Veeam Replication vs Hyper-V - Speed/best practice?

[TimWalker]

Hi All
Fairly new to Veeam, had our server up and running for about a month and a half, mostly very happy.

We've been trying out the Veeam replication instead of the built in Hyper-V replication, and although it is working it seems to run a lot slower than the Hyper-V replication did.

To give you the full story, I work in a school, we have a large split site, with 3 hypervisors in one block, and 2 hypervisors in another - we've found that localised power issues can take down either building, but rarely both at once - we have the capacity, just about, that we can run our network from either block (although we usually run them spread across the five hypervisors for best performance).

There's 10gb fibre between all the servers so there is no issue of network speed/latency.

So, with Hyper-V replication, although I understand that our replicas weren't as clean as they might be with Veeam (eg replica of our exchange server is not application consistent) if the power did go down in either wing, I could shut down the virtual servers on that end, run a planned failover and have them back up on the other wing inside ten minutes, within our UPS run time (and because the servers were shutdown at the time of the final replication operation questions of application consistency were moot).

I'm finding with Veeam that the replications take much longer than they did with Hyper-V - so the jobs are running pretty near constantly to maintain a sync every 20 minutes - and having tried a planned failover with our exchange server the operation took over 40 minutes, beyond what we could reasonably expect our UPS units to sustain.

Is it just the trade off I have to make, between having fast (planned) failover and being better protected against catastrophic hypervisor failure? Is there some best practice guidance for replication that might get my jobs running quicker, or is this just the sort of speed I should expect?

[PTide]

Hi,

Would you describe the replica job settings please (proxy type, proxy location, metadata repository placement etc)? Also please provide your bottleneck stats.

Thanks!

[TimWalker]

Metadata repository is on our primary Veeam backup server, spinning disk array.

For proxy settings I've tried both the default off host (automatic proxy selection) and on host - it didn't seem to make much difference to the performance when changed to on host.

Here's a couple of example log outputs from replications this afternoon.

25/09/2018 15:45:06 :: Job started at 25/09/2018 15:45:03
25/09/2018 15:45:06 :: Building VMs list
25/09/2018 15:45:21 :: VM size: 1.1 TB (1.1 TB used)
25/09/2018 15:45:21 :: Changed block tracking is enabled
25/09/2018 15:45:21 :: Processing for4032-uss-05
25/09/2018 15:45:26 :: Processing for4032-sr-01 2012
25/09/2018 15:48:59 :: All VMs have been queued for processing
25/09/2018 15:53:56 :: 1 restore point removed by retention policy from VM for4032-sr-01 REPLICA
25/09/2018 15:53:56 :: 1 restore point removed by retention policy from VM for4032-uss-05_replica
25/09/2018 15:54:22 :: Load: Source 93% > Proxy 6% > Network 9% > Target 2%
25/09/2018 15:54:22 :: Primary bottleneck: Source
25/09/2018 15:54:22 :: Job finished at 25/09/2018 15:54:22


25/09/2018 15:15:13 :: Job started at 25/09/2018 15:15:10
25/09/2018 15:15:13 :: Building VMs list
25/09/2018 15:15:27 :: VM size: 1.2 TB
25/09/2018 15:15:27 :: Changed block tracking is enabled
25/09/2018 15:15:27 :: Processing FOR4032-SR-02 2012
25/09/2018 15:15:32 :: All VMs have been queued for processing
25/09/2018 15:54:48 :: 1 restore point removed by retention policy from VM FOR4032-SR-02 2012_replica
25/09/2018 15:55:18 :: Load: Source 91% > Proxy 7% > Network 14% > Target 6%
25/09/2018 15:55:18 :: Primary bottleneck: Source
25/09/2018 15:55:18 :: Job finished at 25/09/2018 15:55:18

[TimWalker]

So, I tried a planned failover on a machine that was already shutdown overnight - it still ran two replication cycles despite the machine already being shut down.

I mean, it's surely arguably that the two replications are necessary in a planned failover anyway (I suppose you could argue it minimises downtime of the server if there is a lot of data waiting to be transferred when the planned failover starts) but this is a clear waste of time - could it be an option to just run one replication cycle on planned failovers? Perhaps a tickbox?

[PTide]

The stats show that source is the bottleneck, which means that it is probably the storage what causes the slowdown. You've mentioned that an ordinary Hyper-V replication runs much faster, so I suggest you to contact our support team so they can check what is the exact reason. Please post your case ID afterwards.

Thank you!