A Large-scale Zimbra Deployment

[zimmy]

We are a modest IT solutions based company that presently provides email services to a customer/email-base exceeding 100,000 users (with a forecasted growth out to 250,000 within the next four years). The steady growth we've seen over the past few years means that a redesign and redeployment of a new email solution is inevitable. Our investigations into new and different technologies have lead us to Zimbra. It's been interesting to see that Zimbra has addressed many of the design considerations we've identified as relevant to our company. As such, I've been studying the online Zimbra documentation, reading through the forums and have installed the latest Zimbra beta on a development box for testing. At this point, I still have a number of unanswered questions - it's not to say the answers arn't out there but is to say I've been unable to find the answers thus far... It's my hope that someone out there will be able to offer advice or any personal experience with the following:

- Is there an anticipated release date for a production/stable version of Zimbra?
- Has there been much interest expressed by commercial or third party vendors in "professional" Zimbra support contracts? This isn't a show-stopper for us but might be a consideration.
- The scalability of any email solution is of concern to us. The Zimbra "architecture" should scale well, however, in practice, has anybody hosted/experimented with Zimbra for a *large* number of users with a *large* mail throughput? To what extent has Zimbra been stress-tested? Initial in-house experimentation has shown that the Zimbra suite take approximately 5 minutes to initiate on a 2.0GHz P4 test box with 1GB RAM. I haven't as yet loaded the Zimbra server but will be very interested to see how it copes when I do.
- I've struggle to find much information on the Zimbra clustering architecture and how it should be practically implemented - perhaps I've missed a technical references in my searching? This is a path down which I'm sure we'll need to proceed.
- Has anyone expressed interest in supplying patches to SpamAssassin to enable parsing of the Zimbra message store and in-turn training of system-wide or user-specific Bayesian databases? This would allow for incorrectly classified messages to be re-learned...
- Is multi-lingual support road-mapped for Zimbra?

Any help that could be offered on any of the questions raised above would be greatly appreciated.

Many thanks,
John.

[Andy from Zimbra]

John, thanks for the questions, some answers below. Incidentally, we're working with some other service providers that are similar in profile to what you describe.

Is there an anticipated release date for a production/stable version of Zimbra?

Yes, this is targeted for November. After that, of course, we'll follow with subsequent Beta versions of new functionality (such as full Outlook/MAPI capabilities, synching to wireless devices, etc.), followed by production releases of those versions.

Has there been much interest expressed by commercial or third party vendors in "professional" Zimbra support contracts? This isn't a show-stopper for us but might be a consideration.

A few vendors & potential partners are exploring this, but nothing to report at this time. We do provide this type of support as part of our Zimbra Network. You might have already read some details on our site about it, if not you can read more here: What's in the Open Source Edition vs. Network Standard Edition

The scalability of any email solution is of concern to us. The Zimbra "architecture" should scale well, however, in practice, has anybody hosted/experimented with Zimbra for a *large* number of users with a *large* mail throughput? To what extent has Zimbra been stress-tested? Initial in-house experimentation has shown that the Zimbra suite take approximately 5 minutes to initiate on a 2.0GHz P4 test box with 1GB RAM. I haven't as yet loaded the Zimbra server but will be very interested to see how it copes when I do.

This type of scalability testing is currently happening in various labs; of course our own, but also at some service providers of the size you describe and larger. Most of our developers come from a background building scalable messaging systems for ISPs & other service providers, so we've thought about issues related to developing a scalable architecture.

I've struggle to find much information on the Zimbra clustering architecture and how it should be practically implemented - perhaps I've missed a technical references in my searching? This is a path down which I'm sure we'll need to proceed.

You haven't missed it -- we haven't made that information available yet. We'll be providing that functionality within the next several weeks, as part of our Zimbra Network version. If you're interested in more information sooner than that please feel free to contact me. We just don't have the information in a "publishable" form yet.

Has anyone expressed interest in supplying patches to SpamAssassin to enable parsing of the Zimbra message store and in-turn training of system-wide or user-specific Bayesian databases? This would allow for incorrectly classified messages to be re-learned...

wrt to SA/Bayes training: the Spam/NotSpam buttons in the web UI will, if so configured in LDAP, forward the misclassified message to a pair of "mistake capturing" mailboxes. We are planning on writing a tool that would dump these mailboxes to files that sa-learn could understand. I would imagine you could extend this tool to walk all mailboxes if you want. (The tool will use SOAP to retrieve messages out of the server - understanding the underlying raw data is not good design for such tool - it should go through the APIs).

Is multi-lingual support road-mapped for Zimbra?

ZCS is 'internationalized', meaning it's possible to translate it into other languages. Currently members of the Zimbra Community have reported working on several translations, including Spanish, Italian, French, Brazilian Portuguese, Russian and others. You can see more information on translations/localizations here Translators: Latest msg files
We'll also be continuing to improve the ease with which translations can be done, as well as possibly working on certain translations as needed. We'll be interested to hear what specific translations you have in mind ?

Thanks again for the questions -- we'd be happy to provide more details or answer any other questions you might have -- if so, feel free to contact me via, andy.pflaum [at] zimbra.com