I'm currently using a two node combined virt/storage setup with Ovirt 3.3.4 and Gluster 3.4.2 (replica 2, glusterfs storage domain). I'll call this pair PROD.

I'm then geo-replicating to another gluster replica pair on the local net, btrfs underlying storage, and volume snapshots so I can recover my storage domain from different points in time if necessary. Its also local so restore time is much better than off-site. I'll call this pair BACKUP.

I'm planning on setting up geo-replication from BACKUP to an EC2 gluster target. I'll call this host EC2HOST.

PROD ---geo-rep-lan---> BACKUP ---geo-rep-wan---> EC2HOST

I'd like to avoid saturating my WAN link during office hours. I have some ideas (or combination of):

1. limit bandwidth during certain hours to the offsite hosts. But realistically the bandwidth I would allocate is so low I don't see the purpose of this. Also with 8 guests running, I'm noticing quite a bit of data transfer to the local backup nodes (avg 6-8MB/s), and I'm thinking there is a lot of thrashing going on which isn't useful to backup offsite anyways.

2. stop WAN geo-replication during office hours, and restart for overnight/weekend hours.

3. Not use geo-rep between BACKUP ---> EC2HOST, use rsync on one of the btrfs volume snapshots so we avoid the thrashing. In this case I could limit WAN speed to 1MB/s which should be fine for most differences throughout the day.

So my question is, how do you off-site your storage domains, what constraints have you identified and how have you dealt with them? And of course how would you deal with the scenario I've oulined above?

Thanks,