Replication between clusters/datacenters

[txsastre]

hi there.

I have 2 CPD in different buildings, connected with 10Gb Fibber, 1 is the production the 2 is the backup disaster recovery.

how could I replicate every x time from production CPD VM to the Disaster Recovery CPD ?

Both configuration is similar, 3 servers and 1 SAN storage.

thank you !

 

[LnxBil]

SAN means block storage exposed (possibly multipathed) as LVM to your PVE? If so, you cannot replicate that efficiently, sorry.

I only see these options:
- ceph (built in into PVE)
- do the replication with your SAN (vendors do have solutions for this)
- use a ZFS-based SAN that can replicate for you (possible HA-ZFS solution, not part of PVE)

Disaster recovery is not always a good solution, because it blocks resources and/or costs a high. Often, people do stretched clusters, e.g. with Oracle Database (as Real Application Cluster). Both datacenters run the storage and the application so that one datacenter can fail and the other will still continue to work. That is possible because you can have a software RAID over two SANs, so that every write is send to both SANs, but you always read from your preferred mirror. Something similar is possible with PVE if you use CEPH, but you will not use a SAN at all. For those hyperconverged systems, you do not need a SAN, it is even bad if you have one, you cannot use it. In both setups, you still need a third quorum disk/server that will allow one datacenter to live while it cannot connect to the other datacenter.

 

[txsastre]

hi there, thanks for your fast and clear answer.

Well my SANs are hyper-converged ones, a Huawei Oceanstor, it also can work as file server (NFS, Samba) and block level and the other one is a EMC2 VNXe, also a hyper-converged.

I could connect both as a NAS devices, so each one could see the other one, also remote servers could be part of the cluster (10Gb fiber network)
Don't know if its too "extreme"


What kind of hardware do you use for storage ?

 

[LnxBil]

What kind of hardware do you use for storage ?

The dumbest block storage possible, which normally gives best performance for LVM. We do not use thin provisioning or snapshots on the storage side (both not supported on LVM). For those kind of setups we normally run with entry level Fujitsu Eternus on fiberchannel.