[Dovecot] Redundant shared mail store?
Hi guys
I'm about to embark on a project replacing a legacy single mail server running proprietary software with a cluster of Dovecot servers. We clearly need a shared mail store for these servers, but we are not happy with simply using a server running NFS as this becomes a single point of failure.
What have other people used? AFAICT our options are:
* A shared block device (Fibre Channel, iSCSI, ATAoE etc) running GFS
or similar to allow multiple servers to access it concurrently
* Two or more NFS units, either with built-in support for replication
of FS changes and failover, or with scripts for achieving this functionality
I prefer option 1, although it's expensive. What have other people used? Do we have other options?
TIA
Jasper Bryant-Greene Director Album Limited
jasper@albumltd.co.nz +64 21 708 334 / 0800 425 286 http://www.albumltd.co.nz/
Jasper Bryant-Greene writes:
I prefer option 1, although it's expensive. What have other people used? Do we have other options?
How redudant does it need to be? How big is the setup? Does whatever OS you plan to use, works well with NFS?
I use FreeBSD as the OS and we had terrible experience with NFS. I think NFS servers are probably the way to go... if you have the budget, but that was not an option for us.
We are planning on having a backup/spare server and doign RSYNCS to it, but that may not be acceptable if you don't the possibility of loosing any mail.
We are planning on having a backup/spare server and doign RSYNCS to it, but that may not be acceptable if you don't the possibility of loosing any mail.
We are using drbd and it works very well. Combined with Heartbeat and some APC powerswitches it's foolproof at fail-overs. Use a direct gigabit link between the servers using 9000 frames bytes proves to be enough to keep everything 100% in-sync. The additional mbox'es are placed in the users home directories, also high available NFSv3 servers using drbd. We have more then 20TB realtime syncronized using drbd 0.7.22; so far never lost a single bit. Use the seperate partition for meta- data so you can mount a 'none-drbdified' partition at anytime if troubles would emerge (never had to use it but better to be safe then sorry).
Our golden combination consists off: Beefy hardware with Areca 1260 raid controllers RedHat enterprise 4U4 Drbd 0.7.22 Heartbeat 1.2.4 bunch of scripts for the fail-over to control the power switches ( down means really go down using the stonith feature )
Goodluck, Leroy
Anyone looked at using GFS. I found it very easy to use, reliable and pretty fast. I ran it on RedHat ES.
This gives me the flexabiltiy of having multiple hosts connected to the same volume. The more hosts, the more load balancing and redundancy you have. The storeage would preferably be on a SAN of course.
-----Original Message----- From: dovecot-bounces@dovecot.org [mailto:dovecot-bounces@dovecot.org] On Behalf Of Francisco Reyes Sent: 29 November 2006 12:51 AM To: Jasper Bryant-Greene Cc: dovecot@dovecot.org Subject: Re: [Dovecot] Redundant shared mail store?
Jasper Bryant-Greene writes:
I prefer option 1, although it's expensive. What have other people used? Do we have other options?
How redudant does it need to be? How big is the setup? Does whatever OS you plan to use, works well with NFS?
I use FreeBSD as the OS and we had terrible experience with NFS. I think NFS servers are probably the way to go... if you have the budget, but that was not an option for us.
We are planning on having a backup/spare server and doign RSYNCS to it, but that may not be acceptable if you don't the possibility of loosing any mail.
participants (4)
-
Francisco Reyes
-
Jasper Bryant-Greene
-
Leroy van Logchem
-
Richard Mayhew - Nashua Mobile Broadband Division