vhand - does it affect system performance

[Ajdin Osmanagic]

Hi,

does 'vhand' process affect system performance?
'ls' command takes few seconds to complete its mission.
My system is HP-UX 11.0, has 256 MB RAM, 2x9GB disks (on the other one, HP-UX 11.11 is installed, and doesn't have such performance troubles).

I've noticed 'vhand' process differs, on my machine it is:
TTY PID USERNAME PRI NI SIZE RES STATE TIME %WCPU %CPU COMMAND
? 2 root 128 20 0K 0K sleep 54:58 2.52 2.52 vhand

and on the other machine (also 11.0 system):
TTY PID USERNAME PRI NI SIZE RES STATE TIME %WCPU %CPU COMMAND
? 2 root 128 20 0K 0K sleep 0:17 0.02 0.02 vhand

I wonder, if 'vhand' affects the system's performance?

Best regards,
Ajdin

[harry d brown jr]

Compare your swap spaces.

BTW, 256mb of ram is very minimal, and you should purchase more.

live free or die
harry

[Stefan Farrelly]

vhand is the system swapper/pager process. If you see it using lots of cpu then you have a memory pressure problem (not enough RAM for what youre running). Check swapinfo -mt and look at the dev lines and the USED total - this is how much RAM has been paged to disk. Hopefully its at 0% which is ideal.

[Tom Maloy]

vhand does not affect system performance at all. It can be an indicator of a problem.

vhand should run rarely. In normal operations, vhand should sleep a lot. When it runs frequently, it is a symptom of "not enough swap" or RAM.

Tom

[Ajdin Osmanagic]

Hi,

I've checked swap file, there's more than enough (1GB) of swap size comparing to 256 MB of RAM. I agree, that there is a bit less RAM, than it should be.

I'm still looking for an answer, why is HP-UX 11.11 on other disk, with same system HW resources much faster?

Best reagrds,
Ajdin

[Ajdin Osmanagic]

Hi,

swapinfo -mt shows:
$ swapinfo -mt
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 1024 0 1024 0% 0 - 1 /dev/vg00/lvol2
reserve - 119 -119
memory 171 36 135 21%
total 1195 155 1040 13% - 0 -

Best regards,
Ajdin

[Tom Maloy]

I believe that the memory management model was changed in 11.

So you have two of the same hardware platform, running different versions of the OS. Are all of the applications/users/load the same?

[harry d brown jr]

Ajdin,

Do both (the 11.00 and 11i) instances have roughly the same number of processes?

Does the 11.00 machine have the latest patch bundle installed?

live free or die
harry

[harry d brown jr]

On your 11.00 instance, check these patches out:

http://support2.itrc.hp.com/service/patch/patchDetail.do?patchid=PHNE_22566&context=hpux:700:11:00

http://support2.itrc.hp.com/service/patch/patchDetail.do?patchid=PHKL_18543&context=hpux:800:11:00

http://support2.itrc.hp.com/service/patch/patchDetail.do?patchid=PHKL_25525&context=hpux:800:11:00

live free or die
harry

[Stefan Farrelly]

Weve got servers running 11.0 and 11.11 and see no difference at all in the amount of cpu vhand uses. All our servers run with <100% RAM utilization so no paging.

Yours also has no memory pressure so its not that. Are you uptodate with patches ? were running the 2 June2002 QPK bundles on 11.11 and the Mar2002 QPK bundle on 11.0

Are you apps exactly the same ? If you have one which has a lot more connects then obviously vhand will be a little busier because it has to reserve device swap each time a new process loads into memory.

[Ajdin Osmanagic]

Hi,

I do have one system (physically is other box) also running 11.0, but with 1 GB of RAM - no such problem.
On my system, there are 2 disks, with 11.0 and 11.i .

The same instances are running on both systems (and both disks at first system).

11.0 has the latest patch bundle installed, but I don't think this could be a problem, since it was happening before, and it is happening now (less than 1 year, since I've got a box).

