Zimbra Slow

[jshtimsj]

I'm looking forward to start using Zimbra - migrating away from Qmail servers. However, I'm concerned with the high resource utilization and overall slowness I'm seeing with Zimbra. I'm testing on the following:

CentOS 5.5 x64_64
Zimbra 6.0.7 Community Edition
Dual Core 2.0Ghz Xeon
4GB RAM
RAID 10 (4 disks)

This is a new server with only a handful of test accounts on (< 10). Even stopping and starting the zimbra services takes nearly 3 minutes. CPU use is rather high as well for the server not being used. Is this normal? Installing the OS almost took less time than starting Zimbra... Is this to be expected or am I missing some obvious performance fine tuning after installation? CPU appears to be the bottleneck.

Thanks

[fcash]

This is definitely not normal. On that hardware, Zimbra should just scream.

We run our Zimbra 5.0 install in a Linux-KVM virtual machine (along with 8 other VMs) on a lowly 4-core Opteron 2.0 GHz system with a 12-disk RAID6 array, with 2100 users (95% are web client users). Takes less than 5 minutes for the entire VM host and all VMs to start. So something is definitely wrong with your setup.

I'd look into DNS setup to make sure it's not timing out on hostname lookups. And watch top to see which processes are using all the CPU/RAM. Then go through the logs for those processes to see what they're doing.

[jshtimsj]

DNS is working fine. Other than Zimbra, the server does 'scream'. Once Zimbra is started the web client is fine. The bottleneck is defintely CPU however when starting and stopping Zimbra as well as periodic spikes while Zimbra is running (without anyone using the server). Top shows Java as the culprit. When starting Zimbra, the mailbox, mta and logger services take the longest (by far). I'm looking through the various log files in /opt/zimbra/log but I'm not seeing any obvious errors. Same for /var/log/zimbra.log. Anyone else seeing this slowness?

[zimbra@mail1 ~]$ time zmcontrol stop
Host mail1.example.com
Stopping stats...Done.
Stopping mta...Done.
Stopping spell...Done.
Stopping snmp...Done.
Stopping archiving...Done.
Stopping antivirus...Done.
Stopping antispam...Done.
Stopping imapproxy...Done.
Stopping memcached...Done.
Stopping mailbox...Done.
Stopping logger...Done.
Stopping ldap...Done.

real 0m25.912s
user 0m1.126s
sys 0m0.444s
[zimbra@mail1 ~]$ time zmcontrol start
Host mail1.example.com
Starting ldap...Done.
Starting logger...Done.
Starting mailbox...Done.
Starting antispam...Done.
Starting antivirus...Done.
Starting snmp...Done.
Starting spell...Done.
Starting mta...Done.
Starting stats...Done.

real 2m8.925s
user 2m19.528s
sys 0m7.358s

I am running:

Release 6.0.7_GA_2470.RHEL5_64_20100610220647 CentOS5_64 FOSS edition.

Let me guess, Network Edition doesn't exhibit this behaviour...

[fcash]

Zimbra NE 5.0.13 64-bit running on Ubuntu Server 8.04 LTS starts up in about 5 (maybe 7) minutes (time to be able to login to web interface).

[rbreck]

I have an install that's pretty similar to yours, as well as another that runs Network Edition.

Both servers had issues similar to what you're describing

Check out this thread for some suggestions to tame the CPU java spikes... As for loading time, both editions take ~3-5 minutes to fire up, but I just figured there are a lot of componants that need to start......

High CPU Spikes every 1 min

[jshtimsj]

Thank you for the replies. I've fixed the periodic spikes, but the 3m restart time is still a bit worrisome. From the backup docs in the wiki its recommended you stop ZCS (which is absurd for a production environment). What am I missing? I'd rather not have a mail server down for however long the backup takes every day. Motivation to get NE?

Thanks again.

[rbreck]

Thank you for the replies. I've fixed the periodic spikes, but the 3m restart time is still a bit worrisome. From the backup docs in the wiki its recommended you stop ZCS (which is absurd for a production environment). What am I missing? I'd rather not have a mail server down for however long the backup takes every day. Motivation to get NE?

If you're concerned about backups, then yeah, you might want to consider NE.

NE can do live backups without shutting down any services and that function is working great for us.

There are some posts in the forums about achieving this on OSE with LVM snapshots, but I have not yet gone down that road as our OSE server is pretty low-priority compared to the people we have hosted on NE.