Advice on geo-replication for huge files and snapshotting

[deajan]

Hello,

First of all, sorry if this post does not really belong to github issues. I've had a hard time reading a lof of glusterfs documentation without finding my answers, so I decided to directly ask the glusterfs developpers for help.

I want to geo-replicate KVM virtual machines from machine A to machine B.

So far, I image setting up machine A (single node) to unidirectionally replicate qemu disk images to machine B.
What I'm desesperately searching for is infos about:

• Replication process. I've read it's rsync based. Does that mean that on every replication schedule, the rsync based task will have to read all changed source and destination files to compare and know which chunks to send ? Of course, all qcow2 files change every second. But reading terabytes of data on each schedule seems quite IO hungry. Did I get that right ?
• Snapshotting process. Can glusterfs do XFS snapshots in order to send the file updates without the file being changed while sending it ? If so, is there a hook I can use to launch freeze/thaw guest FS scripts ?

Perhaps Geo-replication exists for "end user file serving" instead of the use case I want ?

Again, sorry if this isn't the best place to post this.
Thank you for any insight.