About HP-UX 11.23 Memory Management

[Kalin Evtimov]

Hello everyone,

I want to share some experience about HP-UX 11.23 and ask a question about memory management.
I am running a SAP/R3 System, which eats lots and lots of memory. I made graphical memory monitor based on the freemem tool for HP-UX. Very often, when some big reports were run simultaneously, the system didn't respond for some minutes. As I could see on the graph afterwards, freemem went down under the value of minfree, desfree went up and the system stopped responding, even login didn't appear. For this reason, I switched xprm and set the maximum memory usage for R3 to 90%. Although, I noticed that memory goes down again, but this time system behaves otherwise. It becomes very very slow for a longer period of time, but at least didn't stop responding.
My Question: Can you give me some memory optimizing ideas (we have 16GB RAM)?
How can I get the freemem, minfree, lotsfree ans desfree parameters on HP-UX without using the freemem-tool??

Thank you.

Here my settings:
freemem= 155163
gpgslim= 15360
lotsfree= 65536
desfree= 15360
bufpages= 209562
dbcminpgs= 209562
dbcmaxpgs= 209563
minfree= 7424

[Steven E. Protter · Accepted solution]

Shalom,

You might wish to check a few things:

1) Buffer cache. Check dbc_max_pct, dbc_min_pct which you can set dynamically in HP-US 11.23. Set the max lower, a few percentage points above min. This will simply free of some memory.

2) Check maxdsiz parameters for 64 bit and possibly increase them.

3) Run a tail -f /var/adm/syslog/syslog.log and re-run your test. You may be receiving diagnostic messages and not even know it.

SEP

[Steve Lewis]

Hi, I suspect that your SAP tuning is the problem. SAP always wants to use every scrap of memory you have, even to point of starting the system to page, which is counter-productive, because when it pages, your performance is reduced 100-fold. You have to be firm with your SAP admins and tell them it is their fault that the system is slow and not to allocate more memory than (RAM-kernel-DB-bufpages). Within SAP you need to check the value of extended memory (em/initial_size_MB). Compare that to the shared memory segments listed in ipcs -ma. In fact, if you add up the segment sizes from ipcs, then you will see where the memory has gone.
If you do a vmstat -S 5 20 you will see if it is paging by checking the PO value. If its ticking over at 5 or so each time, then you are just about OK. If PO hits 50-300 then you will start to see things grind to a halt.
Another common problem is people running their database on their SAP server, getting memory contention between the two systems. If it is shared, then consider moving your database off the SAP server.
See this thread here for more info:
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1079646

[Kalin Evtimov]

Thank you Steve,

The only solution seems to be more RAM. I also think that this is SAP Problem, because the OS is acting as it's supposed to do.
Thank you for your advice, I am going to read more about tuning SAP. I don't think we will get another server to separate DB from SAP :(