Downgrade due to load [or mail/setting migration]

[mhakali]

Hello.

As some people have noticed Zimbra 6 consumes a lot more load (link) than Zimbra 5. The entry is well described, and I can just concur this with my own installation:

<-- why don't I get an image in the forum? :/ Anyhow, click link to reveal.

Upgrade in September.

I use Zimbra OSS for home as it offer a great suite. It is a single server deployment. Currently running in a virtual host with 2GB RAM and 3x 2.4GHz CPUs for disposal, as I had to increase from 1 to cope with the load to get some reasonable response time (w-t-f )

As another writer (link) clearly gets a response to is that a downgrade is not supported.

I would probably just restore the obvious backup that I made before upgrading, but I hate the thought of losing several months worth of mails due to this.

Is someone working addressing this issue?

Can I downgrade?

If not, can someone point me out how I can make a new fresh installation and migrate my mails, settings and accounts to the older Zimbra?

Any advice would be appreciated.

Mikael

[y@w]

If it's on a virtual machine, your best option would be to setup another vm beside it and migrate your mail to it.

There are several tools out there to migration your mail to the new virtual machine. Have a look here in the Zimbra wiki. You might even be able to just export your mailbox from the web client and re-import into the new server's web client, but I haven't tried going backwards..

As for the loads you are experiencing, how many users do you have? You mentioned that this is a home installation? I'm using Zimbra 6.0.4 with 2GB of RAM with 1 vCPU at home with very little load..

[mhakali]

Originally Posted by y@w

If it's on a virtual machine, your best option would be to setup another vm beside it and migrate your mail to it.

There are several tools out there to migration your mail to the new virtual machine. Have a look here in the Zimbra wiki. You might even be able to just export your mailbox from the web client and re-import into the new server's web client, but I haven't tried going backwards..

As for the loads you are experiencing, how many users do you have? You mentioned that this is a home installation? I'm using Zimbra 6.0.4 with 2GB of RAM with 1 vCPU at home with very little load..

Thank you for the tip!

This is odd. I have about 3-4 users and handling ~10 domains (with aliases to these users).

Needless to say it's the simplest setup possible.

Perhaps I should try doing a parallel setup with a fresh 6.0.4 install and see what gives? This Zimbra has been around since version 4.x I would think. Being upgraded as new versions comes along..

Then again upgrade procedures are well taken care of by the script.

Anyhow. Thanks for the tips. :-)

Mikael

[y@w]

Hmm. I have 2 users, but the interface is snappy. Are you on 6.0.4? What OS? What Zimbra components are installed?

[mhakali]

Originally Posted by y@w

Hmm. I have 2 users, but the interface is snappy. Are you on 6.0.4? What OS? What Zimbra components are installed?

Standard components for a single sever install.

root@imix:/opt/zimbra/log# dpkg --get-selections | grep zimbra
zimbra-apache					install
zimbra-core					install
zimbra-ldap					install
zimbra-logger					install
zimbra-memcached				install
zimbra-mta					install
zimbra-snmp					install
zimbra-spell					install
zimbra-store					install

I recently upgraded to 6.0.4. Have earlier been at 6.0.1, 6.0.2 and now 6.0.4. Before that on various 5.x versions where the load increase never occurred.

I have added some DNS blacklists to config for incoming mail. However, this should have an extremely low impact on performance.

It's an Ubuntu-server 8.04 LTS.

Mikael

[y@w]

You could start with disabling memcached.. As I understand it, only Zimbra proxy (nginx) will reference memcached for requests anyway and Zimbra proxy doesn't appear to be installed or enabled.

When you run top (or htop), is there any process that stays at the top of the list?

[mhakali]

Originally Posted by y@w

You could start with disabling memcached.. As I understand it, only Zimbra proxy (nginx) will reference memcached for requests anyway and Zimbra proxy doesn't appear to be installed or enabled.

When you run top (or htop), is there any process that stays at the top of the list?

I could do that. However if it's not used it will not consume processing time But it will free up ~100MB RAM. Then again the RAM consumption of the box is not fully used. (1.7GB out of 2GB, and that includes fs caches of 700MB). I will try it later on, but I don't see it as a likely cause for load.

What's consuming CPU time in top is "java".

13455 zimbra 20 0 207m 15m 8084 S 89 0.8 0:00.29 java

Java is invoked to do a lot of stuff in Zimbra, so just saying "java" does probably not do any good.

What would be interesting is to see exactly what component that java invokes that takes the longest to execute. Not sure how to graph this in a good manner over a long time. But I will think of something.

But the processes spawn frequently and stays at 100% CPU for longer or lesser time.

Besides; Why doesn't other components such as the web frontend etc use memcached? It's a great way for bytecaching objects.

Mikael

[y@w]

Originally Posted by mhakali

I could do that. However if it's not used it will not consume processing time But it will free up ~100MB RAM. Then again the RAM consumption of the box is not fully used. (1.7GB out of 2GB, and that includes fs caches of 700MB). I will try it later on, but I don't see it as a likely cause for load.

What's consuming CPU time in top is "java".

13455 zimbra 20 0 207m 15m 8084 S 89 0.8 0:00.29 java

Java is invoked to do a lot of stuff in Zimbra, so just saying "java" does probably not do any good.

What would be interesting is to see exactly what component that java invokes that takes the longest to execute. Not sure how to graph this in a good manner over a long time. But I will think of something.

But the processes spawn frequently and stays at 100% CPU for longer or lesser time.

Besides; Why doesn't other components such as the web frontend etc use memcached? It's a great way for bytecaching objects.

Mikael

Yeah.. I was mostly just wanting to get rid of anything running so there was no way it could use up CPU.

If you do a

ps aux | grep java

, while the CPU is 'pegged' we can see if there's any zmprov commands that are running via cron or some other mechanism and using CPU. I'm guessing that's the case from the output that you pasted as this java process isn't using much for memory (mailboxd uses 30% for the heap size by default Performance Tuning Guidelines for Large Deployments - Zimbra :: Wiki).

I can't really answer why Zimbra doesn't use memcached for all the services, but in your case I can't imagine that it would help much.. You need quite a bit of memory just to get Jetty, MySQL and other off the ground to have much memory left over for use as an external cache.

[raj]

Try this

High CPU spikes every 1 minute

Raj