Microsoft's vision of virtualization -- Part 1

I had the great fortune to attend a presentation from Microsoft earlier this week where we spent a couple hours with one of the senior engineers in their server group discussion the future of the server OS, focusing specifically on virtualization. For the most part it was a discussion on serve virtualization, but there was some discussion of the desktop as well. We covered a ton of information and I can't get it into a single coherent post so I'm going to split it up over the next few days. Check back for more on virtualization management and some of the desktop bits later in the week.

Moving from "installation" to "configuration"
Microsoft is moving to adapt the VHD format, making it much more flexible for application deployment in addition to server OS virtualization. The goal seems to be a result that is a bit like a WIM in that it can be edited and updated, but also to have applications that can be delivered as VHDs to end users. These wouldn't be demo VHDs that have an OS install and then the application on top of that. Microsoft would expect that the end-user would provide a VHD or physical server install for the base OS and then be able to layer the application VHD on top of that. It sounds a bit like the SoftGrid platform but for servers instead of workstations. We'll see what comes of it, but it is definitely interesting. Not having to wait while a bunch of DLLs and other binaries copy from the install media would definitely be a nice change, though waiting for the one big VHD file probably isn't all that much faster in the long run. Still, the idea is that we would be focusing on configuring the applications rather than installing them.

Hyper-V and the other Hypervisors
Hyper-V is really close to being released, and it seems to have some very interesting benefits as a hypervisor option. I still maintain that it isn't a competitor in the Enterprise space (and the MS guy agreed, several times) yet, but it is clear that Microsoft is moving that direction with great haste, and good things will be coming in the next few product cycles. So, what are the good and bad things?

Hardware Drivers (good)- Microsoft's approach to the hardware situation is a bit different than VMware's (and similar to Xen's). Rather than writing a driver to emulate a specific hardware model (think the LSILogic SCSI controller in VMware), Microsoft has "synthetic drivers" for their hardware. The synthetic drivers can operate with driver code written by the hardware vendors to run in the guest OS sessions. This approach allows for a lot of flexibility for hardware support, both from a breadth of supported hardware as well as the features that can be supported. For example, if you're looking to host a VDI environment and buy a super-powerful video card to handle all sorts of fun UI features, you cannot take advantage of those features unless the guest OS can see the underlying hardware and leverage the power that it has. Synthetic drivers expose a framework that lets that happen. It requires the hardware manufacturers to write the drivers, but that should happen soon enough. The synthetic drivers also requires an OS that is aware of the virtualization platform, also known as an "elinghtented" OS. Fortunately, all the current Windows versions (XP SP3, 2003 SP2, Vista SP1, 2008) are enlightened, so that isn't a barrier.

Data Protection (good) - The VSS integration in Hyper-V takes the idea of snapshots to the next level. Currently, any VSS-integrated backup tool can tell a server that it is going to take a backup snapshot and the files would be quiesced for the snapshot. In a virtual environment, however, that doesn't do a lot of good. The VHD file would be quiesced, but without visibility into the guest OS there is no way to ensure that the applications inside the VHD will be aware that a backup is happening. This leaves you in a crash-consistent state - as if you pulled the plug on the server. Hyper-V exposes a new VSS service in the guest OS sessions. This service allows the host session to tell the guest that a VSS-aware backup is running. Any VSS-aware apps in the guest then quiesce, improving the reliability of the system when you recover it from the snapshot of the VHD. This is very useful for Active Directory, Exchange, SQL and SharePoint servers, as they are all VSS aware.

Host Migration (not as bad as I thought) - One of the major issues I have with Hyper-V as an Enterprise solution is that it doesn't allow for a Vmotion-type migration of a running system in a live state. They do have a "Quick Migration" however, that is a big step ahead of having to shut the server down, migrate it to a new host and then power it back up. The Quick Migration flushes the RAM state to disk, stops processing and then loads the RAM state on the new host and resumes processing. There is still a window where the guest OS is unavailable, but it is still better than having to shut the guest all the way down and power it back up. This would require a 1:1 ratio of VM to LUN, which is going to drive the storage guys crazy; see below for why.

Disk Management (good and bad) - The good news is that Hyper-V allows for pass-through disk access (same as RDMs in VMware), in addition to direct-attached storage, FC, iSCSI and ATA support. The pass-through doesn't have a huge performance benefit - only ~5% - but if you're doing pass-through it is almost always for management, not performance. The bad news is that the NTFS file system is really unhappy when more than one host tries to write to a volume at the same time. This means that if a VHD guest is active on a host and then you want to move it to another host you have to move all the VHDs on the same LUN. Essentially this means that each VHD has to be on its own LUN to maintain the flexibility that you need. Compared to the VMFS-3 file system from VMware this is a huge limitation. The good news is that the Windows Server guys are working on this, but it isn't going to happen until at least R2 of 2008, and probably not until the next version of Server.

That's pretty much what we talked about on the server virtualization front - or at least the parts that I found interesting and took notes on. Look for workstation virtualization notes tomorrow and management stuff heading into the weekend on Friday.