Results 1 to 6 of 6

Thread: Best Practise For Large Mailboxes

  1. #1
    Join Date
    May 2008
    Posts
    12
    Rep Power
    7

    Default Best Practise For Large Mailboxes

    Hi all,

    in the near future we would like to offer a groupware service to our customers (small an medium-sized businesses). Our service should scale up to 10.000 Users and more. Zimbra looks very interesting for that. At the moment we are looking at the system requirements for our service. We think that nowadays you have to offer large mailboxes with 25 GB each to be competitive. Google Apps Professional Edition, Microsoft Online Services and 01.com (Zimbra!) providing 25 GB mailboxes.

    To get 25 GB mailboxes at an affordable price it is very important to select the right storage model and storge components. I think it is all about DAS vs. SAN and SATA vs. SAS.

    I already searched through the forum. It looks like most people are recommending SAN-based systems for large deployments using a combination of SAS and SATA drives with Hierarchical Storage Management. This is of course not the most expensive solution but maybe also not the cheepest one.

    I found a Document for Microsoft which is describing the same situation for MS Exchange systems. The title is "Exchange 2010 Large Mailbox Vision Whitepaper". It can be found here:
    Download details: Exchange 2010 Large Mailbox Vision Whitepaper

    As far as I understand, in this document MS recommends using multiple servers with a DAS storage attached to each server. They seem to use cheep SATA drives with 1 TB only. They say they can use cheap drives because
    a) MS Exchange Server 2010 has lowered IOPS/user to 0,1 and
    b) with large mailboxes of 2 GB each you have 500 users per disc which is fine

    Because the server are replicated they get a lot of capacity, high availability and no need for backups.

    My question is: does the MS approach apply to Zimbra as well? What do the experts in this forum think about the right way of deploying large mailboxes with low budget?

    Best regards,
    Henning

  2. #2
    Join Date
    Sep 2006
    Location
    477 Congress Street | Portland, ME 04101
    Posts
    1,374
    Rep Power
    11

    Default

    Hi Henning,

    We are a Zimbra Hosting provider and also support Exchange environments. Our last Exchange deployment was a 2010 system with WAN replication and DAG groups as described in the article you mentioned, so we have good experience with both environments.

    The architectural differences between Zimbra and Exchange 2010 are great, so to answer your question, no, I don't believe the large/slow disk everywhere approach espoused by Microsoft in the article applies to Zimbra directly.

    In our experience, when you have a 10,000 mailbox system, doing it with DASD requires a lot more physical spindles and add-on disk shelves than doing it with a SAN. At some point, the SAN becomes more cost effective, especially because with a SAN you have the opportunity for more efficient disaster recovery and load balancing.

    For a Zimbra system of that size, I would recommend a two compute head system with 24-cores each connected to a single SAN (your choice of FC or iSCSI).

    If you licensed VMware's DRS, and three of your virtual mailbox servers get hammered, VMware will automagically VMotion virtual machines around to spread the load between the compute heads.

    Upgrading Zimbra or installing a Service Pack to the operating system? Take a snapshot in VMware and if the upgrade totally borks and you run out of time, copy off the log files and roll back.

    Need off site disaster recovery? Take a snapshot, take a backup and copy the backup off site.

    Need off site business continuity? Get a second SAN, drop it in a second data center and do SAN replication over the WAN.

    Don't get me wrong, for single Zimbra servers and even smaller Zimbra multi-server installations, DASD is fine. For a 10,000 mailbox system, I think you will find a modest SAN to be overall more cost effective, especially considering the added reliability and availability benefits.

    Hope that helps,
    Mark

  3. #3
    Join Date
    May 2008
    Posts
    12
    Rep Power
    7

    Default

    Mark, thank you very much! Your answer is very helpful to us.

  4. #4
    Join Date
    May 2008
    Posts
    12
    Rep Power
    7

    Default

    Some more thoughts:

    To get big mailboxes at an affordable price I think it is worth to consider using HSM. If I understand correctly HSM works like this:

    You define a period of time after which messages are transfered to a secondary storage volume where you use cheep and slow disks. If you use a period of 5 days the mail you receive today is now kept on the primary storage (with fast, expensive disks) and gets automatically transfered to the secondary volume after 5 days. All mails on the secondary store are still visible to the user but it takes longer to access them. Because we give our users 25 GB of space we expect them to keep their messages for several years.

