Need help with swap usage on HPUX 11.23 1

[sbrews]

I have been away from HPUX for a very long time...

I have a system (11.23) that appears to be having high swap issues. What are some things I can do to determine what is causing the high swap usage and potentially fix?



Add a little color to your PUTTY terminal: faq52-6627

 

[Annihilannic]

For reference, what is the output of swapinfo -tm?

HP-UX is very "proactive" in the way it preallocates virtual memory... or you could call it pessimistic. Usually when swap is high you aren't at risk of running out for existing applications, however it may prevent spawning new processes.

Sometimes it's simplest just to add more swap devices, even though the existing ones may not be heavily used; we waste quite a lot of disk space this way.

To identify the largest applications in memory you can use glance (if licensed), 'o' to change threshold options, enter a few times down to sort order and type 'rss', then 'y' to accept. Note this will change the defaults, you'll need to do the same and change it back to 'cpu' if that's the way you prefer to see things. Yes, it's dumb.

Or, as I prefer, you can use the following to list processes with the largest virtual sizes:

# UNIX95= ps -eo user,pid,vsz,args | sort -rn -k 3,3  | head

Annihilannic
[small]tgmlify - code syntax highlighting for your tek-tips posts[/small]

 

[sbrews]

Glance output:
PROCESS LIST Users= 2
User CPU Util Cum Disk Thd
Process Name PID PPID Pri Name ( 200% max) CPU IO Rate RSS Cnt
--------------------------------------------------------------------------------
perfalarm 1752 1 154 root 0.0/ 0.3 58420.8 0.0/ 0.0 178.3mb 2
cimprovagt 22423 22402 168 root 0.0/ 0.2 308.3 0.0/ 0.0 98.0mb 35
tnslsnr 7592 1 154 oracle 0.0/ 0.7 12041.1 0.0/ 0.1 96.1mb 1
emagent 4977 4901 168 oracle 0.0/ 0.3 598.9 0.4/ 0.2 89.0mb 6
oracleyml 10220 1 154 oracle 0.0/ 0.1 0.1 0.0/ 0.0 88.9mb 1
oracleyml 9637 1 154 oracle 0.0/ 0.1 0.1 0.0/ 0.0 88.9mb 1
oracleyml 10155 1 154 oracle 0.0/ 0.9 0.7 0.0/ 0.0 88.9mb 1
oracleyml 9643 1 154 oracle 0.0/ 0.1 0.1 0.0/ 0.0 88.9mb 1
oracleyml 10228 1 154 oracle 0.0/ 0.1 0.1 0.0/ 0.0 88.9mb 1
oracleyml 7957 1 154 oracle 0.0/ 0.0 0.1 0.0/ 0.0 88.2mb 1
cmcld 7435 7429 -14 root 0.7/ 0.5 100064 0.0/ 0.0 42.6mb 14
oraclesuniv 5073 1 154 oracle 0.0/ 0.2 286.3 0.0/ 0.0 41.0mb 1
...

The output from the ps is:

--> UNIX95= ps -eo user,pid,vsz,args | sort -rn -k 3,3 | head
root 1752 906824 /opt/perf/bin/perfalarm
root 1770 456688 /opt/OV/lbin/perf/coda
oracle 5073 164352 oraclesuniv (LOCAL=NO)
oracle 25025 147904 oracleyml (LOCAL=NO)
oracle 16569 143872 oracleyml (LOCAL=NO)
oracle 8282 143872 oracleyml (LOCAL=NO)
oracle 1195 143872 oracleyml (LOCAL=NO)
oracle 24901 143808 oracleyml (LOCAL=NO)
oracle 24817 143808 oracleyml (LOCAL=NO)
oracle 24440 143808 oracleyml (LOCAL=NO)

I am not familiar with the perfalarm stuff - so I dont know if those are normal or not. there are also a ton of oracle processes - which may be part of the issue.


Add a little color to your PUTTY terminal: faq52-6627

 

[Annihilannic]

You missed the swapinfo -tm output?

That's a pretty huge memory size for perfalarm (a part of HP's MeasureWare), and I presume there isn't a vast amount of memory in this system since it only has 2 CPUs. coda is also part of HP's performance monitoring stuff. I'd recommend restarting it using /sbin/init.d/ovpa stop ; /sbin/init.d/ovpa start, should do no harm to any other running services.

If that seems to be the problem, and the issue reappears, maybe you need to install some MeasureWare updates; we have a selection of versions C.04.60.000, C.04.70.000 and C.05.00.000 on our 11.23 systems. Use swlist MeasureWare to check yours.

Oracle will use as much memory as it uses; not much you can do there, but many of those processes will be using some shared memory so the total memory usage may be misleading (i.e. you can't just add them all up).

Annihilannic
[small]tgmlify - code syntax highlighting for your tek-tips posts[/small]

 

[sbrews]

Hmm... not sure how I forgot the swapinfo stuff. I have stopped/restarted ovpa - will watch over the next few hours/days and see how the system behaves.

Add a little color to your PUTTY terminal: faq52-6627