[DRBD-user] Moving a DRBD cluster from physical machines to VMWare machines

Pascal BERTON pascal.berton3 at free.fr
Fri May 25 10:30:14 CEST 2012

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


Keith,

I'm not sure it's a good idea, however the first thing to avoid is to
present the replicated disk space to your virtualized peers as vDisks on a
VMFS. Use RDMs instead, in physical mode if possible, to grab some more
perfs.
Then for this RDM, use a vdisk node that will cause the instantiation of a
new vSCSI adapter (not the 0:1 default, but rather 1:0 for instance). It
will probably create an LSI Logic Parallel virtual adapter, change it to the
vmware paravirtual scsi driver, it will help a lot too.
Then, not sure it has a big impact on DRBD but VMware can generate IOs up to
32MB. It's a known perf issue for midrange storage arrays such as HP EVAs
(and even EMC Clariions I guess) causing big slowdowns due to IO
segmentation handling by the storage processors, we use to reduce it to 64k
or 128k instead. Again, I'm not sure it would have an impact on DRBD, may be
the gurus can give their advices about that ? Would it even make sense to
decrease it to 4K since DRBD chuncks are 4K large ? Or is it another part of
the kernel storage chain that's in charge of handling large IOs
segmentation, in which case we don't mind at all from the DRBD standpoint...
Then what else ? I think that's all. I will think of it a bit more and let
you know if something comes up.
BTW, the physical storage space intend to replicate is local storage, right
? Or shared external storage device ?

Regards,

Pascal.

-----Message d'origine-----
De : drbd-user-bounces at lists.linbit.com
[mailto:drbd-user-bounces at lists.linbit.com] De la part de Keith Christian
Envoyé : mercredi 23 mai 2012 17:29
À : drbd-user at lists.linbit.com
Objet : [DRBD-user] Moving a DRBD cluster from physical machines to VMWare
machines

Looking for guidance from Linbit, or the list, about DRBD's disks remaining
on two separate physical servers, except that the peer servers are
virtualized?

For instance: Writes of approximately 18,000 blocks per second for the busy
hours, as reported by SAR.  How much a performance hit is there with the
same hardware after VMWare is in the picture?  (I'm sure there are VMWare
config settings to optimize disk reads/writes for an application like DRBD.)


Thanks.


=====Keith
_______________________________________________
drbd-user mailing list
drbd-user at lists.linbit.com
http://lists.linbit.com/mailman/listinfo/drbd-user




More information about the drbd-user mailing list