Can TS 6.5 not use Solaris swap?

TeamSite 6.5 SP2 + patch 1767
Solaris 9

We are running into a low virtual memory condition being triggered by a memroy leak in OpenDeploy/DataDeploy 6.1.1 but was curious why TeamSite is deactivating stores if we have 1.5 GB of swap virtual memory available?

This is a 2cpu, 2 GB ram server, that went from 1 GB free ram all the way down to less than 100 after we started running some heavy load.

Is there some sort of setting to get the iw.server process to recognize all the memory available? On windows I know it works based on the combination of both real and virtual memory.

Find more posts tagged with

Comments

jed

TeamSite 6.5 SP2 + patch 1767
Solaris 9

We are running into a low virtual memory condition being triggered by a memroy leak in OpenDeploy/DataDeploy 6.1.1 but was curious why TeamSite is deactivating stores if we have 1.5 GB of swap virtual memory available?

This is a 2cpu, 2 GB ram server, that went from 1 GB free ram all the way down to less than 100 after we started running some heavy load.

Is there some sort of setting to get the iw.server process to recognize all the memory available? On windows I know it works based on the combination of both real and virtual memory.

When TeamSite forks a process (such as an external task), it can eat up a lot of memory for a short period of time. (Read this for fork details: http://developers.sun.com/solaris/articles/subprocess/subprocess.html) On a dev/test server with 4GB of RAM we frequently had to restart due to out of memory errors. In short, get more memory or increase the swap setting.

Bill Klish

Jed,

When TeamSite was indicating out of memory, we had 1.5 GB free in the swap. What do you mean by increase the swap size?

Frederik

Sounds like the same thing we encountered. Our only successful countermeasure was to increase the swap space.

http://devnet.interwoven.com/forums_vb/showthread.php?t=16953&forumid=13

jed

Jed,

When TeamSite was indicating out of memory, we had 1.5 GB free in the swap. What do you mean by increase the swap size?

When a fork() is performed within TeamSite, the memory may dip for a short period of time. This is probably triggering the EOM message.

After that short period of time (as in seconds if not milliseconds), memory levels generally go back to normal, but by then it is too late. For quick and dirty resolution (and test case), you can always create a large (4GB?) swap file and see if the problem goes away.

http://www.eyrie.org/~eagle/notes/solaris/swap-file.html

I really think the root cause is fork() and the way child processes are allocated. If Solaris see's swap, it will use it. Scary, but 2GB is really not enough memory to run TeamSite.

birthibtest.zip

Bill Klish

Understand. This is a staging server and not our prod server. We have 4 GB there, and don't have search installed on the box.

I am having a hard time believing for that split second that 1.5 GB of swap vanishes.

beez

I really think the root cause is fork() and the way child processes are allocated. If Solaris see's swap, it will use it. Scary, but 2GB is really not enough memory to run TeamSite.

On Solaris, the TeamSite 6.5.X server uses a separate "Process Daemon" for spawning child processes. This is to avoid the time/space overhead of fork()ing from within a large process like iwserver. So, strictly speaking, there are no child processes of the teamsite server (iwserver).

Teamsite checks available virtual memory and usually initiates a shutdown of content stores when it falls below 100MB. Keep in mind that this is configurable ( "memlow_mbytes" and "memlowpercent" settings in iw.cfg, under the iwserver section), but you really should not need to change these.

Bk99nd,
Your swap allocation seems a little low. I seem to recall Sun recommending using 1.5X physical mem.

--Mike