Large Installations?

[Omeganon]

Hi!

I've briefly looked at Zimbra in the past but have had concerns about it's ability to scale to large numbers of users. After recently reading that Comcast has chosen to use Zimbra, I thought I'd revisit to see what's changed. I see that RHCS is supported now and that's interesting. I've not been able to locate any statistics that can help determine number of machines/number of accounts though. We currently have --

1 - 2x sendmail mail relay machines running RBLs, Mailscanner for A/V, attachment stripping, phishing detection

2 - 4x postfix machines running postfix/SA for per-user spam scanning and final delivery.

3 - 4x Courier-IMAP for pop/imap access

4 - 5x openldap servers for user account information. 1 master, 4 replicas

5 - 1x Squirrelmail for webmail.

6 - 1x Netapp for mail storage.

We currently have approximately 40,000 users with a total of 850GB of mail and about a dozen domains. Most users have <100M of mail about a dozen are in the 1-3GB range. In the last week we processed just at 17,000,000 messages of which, about 1,000,000 made it to users mailboxes. ~75% were deleted at the mail relay level and ~25% were deleted by spamassassin at the MTA level.

Since Zimbra utilizes many of the same programs and the architectural layout is nearly identical, I expect that the hardware requirements would be pretty much the same. Is that the case? What are other large installations utilizing? How's it working? Pointers on where to look would be ok too.

thanks!

--
Marc

I expect that Zimbra would replace 2-5 above.

[phoenix]

Welcome to the forums.

I would have thought that Zimbra could also replace your item 1 as well. I assume that you are talking about a Network Edition install? If that's the case then contact Zimbra sales for some help on sizing your installation, it would also depend on the hardware spec for the servers but it should be similar to what you're currently using.

[mmorse]

This test might perk your interest:
-While they use several machines to do the testing, the goal is to test the load an a zimbra mailstore running on one, dual-core dual-cpu machine (connected to a SAN backend).
=final result 50,000 active users/hr (200k provisioned users)each with 8 recieves, 8 reads, and 8 sends... served at once by one mailserver, (also using an mta, one external ldap, and one SAN)
and that was back in 2006 with zcs v3 I believe
Service Provider Benchmark

Not that comcast will ever release it...but I would personally love to see statistics on their setup; how many servers (and of what type), mail traffic stats etc. But I'll wager on that never happening-as I'm sure they don't want to run the chance of giving away info to competitors

[yeltsin]

hello there,

in regards to your VERY interesting reference to Zimbra benchmark document. I have read it through with great interest as I am tasked to build that test harness for both Zimbra and one of the "BIG guys". The goal is to get first hands comparative benchmarking experience on performance and scalability so that the management could make up their mind as to which way to go. I could not find a hint as to where those 50000 active users are sending their 32k messages: to themselves, to all 200000 provisioned mailboxes, or to the Internet (outside) receivers? The answer is crucial for us to get the correct results in the comparison study, on the same hardware as Zimbra ran the described benchmark.

Thanks in advance for any pointers!!!

[sarmengol]

Hi,

Could any of you send me the document "Service provider benchmark"? It seems that it is not any more on the network.

Thanks in advance!

[mmorse]

It was originally posted as part of the below blog entry & I do not know the status of it in the present (ie: if the url's simply changed or they took it down or it's just down because the partner portal is being worked on)

/blog/archives/2006/08/zimbra_benchmar.html

Though I urge you to contact Zimbra if you would like help with testing/planning.
-while people post their configs (HW/servers/users) all the time; we kinda avoid linking to load testing programs, least people label their test results as 'official', for configs/setups with variables way beyond our control/verification.
-professional services has deployment worksheets and sizing spreadsheets worked out on planning the equipment required etc, etc.

[kservik]

Are there any more recent benchmarks that have been done? Has anyone seen comparisons that others have done comparing IMAP performance?