[Dovecot] GlusterFS

[Pawel Panek]

From: "Ed W" <lists at wildgooses.com>
Sent: Sunday, August 10, 2008 11:09 AM


> I'm also interested to hear how it works out?  It appears that the 
> straightline speed is high for gluster, but it's per file performance has 
> enough overhead that it's a signficant problem for maildir type 
> applications which manipulate lots of small files?  Possibly it works very 
> well if you go mbox though?

FUSE kernel driver from 2.6.24 was unusable. Fuse client saw changed file 
modes ie: from 640 to 666. With fuse driver delivered with glusterfs file 
modes were the same as on exporting server. It was performing very well when 
clients were moving some large files, but when it comes to mail traffic wait 
time and system load on client nodes started increasing. At last glusterfs 
stop working due segfault in io-cache.so or libglusterfs.so. Mail nodes were 
using glusterfs-1.3.7 and fuse-2.7.2glfs8, mail was delivered into maildirs

Pawel.