architecture Zimbra

[gillou51]

Hi,

I want to install Zimbra Collaboration Suite on Ubuntu Server.

I have not managed to find the following information.
Could you tell me what kind of architecture should be put in place to install a platform zimbra for 2,000 users.


Regards.

[Hivos]

With Zimbra, usually IO is your bottleneck, so get fast disks, and no RAID 5 or 6. You could host 2000 users on one powerful box, but depending on how active they are you could decide to setup a multi-server deployment.

[Krishopper]

It also depends on how users are going to use your system, and how many emails are expected to go through the system in a given day.

Are your users going to be using webmail, outlook, imap/pop3? Are they going to receive minimal messages, or do they have a lot of email traffic?

If you are going to be using the Network Edition, it will be very easy to scale out and add another mailbox server and transparently move users to the new mailbox server if the load gets too heavy on the current server. You can also add additional mailbox servers in the Open Source Edition, but you lose the zmmailboxmove capabilities.

[gillou51]

thank you for your answers

I would actually make a multi-server (ldap, mailbox, mta).
I do not know, unfortunately the mail traffic that will be.
Regarding the version, I will use the network edition


If I install zimbra on 3 servers Mono-Xe quad cores, 8 GB RAM, 300 GB SAS
Is this be sufficient?

[Krishopper]

It's hard to say if it will be sufficient without knowing what usage/traffic will be like.

How are you going to spread services across the three boxes? (where will ldap/mailbox/mta's be?)

Since you'll be using network edition, you can easily spread mailboxes out across additional servers (with zmmailboxmove) in the future if you need to add more mailbox servers to handle the load. Adding additional MTA's is trivial if needed.

I'm a big fan of running LDAP on every machine to keep the directory lookups local and fast to each machine. If there is a network link failure, or if your master LDAP server fails, then the other machines are not affected and can continue to process requests.

[gillou51]

thank you for your answers Krishopper.

Last question.
In the case of an installation zimbra open source on a single server.
Is it possible to implement a redundancy on a second server?

[Klug]

If you're going with Network Edition, you should contact one (at least) VAR in your country in order to get informations from them.

Depending on the usage, 2000 users can be on a single server or will need the 3 servers you're talking about (you can start with one and add servers as needed).

If the usage is high, the problem might be the disks, not CPU/RAM.

Using OSE, there's no supported redundancy option supported "out of the box".
Some people are running ZCS OSE + DBRD and seems happy about it. But you need to know exactly what you're doing.
Other people are using virtualization tools to achieve high availability.

[chauvetp]

Those drives are going to be your bottleneck. When you say 300GB SAS, do you mean a single 300GB drive? Mirrored in Raid 0?

If a SAN is not an option, look into getting four to eight disks per system spread out on Raid 10.