Telnet hanging while 2 or more gunzips running

[Daniel Shaha]

Bare with me on this, I am a new Unix Admin.
We have an N-class with 4 cpus and 8GB memory. There are 10 72GB external drives that are striped and mirrored and 2 internal drives that are mirrored only.
Currently there are 13 Oracle databases on the machine. When someone starts up 2 or more gunzip processes telnet processes seem to hang until one of the gunzips are done. During this process the CPU's are pretty much idle(no where near 100%) and the memory usage is very low. I am looking for ideas on what may cause this problem.

[Jim Turner]

Daniel,

I'd prefer to keep my clothes on at this time (bare vs. bear), but I will try to help you with your poser

Do you get any "cannot fork" errors on your console or in syslog when telnet stops responding ? If you made no changes to your kernel, chances are you're soaking up all of your proc table entries with the 13 Oracle instances and other various and sundry processes that are running. You can increase nproc if needed.

Cheers,
Jim

[Daniel Shaha]

If I am looking in the correct place(/stand/build), the nproc is set to 20+8*Maxusers(2048). When I do a ps -ef |wc, there are only 345 processes on the machine. This is a test server that we are just moving/creating the test databases on.

[Paula J Frazer-Campbell]

Hi

Run sar whilst you run multiple gunzips.

sar -A 1 120 > sarfile

Your answer or a pointer to it should be in that file.

Also look at swapinfo -a
and top -q


HTH

Paula

[Daniel Shaha]

Jim, I fogot to mention there are no forking issues listed in syslog.
Enclosed is the output from the sar report Paula mentioned. The only thing that looks a little high is the %wio.
Another thing I forgot to mention is the gunzip source file is on one of the internal drives and the destination is to the striped/mirrored external set.

Thanks
DRS

[tech1214]

Some versions of HPUX 11 have a bug with the tty drivers where anything that does TTY i/o is moved to the monarch processor (cpu 0). The standard symptom of this problem is to have one more cpu-intensive programs virtually monopolize cpu 0, bringing all terminal i/o - telnet or rlogin, but not serial port, to a halt.

You can check if this is the case for you by running your gzip's such that they don't do any terminal i/o, like this (in csh):

(gzip < foo > foo.gz ) >& /dev/null

and then checking to see if your telnet connections continue to hang or not.

If that is the problem, you should upgrade to the latest release of HPUX. I don't know which patch fixes it, but I think that at least the june 2000 patch bundle contains the fix.

[Shahul]

Hi

I can say this much only. U will have to tune ur kernel. For fine tuning U see HP Manuals or there may be some pre-prepared tunable parameter settings for ur gunzip. Try it out.


Best of luck
Shahul