[DRBD-user] Re: MD Raid-0 over DRBD or DRBD over MD Raid-0

Rick Rothstein rickrsr2 at yahoo.com
Fri Feb 16 18:51:07 CET 2007

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


Greg Freemyer <greg.freemyer at ...> writes:
> 
> > However, the OP was talking about running GFS on the DRBD device, and
> > that only really makes sense if you're going dual active.
> 
> Rick, I missed that you wanted to do active-active.  If that is indeed
> the case I'm afraid I see where Graham is coming from.
> 
> You could likely make it work by ensuring md was offline on one of the
> nodes and online on the other anytime you did any maintenance to the
> md setup.  I assume you don't have any way concocted to use a
> hotspare, but you definitely could not do that sort of thing with this
> setup.
> 
> Given that you're talking raid0 from a md perspective, I don't see a
> lot maintenance anyway so that may be a very livable situation.
> 
> As to:
> 
> "You can stack DRBD on top of md, md on top of DRBD is nonsense, however".
> 
> I still don't know why that is true in a active-passive setup.  It
> seems totally logical to me to stack md raid0 above drbd raid1 in a
> active-passive setup, but that may be different thread.
> 
> Greg

Thanks for the advice.

In case there's some experience/interest in related but not totally DRBD stuff 
that has to get done, here's a bigger picture of all that's required:

Actually, over time, reliable raid-0 maintenance 
(primarily growing the stipe using EVMS), will be important.  

Some experimenting is going on to see if the combination of EVMS and GFS2, 
along with some mods, will enable growing the raid-0 stripe in the background 
without having to take the array offline.

The initial configuration will be a raid-0 stripe over three 16_drive boxes, 
using ISCSI, (along with an additional three miror machines); with GFS2 
providing a single, global namespace over all this data. 

So when more storage is needed, the plan is to roll in another box, grow the 
raid-0 stripe to encompass the new box, (using EVMS), and then grow the GFS2 
filesystem to encompass the new storage.

Hopefully, the DRBD mirroring will be manageable because it will be across 
single physical disks, and the raid will be manageable because it will be a 
simple raid-0 (i.e. no parity), with each device of the stripe protected by a 
DRBD mirror.

The DRBD setup does not have to be active-active, but GFS2 is alsreay required 
for the clustered raid-0; and it would be nice to actually get some added 
processing power, and not have those mirror boxes just waiting around for bad 
stuff to happen.

_Rick Rothstein





More information about the drbd-user mailing list