at ur vmstat 2 o/p

Regds,

Kaps

[mike worrell]

Thanks everyone. Forgot to mention last week I installed another networking product called optivity and that's when my trouble started I think. before that Network Node manager ran fine and was quick and my #'s from Top looked good. I have 1 cpu 180mhz
2.0 PA8000
One of the errors I got installing another network product dealt with the kernel parm nfile needing to be increased. Could my problem be stemming from nfile not being large enough.. When I increased one time it took care of an error I was getting.
Thanks

[rick jones]

nfile being too small would either lead to things not working at all, or perhaps things cycling through file descriptors more often than they would otherwise - closing files sooner and perhaps re-opening them later.

that increase in open and close could I suppose lead to an increase in system CPU time - does something like glance show that all the nfile entries are indeed in use?