Intercontinental HA

[bhatch]

Folks:

I have sites in LA and NY. LA is primary - NY stand-by. Hyper-V 2012R2 in both locations.

I have a Veeam B&R VM in LA - but a recent failover and failback test took too long - we had to scrub the test as it was taking 8+ hours for 8 larger servers.

So architecture? I am thinking about moving the B&R server to NY and running a proxy in LA...will this alleviate the absurdly long failovers? Hyper-V replication is also in place and failover/failback takes moments and not hours.

A little help?

[veremin]

May I ask you what particular step took the longest time? Failover, failback, failback latest changes transfer, etc.? Was it an ordinary or planned failover? More details would be helpful. Thanks.

[foggy]

Also, do you currently have a proxy at the NY site?

[bhatch]

Folks:

Duration of planned failback was hours, planned failover took about half of that. There is no proxy at the NY site but I am thinking that the NY site is the best place for the R&D server with a proxy at LA.

[foggy]

You need proxies at both ends for optimal data transfer. Failback is expected to take long since it needs to calculate differences between the original VM and the replica VM. Typically it requires similar time as active full for the VM takes.

[bhatch]

So proxies at both data centers and Veeam B&R server in NY will work?

[foggy]

You can use Veeam B&R as a proxy as well (it is actually a proxy by default).

[bhatch]

OK, so I will Hyper-V replicate the Veeam B&R server to NY and install a proxy on a server in LA. During the next test I will know if speeds improve. Thanks for the help!

[foggy]

Sorry for confusion, I missed the fact we're in Hyper-V branch. In this case data processing is actually performed by the host itself (on-host backup/replication), so you already have proxies at both ends.

How about actual replication jobs? Do they also take long? What are the bottleneck stats for them?

[bhatch]

OK:

Stats for one server - a SQL server: 42:33 duration - Processing Rate is 16MB/s - Bottleneck is source.

[foggy]

Isn't too fast and your source storage seem to be the weakest point in the chain. Since during planned failover the job performs replication of the final changes after powering off the original VMs, it also takes quite long.

[bhatch]

That makes sense.

We have new storage planned for Los Angeles. The rest of the design should be valid with B&R server moving to NY?

[foggy]

Having Veeam B&R in the target site is recommended to be able to failover via Veeam B&R in case of DR.