High Memory Util in GlancePlus

[Tan Shirley]

Hi,

I'm having this new HP L-Class box with 256MB of RAM. However, now that the server is still not production (running live) yet and with no users logging on to it, it'a already having almost 85% of Memory Util as reflected in GlancePlus. Looking at the memory report in GlancePlus, it does breakdown the memory usage, showing the following :

Total VM : 25.1MB
Active VM : 13.5MB
Sys Mem : 32.4MB
Buf Cache : 7.0MB
User Mem : 178.4MB
Free Mem : 37.4MB
Phys Mem : 256.0MB

Attached is the screen captured in GlancePlus.

I really can't imagine how this system is going to go production when no users using it already takes up 85% memory.

Any parameters to tune in kernel ?

All advise is welcome.


Regards,
Shirley

[Michael Tully]

Hi,

Please advise what applications
and databases are installed and being used?

e.g. Oracle, Informix or other RDBMS.
If it is one of these how many instances are
being used.

Regards
Michael

[Tan Shirley]

Hi Michael,

We are using Progress database for our ERP system. However, I've tried shutting down all the databases, but it does not help much.

Also tried a fresh reboot without starting up the databases, it does help a bit, but I still get something like 30 - 40% mem util.
What could have "eaten" up the memory ?

Regards,
Shirley

[Patrick Wallek]

What is your buffer cache set at? Check the dbc_min/max kernel parameters. If they are the default values then a good start would be to change those values so that they are the same (like 15) and there is not a range of memory that it will use.

[Vincent Stedema]

Hi,

Patrick is correct: to reduce the overhead of the buffer cache growing and shrinking it's advisable to set dbc_min_pct and dbc_max_pct to the same value. Calculating from the information you provided, the size of the buffer cache is currently around 3 percent.

Could you also send us a screenshot of the process table (in Glance) and the system tables ( option in Glance)?

Vincent

[Tan Shirley]

Hi,

My dbc_min_pct and dbc_max_pct are both set to 2.

Attached is the screen shot of Glance Process and System table.

Really appreciate your help...
Thanks !

Regards,
Shirley

[Klaus Crusius]

Hi Tan,

you should also inspect the memory report button in glance.
To find oversized user processes, use "ps -el" and lokk at the SZ column. It gives the process size in units of 4k. That will help identify processes with big memory consumption. You could inspect their details with glance.
Also look for shared memory usage "ipcs -ma", where SEGSZ is the size in bytes. The owner of the shared memory gives typically hints for the usage.
Globally seen, high memory usage is not a serious problem per se, more interesting are the page faults per second etc. reported in the memory report of glance, when the system is running.
Can you perform a user simulation test in order to get a feeling for the perfomance.


Klaus

[Shumi Begum]

we have L100 server with 2gig of memory. we installed oracle 8.1.6 and oracle application 11i. when there in no users except 2 of us setting up/implementing Oracle application, we found memory utilization for one instance is getting somewhere 80-90%. this server is not in production yet. after some research and troubleshooting, we found there is a memory leak. there is a application call mib2gat. this was growing every single day. Once applied the patch PHSS_21046, it took care part of the problem. Also changing the dbc_min_pct and dbc_max_pct helped a little. You may check out for the mib2gat problem.
For us, it did not solve our problem completely. with one instance up and total of two users mainly implementing Oracle, we have 60-70% memory usage.

[mike pentecost]

one other thing to look at would be the kernal parameter swapmem_on if it is set to 1 swap will allocate user memory also. try building a new kernal with swapmem_on=0 and check memory use then.

[Tan Shirley]

Hi,

I've got something pretty similar to what Shumi had described. I realised that the mib2agt is taking up lots of mem. And I've searched the Internet and came to know about installing the PHSS_21046. I've downloaded the patch and installed prior to posting my question on this forum, 'cos the patch does not totally help!

As for what Mike suggested, I've checked my Swapmem_on and it's already set to 0.

As for Klaus, I've attached the ps -el output and the ipcs -ma output. Would really really appreciate if you could help me take a look and see if you could find the culprit that's "eating" up the mem.

Thanks so much, guys!
More advise are welcome!

Regards,
Shirley

[Mladen Despic]

Hi Tan,

It looks like your process "_mprosrv" is using a large amount of shared memory. Perhaps your database server is configured to lock certain amount of shared memory at startup. So, even if there are no clients connected, you will see high memory utilization. As your clients connect to the server, I would expect that the memory utilization will not go much higher.

