Concerns about Zimbra Performance

[soxfan]

Early last week I posted about a problem I was having with email getting stuck in my active queue, which went unanswered. That's OK; I understand the nature of these forums. This forced me to re-think how I was handling Spam on our email gateway, which I probably should have done before any way. So, it was kind of a blessing in disguise that no one answered.

However, this issue has me a little concerned. I have been pushing Zimbra for about six months at my company as a replacement for our current Linux-based POP3-only email server. I have finally started to make some headway, and it looks as if we may start the migration to Zimbra soon (OSE, at least initially). When the migration is complete we will have somewhere in the range of 400 email users.

When I has having the problem with email getting stuck in the queue my only guess of the cause was that ZImbra was simply being overwhelmed by an extremely high volume of messages. Although I have fixed the problem, I am worried that once I have all the users migrated over, and there is a pretty high volume of messages that Zimbra won't be able to keep up.

I am planning on using a dedicated server for Zimbra. It has dual Xeon 3.4 GHz processors, with 2 GB of memory.

I'm looking for some feedback about what I can expect with my proposed environment.

TIA,
John

[KevinH]

John, What type of message flow do you expect? Are the users all POP? Any IMAP or Web Client users? How big are your quotas? What are the avg mailbox sizes and range? Avg message size?

Users are different. For example a consumer mail profile getting 10msg/day vs a Wall St trader profile with 100's of messages per and 8hrs of IMAP use. Can you give us some more detail to your user base? As for why your mail was backing up we'd need more info there.. CPU bound? What do you have for disks? For this many users' you'd want several and to go with RAID 10 for maximum IO performance.

[soxfan]

Thanks Kevin. I probably should have posted some more details. A good portion of the users that we migrate will be moving from POP to either IMAP and/or the Zimbra web client. We currently do not have quotas (not really relevant with POP), but the server was spec'd out for an average of 1 GB per user. The Internet email gateway server that we have tracks daily message stats. Currently we are receiving between 10-15k messages a day that are not tagged as Spam. Unfortunately at this point I don't have comparable stats for intra-office email, but I would say it's at least the same, if not more. Most of the messages are pretty small in size, but we also have a fair share of messages with attachments. Presently the disk setup on the server is 5 146.8 GB Ultra320 10K RPM SCSI drives, configured for RAID 5, with one drive as a hot-spare, giving us approximately 420 GB of usable disk space. After I had configured the server I read some forum posts suggesting RAID 0+1 was the way to go. This is something we may consider changing, but I was hoping to at least start out using our current config.

[phoenix]

Don't use raid5 for Zimbra, your performance will suffer. As Kevin mentioned use raid10 you'll not regret it.

[soxfan]

OK, if my thinking is correct I will need to add another 146.8 GB drive to the array and make the hot spare active in order to keep my available space around 420 GB. Let's just assume that getting the additional drive is not an option at this point, and that I don't want any less than 420 GB of available space. Am I better off doing away with the current RAID 5 configuration and going with some other configuration of the existing drives?

[KevinH]

If you really can't get more drives and *must* stay at 420GB then you are stuck but as Bill warned RAID 5 is the worst config for our index and DB files. Lots of small writes and heavy random read traffic. For a smallish system it's ok but when you start talking 100's of active users we always push people to use RAID 10. The only place that RAID 5 is ok is the message files. Even here if you want a fast system you'll want to put the most recent 30days of mail on RAID 10 and use HSM (Network feature) to move older messages to the RAID 5 partition. This just gives you a slightly lower cost for the archive store of mail.

[apocryia]

We have been using Zimbra for our Dev team email for a few months now.

Recent issues with not being able to log into a backup copy easily to verify and now email getting stuck in the deferred queue is forcing us to think about dumping Zimbra and moving on.

What did you do to get your queue mail going?
I have a post with no response so far.
Deferred holding tank

Apocryia

[phoenix]

Originally Posted by apocryia

We have been using Zimbra for our Dev team email for a few months now.

Recent issues with not being able to log into a backup copy easily to verify and now email getting stuck in the deferred queue is forcing us to think about dumping Zimbra and moving on.

What did you do to get your queue mail going?
I have a post with no response so far.
Deferred holding tank

Apocryia

I hardly think that 14hrs is a long time to wait for an answer.

[crp]

i would sum up the situation that the Zimbra is working but not functioning.

The server is running FC5 on a dual xeon 2.4 system with about 2G of RAM.

Peformance issue with just the email client is right off 2 fold:
1: it is so slow even with under a dozen messages
2: on large emails , for example the logwatch report, Zimbra chokes on displaying the whole message. It eventually just displays a black section.

[Klug]

Could you also post your client setup ?
Little/old computers do not like Zimbra when using the WebUI.