Page 1 of 3 123 LastLast
Results 1 to 10 of 21

Thread: [SOLVED] Fast RAID Setup for Zimbra...

  1. #1
    Join Date
    Mar 2007
    Posts
    24
    Rep Power
    8

    Default [SOLVED] Fast RAID Setup for Zimbra...

    ...I want it, but I don't have it.

    We've got performance problems with our Zimbra server at my site. The load average on the box bounces between 2 and 7 during work hours (graph attached), and top shows 40-70% CPU time is in IO Wait. Mail operations like moving/deleting messages have become very slow, especially for imap and web-client users. I can't run filesystem backups during the day for fear of driving the disk I/O even higher.

    Here's our setup:

    Edit: 1 Zimbra server, ZCS 4.5.10
    73 accounts
    Dual Core AMD Opteron Processor 280 (2.4Ghz)
    4GB Ram
    CentOS release 4.5, i686 (with PAE kernel)
    3Ware 9550SX-8LP RAID card, BBU, cache enabled
    4 x 250GB SATA disks (WD2500YS), RAID 10 configuration

    I'm inclined to overhaul the RAID array with a) more, and b) faster disks. But before I go throw my company's $$ at 10k SATA disks, not to mention my time setting up and migrating to a different RAID device, I'm curious if there are software settings I should be tuning instead.

    What kinds of hardware are other people using for similar numbers of users? How have you dealt with scaling Zimbra?

    We started using Zimbra about a year ago with 25 users and great performance. Now I've got unhappy users who want to switch to Google Apps, or worse, hosted Exchange. Egads!

    Thanks for your advice,

    Dane
    Attached Images Attached Images
    Last edited by daneturner; 07-03-2008 at 02:35 PM.

  2. #2
    Join Date
    Nov 2006
    Location
    UK
    Posts
    8,017
    Rep Power
    25

  3. #3
    Join Date
    Jul 2007
    Location
    San Jose, CA
    Posts
    1,027
    Rep Power
    10

    Default

    You shouldn't need that much horsepower for only 75 users unless their traffic is way off the charts. You mentioned Top so you must be looking in the background, but have you looked to see if other non-Zimbra processes are running that might be affecting your performance?

    You also mentioned you are on 4.5.5. That's WAY old by now. You ought to try upgrading to a more current release as lots of bugs and performance issues have been fixed along the way.
    Cheers,

    Dan

  4. #4
    Join Date
    Mar 2007
    Posts
    24
    Rep Power
    8

    Default Memory usage

    I overlooked memory usage. I'm not a pro at Linux paging stats, but this looks bad. sar shows lots of page-in and page-out activity throughout the day. For example, here's a snippet from the morning rush...

    sar -f /var/log/sa/sa01 -B

    Linux 2.6.9-42.0.10.ELsmp (zimbra.greatschools.net) 07/02/2008
    12:00:01 AM pgpgin/s pgpgout/s fault/s majflt/s
    08:10:01 AM 576.22 595.11 1934.28 0.06
    08:20:02 AM 5130.10 2846.09 6250.25 0.07
    08:30:01 AM 672.49 679.60 1796.73 0.13
    08:40:04 AM 4860.63 2626.31 1844.63 0.07
    08:50:01 AM 903.63 1123.91 2136.01 0.11
    09:00:04 AM 3981.61 2219.43 1981.76 0.06
    09:10:01 AM 1919.45 2060.98 2242.85 0.08
    09:20:01 AM 187.42 570.69 6418.49 0.12
    09:30:01 AM 3904.70 2195.18 1792.63 0.06
    09:40:01 AM 1883.21 2265.08 2165.67 0.06
    09:50:01 AM 49.80 436.26 1907.43 0.06

    Sorry for the formatting ugliness. That's averaging 2.5MB/sec paging in from disk and 1.7MB/sec paging out to disk.

    Could this be the cause of high load and high disk activity?

