 Kerio performance tuning? [message #101043] |
Wed, 03 April 2013 19:27  |
chrisrosa
Messages: 16
Registered: May 2006
Location: Novato, CA
|
|
|
|
I have a 7.4.3 server running on an Xserve, with the mailstore on a RAID5 (I know, I know!). I have about 90 users (291 GB mailstore) and day to day usage is fairly decent still, but I'm definitely seeing slow performance for large copies/moves/deletions, in both Webmail and via Apple Mail. IMAP timeouts are common with anything more than 400-500 messages at a time, which makes account chores difficult for some users.
I've been wanting to move to RAID10, but haven't found a device I like yet that would work well with the Xserve. The current mailstore is connected by fibre, which is totally fine, but newer fibre direct attach storage tends be very large capacity with 12-bays+, which is a bit overkill and expensive.
I opted to try out the Areca ARC-8050, which is a 8-drive Thunderbolt RAID, with 8x300GB 10K drives (64K stripe). This is then paired with a new Mac mini server with 16GB of RAM and SSD boot drives. RAW performance of the RAID is pretty great with testing utilities, but I'm wondering if there's anything specific that can be done to maximize performance even further with Kerio mailserver.
Right now, I have 7.4.3 running with a restore of my domain via kmsrecover. I've logged into a couple user accounts and done huge copy, moves and deletions, via webmail and IMAP, and performance seems goood (great compared to the live server), but nowhere near the raw speed of the RAID. Additionally, it seems like the mailserver process only ever uses one processor core, but maybe it's not being pushed hard enough with the couple accounts I'm testing with? If you have any load testing tips specific to Kerio, I'd love to hear them.
My plan is to get all the hardware dialed in with 7.4.3, then upgrade to 8.x sometime after.
Thanks
|
|
|
|
| Re: Kerio performance tuning? [message #101044 is a reply to message #101043] |
Wed, 03 April 2013 20:41  |
tonyswu
Messages: 86
Registered: July 2010
Location: Bellevue, WA
|
|
|
|
|
Your new setup is pretty similar to us, and I think it's good and will last for a long while. I do not have any load testing solutions for Kerio other than have a lot of users do a lot of stuff like you did. The thing is the performance you see on Kerio will never be close to what the test utility gives you. One reason being Kerio still need to process the commands given to it, and two is test utility usually tests using big block of data (4GB, 8GB, etc.) while Kerio really deals with thousands of small files, and that's very different. If you want to see how fast your RAID can go as far as copying / writing small files, write a simple script to write 1billion small files ranging from 1K to 999K and time it.
|
|
|
|
|
|
| Re: Kerio performance tuning? [message #101046 is a reply to message #101045] |
Wed, 03 April 2013 22:13 |
chrisrosa
Messages: 16
Registered: May 2006
Location: Novato, CA
|
|
|
|
Successfully killed mailserver during testing today. I had 3 IMAP clients running (2 Apple Mail, one Thunderbird on Win7), and 3 Webmail clients running. All doing large copy/moves of >8K messages...even a few 14K. Stepped away for a bit, and returned to find the mailserver process maxed on two precessor cores, and all connections locked up. Had to force quit the process.
Of course, it's super unlikely copies/moves of this sort would be going on concurrently on a typical day, but it would be nice if it was a bit more bullet proof. I guess a timeout is preferred to a server lockup, but neither are appealing.
|
|
|
|
| Re: Kerio performance tuning? [message #101047 is a reply to message #101046] |
Wed, 03 April 2013 22:38 |
tonyswu
Messages: 86
Registered: July 2010
Location: Bellevue, WA
|
|
|
|
|
Interesting. I am gonna have to try the same thing on my server and see if I can bring it down. I've always wondered how Kerio does their stress test, if some Kerio rep can shed some light that would be helpful. Not necessarily how it's done, but what the numbers are would be nice.
|
|
|
|
|
|
| Re: Kerio performance tuning? [message #101118 is a reply to message #101073] |
Fri, 05 April 2013 01:01 |
chrisrosa
Messages: 16
Registered: May 2006
Location: Novato, CA
|
|
|
|
HoosierMac wrote on Thu, 04 April 2013 04:54Honestly, HFS+ isn't a great filesystem to host Kerio on for larger mailstores.
Yeah...definitely an achilles heel. Would be nice if Kerio offerend some internal benchmarking so we could directly compare.
|
|
|
|
|
|
|
|
| Re: Kerio performance tuning? [message #101143 is a reply to message #101123] |
Fri, 05 April 2013 19:42 |
j.a.duke
Messages: 239
Registered: October 2006
|
|
|
|
rsip wrote on Fri, 05 April 2013 04:11AFAIK we cannot publish or comment performance because Apple's NDA.
NDA for what? HFS+? That's been out for 15 years! My experience with NDAs from Apple, Microsoft and other developers is that they typically expire when the product or feature ships. I've had more than a few NDAs reviewed by lawyers and they concur with this understanding. There may be communications (emails, etc.) during the NDA period that remain covered by the agreement even after the product/feature ships, but this certainly doesn't fall into that category.
I could see if there was a request for performance under, let's say Mac OS 10.9, which would be covered by a non-expired NDA, but we are discussing released OS versions like 10.6 (really old and not likely under NDA), 10.7 (old and not likely under NDA), and 10.8 (newer and not under NDA, unless you are referring to 10.8.4 and other future updates, which are likely covered by a developers agreement NDA).
A simple set of tests on a released OS version regarding Kerio Connect performance when attached to a specific type of storage and on a given host CPU wouldn't even raise an eyebrow from my experience. In fact, back in the day, these benchmarks were de rigueur in the specifications of any server/storage/etc.
I think, that while KC can be run on any number of combinations, contributions from the community and Kerio would help us determine what works best given certain constraints.
Cheers,
Jon
|
|
|
|
|
|
|
|
|
|
|
|
| Re: Kerio performance tuning? [message #101189 is a reply to message #101043] |
Mon, 08 April 2013 15:42 |
MacLab
Messages: 233
Registered: May 2012
|
|
|
|
|
I recommend waiting until a period of lesser activity such as evening or early morning to do the shutdown. If you do that, there are no problems. Also, all the self-healing the server performs of index files now makes a big difference.
|
|
|
|
Members
Search
Help
Register
Login
Home
