[DRBD-user] Question about server structure

Bart Coninckx bart.coninckx at telenet.be
Thu Dec 23 21:51:57 CET 2010

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On Thursday 23 December 2010 21:24:07 Tim Mauerbach wrote:
> 2010/12/23 J. Ryan Earl <oss at jryanearl.us>
> 
> > On Thu, Dec 23, 2010 at 11:23 AM, Tim Mauerbach <
> > 
> > tim.mauerbach at googlemail.com> wrote:
> >> I am thinking about the structure of a new mailserver on top of
> >> raid1/lvm/xen.
> >> 
> >> Two different approaches come to my mind:
> >> 
> >> 1. One big LV as DRBD backing device for the whole guest (os + data).
> >> 2. One small LV without drbd for the base system + one big LV as DRBD
> >> backing device only for data (maildirs+mariadb).
> >> 
> >> Is the maintenance overhead of 2 worth the performance gain?
> > 
> > I'd definitely go with 1, and it has nothing to do with performance.  If
> > you put the whole VM on top of DRBD, and then make that a primary/primary
> > (aka active/active) DRBD then you can live-migrate the VM between host
> > and/or use Remus VM-mirroring for VM-HA.  I don't see why anyone would
> > ever go with 2 as leaving the OS unreplicated on the small LV exposes
> > you to big potential downtime as you rebuild a new OS for the
> > replicated-data in the event of a failure.  Am I misinterpreting what
> > you mean?
> > 
> > -JR
> 
> Thanks for your input.
> You´re right. It has nothing to do with performance, I meant efficiency:
> 
> The idea of 2 is to prevent live-replication of not that important non-user
> data (os + tmp|log files) that could be replicated via rsync at midnight.
> Thus I could use my DRBD capacity more efficiently and only for important
> data, though at the expense of maintainability.
> 
> Best Regards
> Tim

When you scale your hardware correctly, you can get awesome results with DRBD. 
I second JR's suggestions. Just out of curiosity: what mailsystem, how many 
mailboxes and what storage are we talking?

b.



More information about the drbd-user mailing list