  5. #5
    Join Date
    Mar 2007
    Posts
    24
    Rep Power
    8

    Default zimbra version

    Oops, I reported the wrong Zimbra version. We're using 4.5.10

    Release 4.5.10_GA_1575.RHEL4_20071116233533 CentOS4 NETWORK edition

  6. #6
    Join Date
    Jul 2007
    Location
    San Jose, CA
    Posts
    1,027
    Rep Power
    10

    Default

    Quote Originally Posted by daneturner View Post
    I overlooked memory usage. I'm not a pro at Linux paging stats, but this looks bad. sar shows lots of page-in and page-out activity throughout the day. For example, here's a snippet from the morning rush...

    <snip>

    Sorry for the formatting ugliness. That's averaging 2.5MB/sec paging in from disk and 1.7MB/sec paging out to disk.

    Could this be the cause of high load and high disk activity?
    Yes, but this doesn't really tell us WHY it's happening. SOMETHING is overloading your memory and swap, but we don't know WHAT.

    Seriously, I would recommend you upgrade AT LEAST to 4.5.11 if not to a 5.x release and see if that doesn't solve your problems. Those middle 4.5.x releases did have some bugs that inflicted a variety of kinds of pain.

    But the other thing you might consider is just stopping all Zimbra services (zmcontrol stop) and then doing a top and sar and see whether something is still hammering on your system. Because as I said before your problem is categorically NOT hardware. . .the hardware you have is more than sufficient for your user base (unless a component is actually failing). Throwing more hardware at the issues you are experiencing is a Micro$oft solution, not a Zimbra/Linux one!

    Edit: OK if you are on 4.5.10 I'm less sure the version could be the problem, (though you still should consider upgrading to 5), but it'd still be a good idea to stop Zimbra services and see if the bad usage continues.
    Last edited by dwmtractor; 07-03-2008 at 12:05 PM. Reason: Version change reported by OP
    Cheers,

    Dan

  7. #7
    Join Date
    Jul 2007
    Location
    San Jose, CA
    Posts
    1,027
    Rep Power
    10

    Default

    Quote Originally Posted by daneturner View Post
    Oops, I reported the wrong Zimbra version. We're using 4.5.10

    Release 4.5.10_GA_1575.RHEL4_20071116233533 CentOS4 NETWORK edition
    You realize that with Network edition (I had assumed you were FOSS) you should be able to get a Zimbra support tech to actually get under the hood with you and see what's ticking? Have you tried a support call?
    Cheers,

    Dan

  8. #8
    Join Date
    Mar 2007
    Posts
    24
    Rep Power
    8

    Default other services running

    You mentioned Top so you must be looking in the background, but have you looked to see if other non-Zimbra processes are running that might be affecting your performance?
    Top shows only zimbra processes using CPU time. I've attached a screenshot showing the typical top output (pasting text loses too much formating).

    I see two mysqld processes running. Is that normal?
    Attached Images Attached Images

  9. #9
    Join Date
    Jan 2008
    Posts
    658
    Rep Power
    8

    Default

    Two mysqld processes is normal. I believe one's logger and the other is the Zimbra mysql. What does 'free -m' show?

  10. #10
    Join Date
    Mar 2007
    Posts
    24
    Rep Power
    8

    Default free -m

    [dane@zimbra ~]$ free -m
    total used free shared buffers cached
    Mem: 4042 4026 15 0 18 1465
    -/+ buffers/cache: 2542 1499
    Swap: 4095 164 3931
    Attached Images Attached Images

Similar Threads

  1. Setup Yahoo business email account
    By muscles614 in forum General Questions
    Replies: 9
    Last Post: 11-20-2012, 03:10 PM
  2. Replies: 4
    Last Post: 06-20-2008, 05:57 AM
  3. Upgrading to 5.x... and on raid 5 now...
    By timothyalangorman in forum Administrators
    Replies: 1
    Last Post: 05-01-2008, 07:35 AM
  4. RHCS setup
    By Klug in forum Installation
    Replies: 1
    Last Post: 05-29-2007, 01:59 AM
  5. zimbra email dns setup
    By robharpham in forum Installation
    Replies: 4
    Last Post: 05-13-2007, 07:25 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •