[DRBD-user] Slow NFS behaviour (0.7.5)
crosser at rol.ru
Wed Nov 24 15:55:58 CET 2004
Philipp Reisner wrote:
>>/me curses the sysadmin fate...
>>At the moment I'm thinking how should I explain to users, that every mount
>>is now exported async, over tcp and we're having ~ 1GB buffers on the
>>server just to mitigate the commits load. Then you get the drbd (which, in
>>it's fairness is protocol C; but I don't know why, as I could as well use
>>A with with current settings of al-extents 521; and max-buffers 131072;).
>>Is there really no way to have transaction safe NFS server on DRBD that
>>doesn't crawl like rats-assed three legged turtle on valium? :>
> My conclusio is:
> DRBD: Protocol C
> Filesystem on DRBD: async (default)
> NFS export: sync (new default)
> If performance is too low "tune" NFS by making the export async.
my setup for mailbox store NFS server:
DRBD: protocol C
Filesystem on DRBD: journal=ordered(? - default for ext3), noatime
NFS export: sync,no_wdelay
Mounted on clients as: nolock,nfsvers=3,rsize=32768,wsize=32768
(I use lock file locking in delivery agent and pop/imap servers)
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 256 bytes
Desc: OpenPGP digital signature
More information about the drbd-user