Replication at the Hypervisor layer or storage layer?

[TheVMinator]

I'm hearing a lot now about replication at the hypervisor layer.  What I'm wondering is why it would be desirable.  For example, if I have an HA cluster that is highly utilized and am using DRS to manage my compute capacity on each physical host, why would I want to make my physical hosts use their processor capacity for replication?  Why wouldn't I want the storage array to do it?  The storage array is already equipped with things like throttling and is designed to do replication well.

For special use cases like replicating to a site with different storage system or different administration perhaps it would be necessary from an administrative standpoint.  From a technical standpoint I am thinking about managing compute capacity in my DRS cluster - when if ever is virtualization layer replication preferable to doing it at the array level?

[jfrappier]

When your array doesn't support replication?

[TheVMinator]

Yes obviously if your array doesn't support replication then you have no other choice but to have your physical hosts use their resources on replication instead of running virtual machines.   Is it safe to say that replication at the virtualization layer is intended only for environments that can't do array-level replication for some reason?

[jfrappier]

I think I said this in one of your previous post - balance. Where do you have resources to manage the replication? As I think about it you will always need some SAN resources, even if its the hypervisor doing lots of reads.

To me vSphere replication is for SMBs who may not have the ability on your SAN.

[TheVMinator]

Thanks for the input.  As far as this:

"Where do you have resources to manage the replication? As I think about it you will always need some SAN resources, even if its the hypervisor doing lots of reads."

One feature I have seen on storage arrays is the ability to deprioritize replication traffic over other more critical workload reads and writes.  For example, replication starts and then a spike occurs of critical workload reads and writes - so the array throttles back replication and prioritizes providing resources for the critical application.

As you mentioned, it is true that you will always need some SAN or array-level resources even if the hypervisor does the replication.   If the hypervisor does do the replication, those IOPS are going to look like regular workloads to the array.  So it won't be able to prioritize its resources to the actual applications over replication traffic.

So I'm thinking that in these ways and other ways the array is more advanced or more mature in its ability to handle replication and that the hypervisor's capabilities are only a stop-gap solution applicable were array replication can't be done.  It's just that all the hype on hypervisor replication I have been hearing makes me wonder if I am missing something. 

[EGarbuzov]

Another example is replication vms in virtual DC in provider's Cloud (aka DRaaS). Array base replication can't be used in this case, and I think it's not only for SMB.

[TheVMinator]

Replicating to a public cloud provider is one case where definitely there would be an advantage to having the option to replicate as the hypervisor layer. 

(As I mentioned in my initial posting:

     "For special use cases like replicating to a site with different storage system or different administration perhaps it would be necessary from an administrative standpoint")

In that case, it seems also that the driver for using virtualzation layer replication is not that there is a performance advantage of any kind.    Since you can't replicate to a public cloud's storage system directly, you would be forced to replicate at the hypervisor layer.