Memory problems with zimbra

[rajeshkodali]

Hi,

I am facing a big problem with Zimbra's memory handling capabilities. The problem is that my server has been upgraded from 2GB RAM to 8 GB and I see the RAM utilization is around 3 GB-4GB and suddenly the Free memory gets down to 50MB. I am here by attaching chart from hyperic, which shows the same. I have same problem at one other installation. I have made changes to mailboxd_java_heap_memory_percent = 20 and mysql_memory_percent = 25 but still this sudden unavailability is free memory is a puzzle.

I had to restart the machine and the free memory is back to normal. I have tried restarting zimbra services but it did not help. I think there is some problem with zimbra's memory handling.

My version is: Release 5.0.10_GA_2638.RHEL4_20081003035619 RHEL4 FOSS edition

[uxbod]

Just because it is taking all the memory does not mean that there is a memory issue. Linux will eat all the memory but also free it up when other applications require the use. Is your server swapping ? perhaps show us a top and freemem from the ZCS server.

[rajeshkodali]

I could bee most of the memory utilized by cache and the swap is not used. I have swapped 16 GB of space. And I could see the swap usage 200KB at the point of less free memory. As I have alerts configured at memory less than 100 MB to the network it causes a huge problem as I have to explain why this happens.

[uxbod]

Thats Linux for you Linux Howtos: System -> Linux Memory Management

[Jbrabander]

Good article, easy to understand for those of us that don't know much about Linux!

[LMStone]

Originally Posted by rajeshkodali

I could bee most of the memory utilized by cache and the swap is not used. I have swapped 16 GB of space. And I could see the swap usage 200KB at the point of less free memory. As I have alerts configured at memory less than 100 MB to the network it causes a huge problem as I have to explain why this happens.

Later Linux kernels seem to be pretty aggressive at moving things out to the swap file. Adding RAM to a Zimbra box after Zimbra has already been installed seems to exacerbate this tendency in our experience. I don't know why; it just happens.

Consequently, we now as a routine matter add to /etc/sysctl.conf:

vm.swappiness=0

as described in the Performance Tuning Guidelines for Large Deployments
wiki article. Swap file usage then goes down to next to nothing.

Hope that helps,
Mark

[Viking0]

I was worried for the same thing.
I had a zimbra installation on a 4 GB server and upgraded it up to 8.
Only problem I had was that maiboxd would not start so I had to set
mailboxd_java_heap_memory_percent = 18
Zimbra used all my ram as cache after this change, but after a cuple of days it relased it and now I have 2-3 gb of free ram. I't newer swapped or crashed in my case so don't worry.

[LMStone]

Originally Posted by Viking0

I was worried for the same thing.
I had a zimbra installation on a 4 GB server and upgraded it up to 8.
Only problem I had was that maiboxd would not start so I had to set
mailboxd_java_heap_memory_percent = 18
Zimbra used all my ram as cache after this change, but after a cuple of days it relased it and now I have 2-3 gb of free ram. I't newer swapped or crashed in my case so don't worry.

Yes, we've also reduced the Java memory footprint to a very small percentage to reduce swap file usage. It works until a full backup runs (300+ mailboxes) and then we would see swap file usage of 100MB or more.

After a lot of internal testing, we've settled on keeping the Java memory percentages at their default and setting vm.swappiness to zero as our standard build policy.

As a sidebar, on servers with 12GB or more we've been able to reduce the MySQL memory percentage (you have to change the innodb buffer size manually in my.cnf as well to fully implement) with no reduction in the innodb buffer pool cache hit rates.

Hope that helps,
Mark