Prev: How to reject bad hosts
Next: statvfs errors
From: Jeroen Geilman on 10 Aug 2010 14:29 On 08/10/2010 10:05 AM, Bjorn Mork wrote: > Hi, > can POSTFIX handle load of 120k mailboxes.... Since "postfix" has relatively little to do with "mailboxes", and a "mailbox" is most definitely not a unit of load, I'd say... VERMILION. These mailboxes/accounts presumably have users associated with them, who either send lots of mail, very little mail, or no mail. They will also receive lots of mail, a bit of mail, or no mail. However, they will ALL receive massively ridiculous amounts of SPAM. Furthermore, they will check or download their mail every minute, every day, or never. They will use a MUA or a web client, or a forwarder, or even complicated scripting and sieve filters on their mailboxes. All of these are factors you need to measure and analyze in order to sanely scale such a setup. For instance, if you will mostly be receiving mail and sorting it into mailboxes, you need both a capable (redundant) frontend as well as a fairly massive enterprise-worthy redundant storage backend. Neither iSCSI nor NFS (shudder) will suffice for 120 THOUSAND mailboxes, if you're talking about one single (presumably not very large) backend. I'd strongly suggest you investigate FC, or else cook up a solution that can line-bond 2 or 4 gbit links to an iSCSI backend. Even better would be to run a distributed filesystem (such as AFS) on multiple backends, so that your world doesn't end if one of them happens to go down in flames. But if you absolutely must run into this plan half-cocked, you should start by using components that you know you can scale up and scale out later on. Good luck, you have a lot of research and testing to look forward to! > What would be required additional with postfix to bear such load.... I'd probably run all of it on one of our 8-core vm hosts - 22GHz, 72GB memory and 3000 IOPS FC should do nicely... (That was a joke - it wouldn't be redundant at all) J. |