Dovecot mailstore performance tuning

Timo Sirainen tss at
Tue Jul 29 11:16:10 UTC 2014

On 22 Jul 2014, at 04:57, Murray Trainer <mtrainer at> wrote:

> We have a couple of dovecot director proxies and six backed mailstores
> each accessing mailboxes stored on five NFSv4 filsystems with about
> 1TB of mail on each in maildir format.  We have about 800 max users
> on each mailstore at peak times and performance appears to starting to
> degrade at these times.  The mailstores are pretty recent hardware
> with 64GB of RAM and 24 cores.   The NFS storage is EMC VNX and we
> are doing about 250 I/O per sec upto max of 500 on each
> filesystem.   I need to squeeze more performance out of these
> servers whether that is in the NFS client, Dovecot or Linux OS/kernel
> areas.   We use LDAP for auth and I am doing some tuning in that
> area.   The NFS filesystems are mounted with the options below:
> on /home1 type nfs4
> (rw,relatime,vers=4.0,rsize=65536,wsize=65536,namlen=255,hard,nordirplus,proto=tcp,port=0,timeo=600,retrans=2,sec=sys,clientaddr=,local_lock=none,addr=

Does relatime work with NFS? If yes, changing it to noatime would save some I/O.

maildir_very_dirty_syncs=yes should be helpful.

> # 2.2.9: /etc/dovecot/dovecot.conf

mailbox_list_index=yes might be useful, although it has had some further performance improvements since v2.2.13. I should try to make v2.2.14 soon..

>   quota = maildir

Dict file quota would be a bit faster than maildir++ quota.

More information about the dovecot mailing list