Running Zimbra
Has anyone had any experience running Zimbra Open Source edition on a Linode?
I am considering switching to Linode 2048, and I would be interested to know if I would be able to run approx 25 mailboxes on this size of VPS.
Would the vps have the ram to handle with this number of mail boxes?
jk
11 Replies
~~![](<URL url=)http://i.imgur.com/FtvrJ.png
![](~~
Evaluation and Testing
Intel/AMD 32-bit or 64-bit CPU 1.5 GHz
1 GB RAM
5 GB free disk space for software and logs
Temp file space for installs and upgrades*
Additional disk space for mail storage
Production environments
Minimum - 32-bit OS with Intel/AMD 2.0 GHZ+ CPU
Recommended - 64-bit OS
Minimum - 2 GB RAM
Recommend minimum - 4 GB RAM
Temp file space for installs and upgrades*
10 GB free disk space for software and logs (SATA or SCSI for performance, and RAID/Mirroring for redundancy)
Additional disk space for mail storage
*Temp files space- The zimbra-store requires 5GB for /opt/zimbra, plus additional space for mail storage. The other nodes require 100MB.
General Requirements
Firewall Configuration should be set to “No firewall”, and the Security Enhanced Linux (SELinux) should be disabled
RAID-5 is not recommended for installations with more than 100 accounts.
We ran Zimbra Open Source in house for a bit, and even on a 16G Single 2.6Ghz Xeon with RAID 10 it was never a speed demon. Eventually we migrated to SmarterMail which had much better Groupware features for way less money then Zimbra Commerical.
When my connection dies I do not receive any e-mails.
My current server has 4gb (as recommended) but it used to run with 2GB of memory and it is only an Atom based ASUS Hummingbird. So I wanted to see if anyone was sucessfully using Zimbra on a Linode.
A 2GB linode is also a little cheaper than a 4GB so I was wondering if I could get away with a Linode 2048.
jk
Start with the 2G, see how it works for you, then upgrade to 4G if needed.
Or spin a 2G and 4G up and compare them. You only pay for the time you use, so a few days testing each config side by side doesn't cost very much.
Easy to try in that you can very easily resize a linode, entirely automatically, without human intervention (apart from your own). Shut down the linode, select a new size, wait for the data to transfer to a new box (just a few minutes, it's quick), start up the now larger linode.
In terms of paying for what time you use, Linode does pro-rated refunds to the day. So if you activate a 4096 linode today, and cancel it tomorrow, you'd pay $160 when you activate it, and then be refunded $149.33 when you cancel it. People do use this to spin up additional nodes based on demand, and reduce the number when they're not needed, and Linode does't mind because it's entirely automated.
@Guspaz:
Brian Puccio: It looks like your 1024 linode is horribly overloaded in terms of memory consumption. You appear to need about 2.5x more RAM than you actually have, although a 2048 would probably be fine.
Oh, I'm aware. I was just showing that it's entirely possible to run it without issue on a 1024, let alone a 2048. I don't recommend 1024, but certainly think 2048 will be enough.
@Brian Puccio:
Oh, I'm aware. I was just showing that it's entirely possible to run it without issue on a 1024, let alone a 2048. I don't recommend 1024, but certainly think 2048 will be enough.
From the graphs posted on Monday, it looks like you're not only swap-thrashing frequently, but you've had to increase swap to 4x the normal amount and still ran out at least once. It could be said that "without issue" is not quite the right phrase for this situation.
That said, I'm with Guspaz: 2 GB would probably do nicely, with perhaps a sanity reboot every once in awhile. I'd ask what Zimbra does with all of the RAM it allocates, but I suspect it doesn't know either.
Caker is a lot more hardcore about this than many of us are (he advocates no swap at all, if memory serves), but I'm of the opinion that a little bit of swap, to get rid of the inactive stuff that doesn't really need system RAM even on a server, is probably warranted. Maybe 1/4th or 1/8th of system RAM. I think I'm actually beyond that, I think my Linode 512 has 256MB of swap going, but I'm only using about 50-60MB of that with default swappiness (60) since my system RAM isn't maxed out.
@hoopycat:
From the graphs posted on Monday, it looks like you're not only swap-thrashing frequently, but you've had to increase swap to 4x the normal amount and still ran out at least once. It could be said that "without issue" is not quite the right phrase for this situation.
I ran out once due to importing a 15 GB database to this server instead of the intended server.
Just to see how much more I need, I upgraded the Linode two nights ago. Since then my swap use is, according to munin, 1.7m pages in/second (so really 0.0017 pages/second) which means hitting swap every 588 seconds or almost once every ten minutes, which looks about right, looking at the graphs.
According to the Linode control panel, I'm doing about 68 blocks per second of IO:
~~![](<URL url=)http://i.imgur.com/1zd2L.png
This is less than each of my web and database servers and neither of those make any use of swap whatsoever.
By without issue I mean aside from dumping a 15 GB database to the wrong server, I make use of swap, but have never experienced an issue with zimbra itself, nor is my IO use higher than any other linode I have.~~