Memory usage on Zimbra 5.0.2 NE on RHEL 5.1?

[Miklos Kalman]

Hi,

I am running a Zimbra NE on RHEL. The machine has currently 65megs free memory which is a bit alarming. I am setting up a new environment which is being migrated from the FOSS server.

---
15821 zimbra 18 0 2010m 809m 10m S 1 20.5 3:37.67 java
2625 root 18 0 10180 644 552 S 0 0.0 0:00.86 hald-addon-stor
15457 zimbra 15 0 1211m 163m 4300 S 0 4.1 0:17.77 mysqld
1 root 15 0 10316 688 576 S 0 0.0 0:00.43 init

---

The machine is a Core2Duo 3GHz with 4GB of RAM running on 64bit RHEL5.1. It seems that Java and mysqld take up more than 90% of the memory and this machine is NOT even receiving any mail.

How can I lower the memory usage, since I am a bit worried that I will need more memory when it hits production.

[Miklos Kalman]

Is there a way to limit the memory usage of java and the mysql daemon?

[dijichi2]

zimbra pre-allocates huge amounts of memory by default. you can change this with:

zmlocalconfig -e mysql_memory_percent = 40
zmlocalconfig -e mailboxd_java_heap_memory_percent=30

I think these are the defaults - lower them if you want to use less memory at startup, however if the box is dedicated to zimbra leave them as they are.

[apatosaur.9]

I use 5.0.2 GA on ubuntu 7.10 server, with memory useage seldom above 500 Megs. (only time was actually during install)

Admittedly I am running it for demo purposes and the load isn't especially high, but it is delivering mails and working with Thunderbird with Lightning and zimbra desktop.

I have reduced ram on this box to 768 Megs.

Disclaimer. VMWARE.

/etc/security/limits.conf on debian systems may help
/etc/sysconfig/limit* (or something like it on RPM distros)

[Miklos Kalman]

the machine is dedicated to Zimbra entrirely. However if it only leaves 65 megs during "no traffic" how much will it eat when it goes live? The only traffic it gets currently is uploading the messages from the old zimbra foss server.

The FOSS machine is running on VMWare as well. THere I gave it 3GB which it leaves usually around 500megs. And that has around 20 users and mails coming in all the time, greylisting, spam...etc

Is this memory difference due to the Network and FOSS differences ? I thought 4GB would be engough for around 20 users right? But with the rate its going I am not sure anymore. It would be good to know if this would improve over time, or should I just buy more memory before putting it into production?

[quanah]

Editing the number of file descriptors available to slapd in 5.0.2 will significantly reduce memory pressure. This fix will be in 5.0.3. To fix this, as root edit /opt/zimbra/libexec/zmslapd so it looks like:

ulimit -n 32768
ulimit -c unlimited
export LD_PRELOAD=/opt/zimbra/tcmalloc/lib/libtcmalloc.so
exec /opt/zimbra/openldap/libexec/slapd "$@"

Basically, just add the "ulimit -n 32768" line.

--Quanah

[fcash]

Note: "free" memory is wasted memory in Linux. Any memory not currently and actively being used by programs *will* be used as a disk/data cache. When programs need memory, pages from the cache are freed for them to use.

You only have to worry when SWAP usage starts climbing into the gigabytes. Until then, relax in the knowledge that the kernel is doing it's job managing memory in such a way to reduce the amount of disk traffic.