Insufficent Memory Problem

[Preeti Khanna]

Hi
I have one N-Class server running HP-UX11.0 . This is the production server of SAP application. I am having 4GB of RAM and 32GB of Swap space in my server but when i start SAP application and user's starts logging in . At the SAP front end it says insufficent Memory . When i see at OS level only 13 MB free memory is available out of 4 GB . When i see the process which was taking much memory I found only SAP application which was taking much memory.
If i increase the Swap space shall this slove my problem or i have to increase the memory.
Can anybody suggest.

Thanks

[Denver Osborn]

When you get the insufficient memory error, if you think it might be swap execute this command when you encounter the problem:

# swapinfo -tam

Look at the bottom line under the Percent Used column. If the pct used is +90% maybe you could increase swap. If it doesnt appear to be swap you're probably looking at tuning some kernel parameters. Was there anything in /var/adm/syslog/syslog.log? Maybe you should look at dbc_max_pct, if it is at the default of 50% possibly reduce it to 10% or 15%.

[James R. Ferguson]

Hi:

Take a look at this thread from last week. There is quite a discussion of swapspace and the analysis some real "swapinfo -t" data.

http://forums.itrc.hp.com/cm/QuestionAnswer/1,1150,0xbc8a0559ff7cd4118fef0090279cd0f9,00.html

You might also want to adjust (upward) the kernel parameter maxdsiz. For an overview of kernel parameters see:

http://docs.hp.com/hpux/content/KCparams.OverviewAll.html

If you have Glance it may provide you with some additional clues. Tuning may be needed in addition to more physical memory.

...JRF...

[john strumila]

without knowing much about unix or what exactly your message says I still think it's got nothing to do with swap.

The problem is d/t/s size. Most likely maxdsize.

[Vincente Fernandes]

Looks like you need to have a walk thru with you current kernel parameters and the one's suggest as per SAP applications config. Most of the time the kernel parameters needs to be fine tunned or increased/adjusted as per the applications specifications/requirememts.
Look and see if your SAP applications can be fine tunned to your environment, i.e don't use modules which are not used. Also review you SAP application config before increasing memory.

[CHRIS_ANORUO]

Hi Preeti,
You have a very large swap space, your really don't need that size of swap space. Set your swap space to 2*Physical memory and please look at this linked document and also set your kernel parameters as stated. It will help you to tackle job. You can also use sysdef; adb command (echo "bufpages/D" |adb /stand/vmunix /dev/mem) to determine your bufpages. From kernel configuration, set swapmem_on=1; allocate_fs_swapmap=0; dbc_max_pct=25; dbc_min_pct=5; bufpages=0; nbuf=0 .This will give you pseudo memory of 75% of you total physical memory and you buffer caching will be dynamic.

http://forums.itrc.hp.com/cm/components/FileAttachment/0,1167,0x618b0559ff7cd4118fef0090279cd0f9,00.htm

[Anthony deRito]

32 GB of swap!!! I believe your issue is directly related to virtual memory but I am very surprised at this. I would use vmstat and check out your stats on VM to confirm this theory. Also use swapinfo -tma and let us know what these values are. What does your SAP vendor say about memory requirements??

Tony

[Andy Monks]

I recently worked on a performance issue with SAP on an N-Class. I turns out that the latest version of SAP (64bit version of SAP and 64bit version of Oracle), uses 10 times more memory than the 32bit version. Now each SAP process can require upto 2GB of virtual memory.

We found that huge amounts had be paged/swapped out, and the machine was under so much memory pressure it deactived processes. This included the nfs processes, which effectively hung the machine as it was in a MC/SG environment. This N class also had 4GB of memory and 20 SAP processes. We increased the physical memory to 8GB and it behaved much better (although shutting down SAP every 2 weeks helps a lot!).

If you monitor the paging activing using vmstat (or glance if you have it), you will probably see heavy page out (po column). Don't worry too much about page in's (pi column).