Query on LAN based Hyper-V Replica Server 2012 R2

TrentSlater
TrentSlater used Ask the Experts™
on
We have been asked by a customer to setup Hyper-V Replica internally on 2 servers as part of their improved Business Continuity plan.
I understand that the HV Replica scenario is more suited to DR - where the Production site may be unavailable, rather than BC and to do HV clustering we would require a centralised data storage -i.e. SAN. To keep their costs down their Internal IT have asked if we can migrate all their VM's to 1 physical server (Currently  split over 2 servers) and then setup a separate 10Gbit link directly between the 2 servers and use the second server as the HV replica - setting a low replication time to reduce data loss. Does anyone have any thoughts on this?
Comment
Watch Question

Do more with

Expert Office
EXPERT OFFICE® is a registered trademark of EXPERTS EXCHANGE®
Andrew Hancock (VMware vExpert / EE Fellow)VMware and Virtualization Consultant
Fellow 2018
Expert of the Year 2017

Commented:
I understand that the HV Replica scenario is more suited to DR

we would agree, in the event a VM fails on Production Server, you can quickly turn on the Replica.

rather than BC and to do HV clustering we would require a centralised data storage -i.e. SAN.

Failover clustering does require the prerequisite of centralised data storage, this could be SMB3 file share, not necessarily a SAN.

To keep their costs down their Internal IT have asked if we can migrate all their VM's to 1 physical server (Currently  split over 2 servers) and then setup a separate 10Gbit link directly between the 2 servers and use the second server as the HV replica - setting a low replication time to reduce data loss.

Does anyone have any thoughts on this?

Well that works, it depends on, on how much value, they give to lost business, with time taken to spin up the replica, and also data loss, between the last replication time.

and finally - Customer/Client is always right!

We do what we are told
... Milgram's 37
Top Expert 2016

Commented:
workable but there is approx 15 minute lag
Distinguished Expert 2018
Commented:
A lot to consider:  Workloads. I wouldn't be failing over anything that requires transactional integrity. So most databases are out. Exchange is out (and the Exchange team officially has stated they won't support Hyper-V Replica) and most accounting packages are out. Hyper-V Replica fairs better with file workloads and web workloads.

Overall, I'd consider rethinking things. First, there is licensing. If the VMs will be running on the "cold" server at all, you have to license for them, even if they aren't actively running most of the time. If you license Standard, there is overprovisioning regardless of whether you cluster or replica, but if you go datacenter, since that is unlimited VMs, you get much better utilization out of a cluster and "free" VM roaming as needed.

 Then there is the issue of under-utilizing hardware. You can split your VMs up and configure replica going both ways....a nightmare to maintain and properly fail-over *or* fail-back since you actually have multiple replica relationships that can go wrong. And the ensuing risk of data loss. Your cost is soaked in support.

Or you can pay to have a server sitting there doing nothing but receive replicas. Much easier to manage, but you have licenses on that server doing nothing and you have hardware you paid for doing nothing. A lot of "empty" up front cost just for a potential failover.

Or you can invest in shared storage. The cost of shared storage is not usually a barrier to entry these days, plus you get much better hardware utilization. SAS enclosures are relatively inexpensive and scale well up to four servers in a cluster with multi-path which virtually eliminates any and every single point of failure. And if you have more than four servers, replica pretty much goes out the window anyways, and storage is even less of an issue (you can stretch multiple SAS enclosures or go bigger to iSCSI, FC, etc.

And you have to reconfigure clients in an HV-Replica scenario. You don't in a cluster scenario.

Regardless of how you slice it, for true HA, you want a cluster One way or the other, the hidden costs will soak you if you try to short-cut it.
OWASP: Threats Fundamentals

Learn the top ten threats that are present in modern web-application development and how to protect your business from them.

Author

Commented:
15 minute lag is not a problem - within 2 hours is reasonable with minimal data loss in the event of major hardware failure
Andrew Hancock (VMware vExpert / EE Fellow)VMware and Virtualization Consultant
Fellow 2018
Expert of the Year 2017

Commented:
So there you go, but have a read of Cliffs post!

You may want to re-think!

Author

Commented:
Thanks Cliff

There are 3 virtual servers with 1 of these being SQL. Currently emails are with O365 so no onsite exchange server. IF we went for shared storage - i.e. SAS enclosure would this technically be a dual port DAS connected to both servers or would it work OK SMB3 via 1GB ethernet?
Distinguished Expert 2018
Commented:
I don't like mixing terminology. Shared storage is not DAS, and shouldn't be considered as such. While the enclosure would be directly connected to multiple servers, is *is* shared in the fact that two servers could request simultaneous writes to the same block on the same disk and...if not configured correctly, corrupt each other's data. The disks *are* truly shared.

While Microsoft *did* introduce SMB3 as a legitimate storage option for Hyper-V in 2012, most people misunderstand *why* Microsoft did this. There are super-cheap NAS boxes (not SAN, but NAS) that support SMB3, but they too have a motherboard, a power supply, and an OS (usually Linux) and therefore *can* fail. You've basically moved the single-point of failure from the server to the storage box. But it is *still* a single point of failure.

Microsoft's intent with supporting SMB3 wasn't so you could get away with a single box hosting storage. It was to make large clusters much more economical. If you had, say, 100 nodes in a cluster, you used to have to invest in a very expensive SAN (fibre-channel, *maybe* iSCSI) which, itself internally, had redundant everything...including mainboard and OS so it *wasn't* a single point of failure....and ran in the price-range anywhere from a family house to a small mansion. Plus fibre switches and such just to get all the nodes connected to the same SAN. Deployments were *expensive* ...even by the standards of companies who make enough ot justify needing a 100-node cluster.

That all changed with SMB3. Now you can take SAS enclosures, connect them to two to eight servers, and cluster those servers running the file-server role (not the Hyper-V role), and if desired, buy more enclosures and more servers and scale out from there. Thus the term "scale out file server."  Then your Hyper-V servers could connect to your file servers via SMB3.  Once again, you've eliminated a single point of failure because if a file server dies, it is clustered and another can take its place. The enclosures can be redundant (multiple power supplies and multiple SAS connections) but are otherwise "dumb" in that they don't have an OS or processing *at all* so nothing to fail there. And you can configure windows to even duplicate data across enclosures if you were really paranoid.

But the point is, even with SMB3, Microsoft's architecture still has shared storage (via SAS) in the mix. They are just doing it at the file-server stage instead of connecting Hyper-V nodes directly to the SAS enclosures. BOTH topologies are supported. But neither is using a single file server (or NAS, or other) as a single point of failure. Implementing such a thing defeats the purpose of a cluster in the first place.

So, long story short, in a small deployment (two to four nodes) I wouldn't use SMB3, and don't see a justification to do so. To be HA, you'd effectively need to double the number of servers in the mix.

Do more with

Expert Office
Submit tech questions to Ask the Experts™ at any time to receive solutions, advice, and new ideas from leading industry professionals.

Start 7-Day Free Trial