If you want to list all processes using shared memory, take a look at the glance adviser script that I posted in:
http://forums.itrc.hp.com/cm/QuestionAnswer/1,1150,0xdd07a22d6d27d5118fef0090279cd0f9,00.html

HTH

[Tan Shirley]

Hi,

I've managed to reduce the amount of memory locking for my databses. Slight improvement. Now the mem util is around 68%. But I would still consider it on the high side since no processes and no users are using it!

I've realised that the system memory seems abit too high. 32.4MB of sys mem out of 256MB of Physical mem, which is about 12.66% !

Isn't that too high a percentage for a sys mem?
Any parameter to reduce the amount of sys mem?
Or what can I do to reduce sys mem?

Any advise are welcome. Thanks !

Regards,
Shirley

[Vincent Stedema]

Hi,

You can start with disabling daemons that aren't used. This will free up some resources.

For instance:

CDE desktop:
disable with /usr/dt/bin/dtconfig -d

NFS services:
set NFS_CLIENT, NFS_SERVER, START_MOUNTD to 0 in /etc/rc.config.d/nfsconf

automounter:
set AUTOMOUNT, AUTOFS to 0 in /etc/rc.config.d/nfsconf

pwgrd (password caching and hashing daemon):
set PWGR to 0 in /etc/rc.config.d/pwgr if your passwd file is relatively small, but not if you have a rather large passwd file.

Only disable these services if you don't need them for normal operation!

Hope this helps.

Regards,

Vincent

[Klaus Crusius]

Hi,

there is approximately 100MB of anomymous (key=0x00000000) shared memory used by owner root. The processes using these areas are named "_mprosrv" and "_mprshut". So I guess the postgress server is demanding this amount of memory.

Kind Regards, Klaus

[Bill Hassell]

Let's start with the practical part: 256 megs is crippling your L-class system, especially for a database application. If you want good performance, add between 2000 and 4000 Mbytes of RAM. From the HP-UX 11.0 webpages on www.hp.com, the minimum RAM for full benefits of 64-bit memory management requires 4000 Mbytes of RAM.

100% RAM usage is a GOOD thing. HP-UX is a virtual memory system and if more space is needed to run additional processes, the system will page deactivated processes out to the swap area. You can easily run thousands of megs of processes on your system as long as you have enough swap space.

Now I said 'easily run', not 'efficiently run'. Once paging starts, the throughput of the system drops in proportion to the amount of paging. You have a very small buffer cache (under 8 megs) which severely limits filesystem I/O. The percentages you are seeing are a function of the incredibly small amount of RAM you have available.

You could spend a few days removing various network and operating system daemons and get free memory up to 150 megs...but for a database program, that's not enough for a single instance of a well-configured program.

If you are not concerned about performance, add 2000-4000 Mbytes of swap space and you'll be fine. HP-UX will allocate space for all the processes and then page them back and forth as they are needed. To monitor the paging rate, run the following commands:

vmstat -z
then
vmstat -s

vmstat -s will summarize memory activity. Look specifically at page out as this is real swap space being used, while page in is counted for starting programs as well as paging back in.

[CHRIS_ANORUO]

Hi Tan,

Set dbc_max_pct and dbc_min_pct to 12, nbuf and bufpages to 0(zero) and swapmem_on=1. Your total swap space should be at least 2*Physical memory.
This will give you a dynamic buffer caching and pseudo memory of 3/4 of Phy mem. Try out these changes through kernel configuration settings.

[Tan Shirley]

Hi Chris Anoruo,

You were suggesting changing swapmem_on=1.
However, Mike Pentecost was saying that if swapmem_on is set to 1, it will allocate user memory. So I'm not sure whether setting swapmem_on to 1 is a wise choice or not. Please advise.

[Sandy Orr]

I am running HP-UX 11.0 on a 6 way N-Class with 2GB RAM, 1GB swap, swapmem_on=1 and dbc_max/min set to 50/5.

Sept 2000 XSWGR/XSWHWCR bundles are installed with oracle 8.1.6.2 and our companies products. Afer a few days memory runs out and login is impossible. The initial assumption was a product memory leak, but investigations have shown the problem to be a more general one:

- Product shared memeory segments not released
- Oracle shared memory not released
- Glance shared memory not released
- buf cache grows to 1GB

swapinfo shows 'memory' % growing.
vmstat shows free pages decreasing (possibly co-inciding with ftp sessions)

o Am I missing more recent patches?
o Will setting the swap area to a better size resolve this and, if so, is this not a defect in memory management based on the above configuration?

I have been investigating this for two weeks with no single symptom and several red herrings including mib2agt high memory usage resolved by patch PHSS_21046

Please can someone show me the light?

Thanks
Sandy