[DRBD-user] One vs many DRBD resources

Maciej Gałkiewicz maciejgalkiewicz at ragnarson.com
Fri Sep 23 14:47:33 CEST 2011


Hello,

I would like to implement DRBD with LVM volumes exported by iSCSI to
virtual machines. Here (http://dl.dropbox.com/u/5820195/storage.svg)
are two stacks which I am considering. I have tens of virtual
machines. Each of them mounts 5GB partition(s) (the size may change)
accessible through iSCSI.

In the first case (#1) I have one big drbd volume which stores LVM
volumes for instances. When primary node crashes all I have to do is
to promote the other one. The problem is that in case of maintenance
when I need to shutdown primary node I have to migrate all LVM volumes
at once. All VMs do not have access to remote partition.

The second stack (#2) is more flexible. I can migrate one volume
without interrupting the rest. I am concerned about network traffic
during initial drbd synchronization. Each resource will have about 5GB
so when I add a new one the synchronization generates high traffic. Is
there any way to avoid it? What about resizing LVM volume? How much
data drbd will transfer assuming that the volume size will be +/- 1GB?

Could you please advice me which implementation would be better and
point any expected inconveniences?

-- 
Regards
Maciej Galkiewicz



More information about the drbd-user mailing list