In-Depth

Q&A on Microsoft Virtualization, Part 5

Ben Armstrong discusses virtualization options for storage and Azure.

• By Paul Schnackenburg
• 01/04/2016

More on this:

• Part 1
• Part 2
• Part 3
• Part 4
• Part 6

Contributing Editor Paul Schnackenburg continues his interview with Ben Armstrong, Microsoft's program lead for Hyper-V.

Paul Schnackenburg: Are you going to bring down the number of hosts required for Storage Spaces Direct?

Ben Armstrong: That is not my team! I hope we do.

PS: Four [hosts] seems an odd number. It makes perfect sense if you're a medium to large business. You can build your storage clusters and you can add nodes; but it's such a perfect technology for a small to medium business that might want two or three hosts.

BA: So there are a couple of answers to this. The first one is two hosts is just particularly problematic from an engineering point of view. We're well aware of the demand for it. We have a huge amount of deployments in the two host range. Just from an engineering point of view, figuring out how to do a good availability story with a two host solution, it's a thorny topic to bite off.

The biggest debate has been internally about like should it be four hosts or should it be three hosts. At this point in time we are making people have a bit more pain up front, but knowing that they'll have a great experience when they do have hardware failure… that's the problem with the three host deployment; you get a node fail and now you're down to two, that's a real... algorithmic challenge at that stage, whereas with four hosts we can say, okay, if you have a host fail we're down to three and that's still good.

Now the counter, I will say… is we are supporting hyper converged deployment of Storage Spaces Direct in Hyper-V. So you can have four nodes where it's running Storage Spaces Direct and it's running Hyper-V on top of it, and we actually think that's going to be a really interesting deployment option for a lot of people.

We are aware that there are situations where people want the two node thing; and I'll be frank, we don't have a great story for that today. It's something that I wish we did have and it's something that we talk about a lot in our team.

PS: So can you do the hyper converged set up in TP3 [Technology Preview 3]?

BA: In TP3, yes you can.

PS: Okay. Is RDMA [Remote Direct Memory Access] networking a requirement?

BA: Once again this is not my team, so I'd have to go and check the paperwork, and I will say that we have this around Storage Spaces Direct. We do have a bunch of requirements that aren't enforced. They're more best practices where we say, look, we know that you'll get the best experience in this configuration and that's really what we recommend.

That said, once thing I always have to point out because we don't have enough people using RDMA: RDMA adaptors have gotten really cheap. If you're building out a new system, getting RDMA-enabled adaptors does not add a substantial amount to the bottom line.

PS: So obviously, the next step will be that Azure will move to Nano Server as the host for Hyper-V for IaaS and everything else up there. What you're telling me really is that it's more a commercial decision from Azure's point of view: "Sure, we've got these things but we're not necessarily going to light them up unless it makes business sense."

BA: Yes, absolutely.

PS: So, are we likely to see Generation 2 VMs, are we likely to see live migrations in Azure?

BA: I cannot honestly give the answer. This is a discussion that goes on frequently, and now the discussion is primarily focused on: Now, what [makes] the most business sense, what's going to get the value back? 

Now one of the things I am really excited about is that we're working on bringing nested virtualization to Azure, because that's going to make it really easy for people to play with different features, even if we haven't lit them up; they can still with them using Azure.