Best regards,
Ajdin

[Wodisch_1]

Hi Ajdin,

could it be you (someone else) has changed the kernel "watermark" parameters? Like "maxdesired" and such...

Then that kernel would start eralier to page, and stop later, keeping more free pagaes at all(weel, most) times.

What does the column "po" (page out) of "vmstat" show? Is there a lot of paging-out happening?

Just an idea,
Wodisch (still on his way ;-)

[Tim D Fulford]

If you have measure ware installed (& on) there is an application group called memory_management, which includes vhand

See what they relative CPU utilisation of the two are. This may well give you a clue to if you have a memory problem or not

Tim

[Ajdin Osmanagic]

Hi,

have installed additional 512 MB of RAM. Looks good for now.

'po' was showing '21'. On the other box was '0'. I lowered quantity of RAM on another box to 512 MB, and 'po' is still showing '0'. On my box, where is currently 512 MB RAM more, 'po' is near to '32'.

Thanks for all replies.
I hope I find one day, what happened there...

Best regards,
Ajdin

[Bill Hassell]

po (page outs) is an excellent measure of swapping, although 32 isn't a lot UNLESS it is steady for long periods (minutes to hours). In that case, you have (as you've already seen) way too little RAM. Typical RAM for production systems starts at 1000 megs and goes up to 8000 megs (or more) for 64bit opsystem and 64bit programs. As always, it all depends on your applications.

If these systems are supposed to be identical, then run this on each system:

UNIX95= ps -eo vsz,ruser,args | sort -rn | more

This sorts all the processes by the amount of RAM they use. If you see a big difference, the names of the processes will be given and you can investigate.

BTW: If one of the processes is mib2agt, this is a known problem that was fixed a long time ago. If you shop does not run LAN-based monitoring tools such as HP OpenView Node Manager, then stop all the SNMP processes using the /sbin/init.d scripts and turn off the autostart capability by editing the SNMP files in /etc/rc.config.d

[Wodisch]

Hi Ajdin,

actually "vhand" does have an impact, too.
And it is about CPU time, as the referencing of all pages by the two "hands" can accumulate in case it is needed all the time.
But the effect of NOT having sufficient amounts of free memory pages available is causing so much more trouble, that this effect is only a hint at the main problem, but not a part of that problem.

You could use "glance" to check which processes trigger the paging, IIRC.
SNMP is always a problem, as Bill points out, but so are some of the OpenView products' processes, too.
Do you have shared memory segments in use, which are just a little too big to fit into your scarce RAM?
HP-UX *does* page shared memory, until it is "mlock"ed (memory locked), which needs the *privilege* "MLOCK" in /etc/privgroup.

Reducing the unix buffer cache could safe you a little, reducing some unneccessary kernel tables could help either (like NINODE - you are using VxFS not HFS, don't you?), and unused device drivers eat up memory-locked kernel address space.
Actually I would shrink all oversized kernel tables to about 5/4 of their actual peak level, remove all unneccessary subsystems and drivers, maybe try to use bigger, but fewer (LVM) PEs/LEs in all the volume groups (Ignite/UX is your friend) to reduce the amount of RAM needed by the kernel to keep the PVRAs and VGRAs in memory, maybe even use "chatr" to change the data segment size for the few processes allocating most of your RAM (so they need fewer pages, and hence smaller page-tables), and such.

Oh, check all the codefiles of the running/active processes with "chatr" if there are some parameters different between your two systems, and the used libraries, too! That could allocate much more RAM on one system than on the other...

A completely different case could be oversized directories, i.e. directories of much more than about 8KB, as every UN*X reads and writes directories at once every time, and hence eat up a lot of buffer cache, almost permanently, so that it get unneccessarily big.

Well, FWIW,
Wodisch

[Shannon Petry]

vhand will run on the system with little memory, as it runs with swap.

Your other sytem does not have the same problem, as it is not swapping to disk as often, or at all.

What you see in vhand is just a process to control disk i/o to swap. The less you swap the less this process will run.

Regards,
Shannon