[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Clustering mail servers - Cyrus or Courier ?

On Sat, 11 Aug 2001, Russell Coker wrote:

> On Sun, 5 Aug 2001 16:10, Przemyslaw Wegrzyn wrote:
> > I'm looking for good solution for big, scalabale mailserver installation,
> > for 500,000 accounts and more...
> >
> > As for now I think:
> >  - sharng NFS-mounted mail storage is not so good idea.
> Why not?  If you use Maildir storage then it's NFS safe.  If you use a NetAPP 
> Filer, a VA NAS device (or any machine running 2.2.x kernel with LVM, Ext3, 
> and appropriate NFS kernel-server patches as the VA NAS devices did), or some 
> similarly powerful NFS server then performance should be more than adequate.

Hmmm, I'm really afraid about NFS, I've heared that Linux NFS
implementation is broken, many times. Can anyone explain it a bit ?

> For Ext3 you need to keep the number of entries in a directory down to <1000 
> for best performance, but most users don't have that many messages in their 
> mail folder.  So you just have to do suitable directory hashing to make sure 
> you don't have too many users in a directory (all mail storage software has 
> support for this).

Yeah, I do such hushing on every bigger mailserver, it's a common

Currently I'm testing qmail-ldap + maildrop + Courier-IMAP. 
Works well, but it's not HA solution :|
> Then you can have multiple mail servers running at the same time for 
> increased reliability.  This is much better than the Perdition approach of 
> increasing the number of points of failure.  Having an NFS file server as a 
> single point of failure is as good as it gets, making an NFS file server 
> reliable is much easier than making a mail server reliable...

Running multiple mail servers over NFS will actually makie it HA-like
system. But is it good solution for high load ? 
Using perdition-like solution to split mailspool between separate NFS
servers, and using multiple mailservers for each NFS backend would
probably build good scalable solution.


Reply to: