Replication Architecture question

[mfc]

I was hoping you guys could provide some advice on the best way to architect a Veeam B&R off-site replication setup.

So currently we're using Veeam to replicate all production VMs to an off-site datacenter for DR.
Link is 10Mbps.
Veeam proxies are VMs using SCSI hotadd. This can be changed to physical is needed.
Our priority is completing the replication in the shortest time/bandwidth used as possible. I'm willing to give up recovery time to achieve this.

Here's what we're considering switching to:

-Backup Copy jobs with built-in WAN acceleration. I'm pretty sure this would be the highest performing option. Would we notice a performance increase using this method with VM proxies, or would we need to switch proxies to physical boxes with SSD for the WAN cache in order to really see the benefit?

-Using Secondary Backup to NetApp Snapvault feature in the upcoming Veeam/Netapp integration. Datastores live on NetApp storage (iSCSI) on both sites. Our concern with this method is whether or not NetApp SnapVault replication will outperform Veeam replication? Offloading this to the storage layer is attractive, but will it help reduce our backup time over traditional replication?

[veremin]

How long does the replication job take at the moment? Were you able to complete the initial replication cycle? I'm asking because after the initial cycle replication job is always incremental, and shouldn't take long time. Also, what is the amount of data transferred? What is identified as the major bottleneck? Thanks.

[foggy]

-Backup Copy jobs with built-in WAN acceleration. I'm pretty sure this would be the highest performing option. Would we notice a performance increase using this method with VM proxies, or would we need to switch proxies to physical boxes with SSD for the WAN cache in order to really see the benefit?

Proxy servers are not involved into backup copy process. You will have to backup locally first and then use backup copy with WAN accelerators to transfer data offsite (could happen outside the main backup window).

-Using Secondary Backup to NetApp Snapvault feature in the upcoming Veeam/Netapp integration. Datastores live on NetApp storage (iSCSI) on both sites. Our concern with this method is whether or not NetApp SnapVault replication will outperform Veeam replication? Offloading this to the storage layer is attractive, but will it help reduce our backup time over traditional replication?

Just to make sure you're not missing it: with either method you will not have ready-to-be-started VMs registered in the target VI, like you have now with traditional replication.

[mfc]

We do have all the replicas seeded, we have been using replication for some time. On average, the job runs 5 hours and transfers 9GB (runs daily). However, we do plan to add an Exchange 2013 and SQL 2012 server to this job as they are migrated to a virtual platform. This will definitely increase job time as they have a high change rate.

We do realize that backup copy and snapvault would not provide ready-to-start VMs, we're OK with performing restores to get usable VMs (though we still may replicate one domain controller and vCenter to expedite the recovery process in the event of a site loss).

We currently run a nightly backup job of these same VMs to an on-site repository, so we could setup a backup copy job from that. Since the bottleneck is definitely the 10Mbps WAN link, does it make sense to utilize SSDs if we implement WAN accelerators? Would we notice any appreciable speed increase or would the bottleneck of the WAN negate the benefit of SSDs for the WAN cache?

[veremin]

In most cases, disk I/O performance on the target repository is identified as the primary bottleneck due to WAN accelerator effectively trading disk I/O for WAN bandwidth savings. The recommended steps allowing you to increase WAN Accelerator performance include placing global cache on a separate SSD disk (that provides fast random I/O) from the repository.

Thanks.