For a new e-mail cluster that will eventually consist of a number of #
IMAP servers, I need some shared storage that all servers can read from and write to at the same time. I have some experience with #
DRBD, but I was told that DRBD isn't going to be the solution for what I want.
It will start small, with only several hundreds of mailboxes, but it should be able to scale up to many thousands or even hundreds of thousands in the distant future. What I want is a variable number of #
Dovecot servers with an HA Director in front of them, so that I can upgrade and reboot individual nodes without users noticing.
#
NFS would -on paper- be ideal, but when working with lots and lots of small files it gets very slow. #
ZFS is pretty cool, but that wasn't designed as a cluster filesystem and I'm not sure if it can be made to do that reliably. I find #
GlusterFS and #
GFS2 in many articles, but I don't have any experience with those. I have a bit of experience with #
Ceph, just enough to know that I don't want that.
What do you guys think, what is the system I should go for? And, or course, why? Did I overlook systems that are worthwhile?