cluster replication between two locations

[Gerhard W. Recher]

Hi Folks,

I plan a proxmox cluster for a customer, he has a dedicated line between two location, but only 10 Mbit/s synchronous.
vm's will be (windows ad terminalserver , exchange, m$-sql server) clients connect via rdp on Terminalserver .

We want to setup a cluster on both sites , so in case one site dies (fire/Power outage/.... ) users on the other site can work with the local replicated instance.

Is this possible, or will sync traffic saturating a 10 Mbit/s line, so no regular rdp traffic will get through ?

Backup site vm's will be started manually.

any directions ?

 

[jeffwadsworth]

Did you see this thread?

https://forum.proxmox.com/threads/proxmox-ha-cluster-disaster-recovery.30791/

 

[Gerhard W. Recher]

Did you see this thread?

https://forum.proxmox.com/threads/proxmox-ha-cluster-disaster-recovery.30791/

yep i saw this, but we want to build ceph storage and replicate this other this thin-pipe .... this article refers to zfs

 

[jeffwadsworth]

This might shed some light on that task:

http://docs.ceph.com/docs/master/rbd/rbd-mirroring/

https://forum.proxmox.com/threads/rbd-mirror-support.33298/

 

[Gerhard W. Recher]

This might shed some light on that task:

http://docs.ceph.com/docs/master/rbd/rbd-mirroring/

https://forum.proxmox.com/threads/rbd-mirror-support.33298/

@jeffwadsworth Thanks for these directions, but i see no real live solution yet, or have i missed something.
also bandwidth calculation and latency demands for rbd-mirror are not known for me ...
has anyone implemented this scenario ?

 

[guletz]

Hi,

When you want data replication between 2 remote sites, the best is zfs send/receive(like incremental backup at block level). I do not know if your dedicated inter data center link speed (10 Mb/s is enough or not. In this situation your best efforts is:

- use ceph with zfs as backend (so you can do zfs replication)
- replicate very often the most important data (will be finish faster)

When your primary DC is recover after a fail, you will be able to reverse the replication ( maybe during off-work period).

Try to imagine the non-zfs scenario: maybe multi-TB sync? Or only 1 TB with multi-billion files?

 

[Gerhard W. Recher]

Hi,

When you want data replication between 2 remote sites, the best is zfs send/receive(like incremental backup at block level). I do not know if your dedicated inter data center link speed (10 Mb/s is enough or not. In this situation your best efforts is:

- use ceph with zfs as backend (so you can do zfs replication)
- replicate very often the most important data (will be finish faster)

When your primary DC is recover after a fail, you will be able to reverse the replication ( maybe during off-work period).

Try to imagine the non-zfs scenario: maybe multi-TB sync? Or only 1 TB with multi-billion files?

typical windows server ... workload tons of files in windows filesystem, dedicated sql-server, and exchange server ...
so transparent blockdevice replication would be best ....
existing line 10mbit/s leased line is now only used for 15 Workstations RDP traffic.. QoS is just not really possible (no layer-3 switch at customer site...)

 

[czechsys]

The first thing, that you need to check, is how much data is read/written in time. 10mbps is very low for anything as replication.

 

[Magneto]

I am looking for the same kind of solution, though I have a faster line to the remote site. Has anyone actually done anything like this, with CEPH?

 

[jeffwadsworth]

Just set up a test environment and check it out. The instructions:
http://docs.ceph.com/docs/master/rbd/rbd-mirroring/

 

[Kmgish]

Just set up a test environment and check it out. The instructions:
http://docs.ceph.com/docs/master/rbd/rbd-mirroring/

How has your test worked out?