    My questions are:

    1. Is my understanding of HSM correct?
    2. Are messages on the secondary store (same SAN but SATA drives) accessible in an acceptable time?
    3. What do you think about the time period of 5 days?
    4. With HSM and a time period of 5 days we should see 99 % of all messages on the secondary store after a while?

    Thanks in advance.

    Rgs,
    Henning

  5. #5
    Join Date
    Sep 2006
    Location
    477 Congress Street | Portland, ME 04101
    Posts
    1,374
    Rep Power
    11

    Default

    Quote Originally Posted by schogge View Post
    Some more thoughts:

    To get big mailboxes at an affordable price I think it is worth to consider using HSM. If I understand correctly HSM works like this:

    You define a period of time after which messages are transfered to a secondary storage volume where you use cheep and slow disks. If you use a period of 5 days the mail you receive today is now kept on the primary storage (with fast, expensive disks) and gets automatically transfered to the secondary volume after 5 days. All mails on the secondary store are still visible to the user but it takes longer to access them. Because we give our users 25 GB of space we expect them to keep their messages for several years.

    My questions are:

    1. Is my understanding of HSM correct?
    2. Are messages on the secondary store (same SAN but SATA drives) accessible in an acceptable time?
    3. What do you think about the time period of 5 days?
    4. With HSM and a time period of 5 days we should see 99 % of all messages on the secondary store after a while?

    Thanks in advance.

    Rgs,
    Henning
    Henning,

    I didn't mention HSM originally because it is documented as the standard method for using lower-cost storage so I presumed you would make use of it.

    HSM policies can be configured in the Admin Console or at the CLI, and are run as a job. You can schedule the HSM job to run via cron if you like.

    How often to run the HSM job, how many days should be in your policy, and what items should be moved to HSM are entirely up to you and should be based on how your users will use the system.

    For example, we never move Contacts to HSM drives. just because a contact was entered two years ago doesn't mean the user wants to wait any longer to see it.

    Unless you are running very small disks for /opt/zimbra/store, 5 days sounds aggressive. If the majority of your mail store is on a set of slow HSM drives and users access old items frequently, then yes, you will experience noticeable performance degradation.

    SATA drives have gotten much better in recent years, but their random I/O numbers still lag fast disks and/or a SAN. And to be clear, mail store access is pretty much nothing but random I/O.

    Consequently, though others do use SATA drives, we and our clients are very impatient and so we do not have any SATA drives deployed at all, even for our HSM DASD volumes.

    Hope that helps,
    Mark

  6. #6
    Join Date
    Dec 2010
    Posts
    38
    Rep Power
    5

    Default

    I got a quick question. Regarding the mail boxes. I am newbie on using zimbra, and have been following the discussion.

    I want to make zimbra read and write any data, regarding the user frim a remote data server. Would zimbra allow to make any changes to the service to copy all the emails and personal data to a remote server on the network and read from the same. For example, External LDAP, it reads from external AD to check the user authentication.

    Is there any provision like tht in Zimbra. Any inputs are very helpful.

    Thank you.

Similar Threads

  1. Multiple Mailboxes in Zimbra Web Client?
    By talljon in forum Administrators
    Replies: 17
    Last Post: 08-29-2014, 05:17 AM
  2. auto-provision mailboxes in postini service
    By speculatrix in forum Administrators
    Replies: 2
    Last Post: 12-03-2007, 10:38 AM
  3. Replies: 11
    Last Post: 11-02-2007, 10:23 AM
  4. Restoring old mailboxes?
    By RobLarose in forum Installation
    Replies: 6
    Last Post: 09-25-2007, 01:55 PM
  5. From POP3 and 100 scattered mailboxes to Zimbra
    By spiderman in forum Administrators
    Replies: 5
    Last Post: 08-11-2006, 09:00 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
  •