Another major enhancement that is coming to Windows Server 2008 is the Microsoft hypervisor. Microsoft has stated the hypervisor will be released within 180 days of RTM, so we won’t see it until the middle of 2008.
As you are probably aware, there are a number of virtualization products on the market. Vmware probably comes to mind first, as well as Zen. Microsoft has taken a somewhat different approach which I will cover shortly.
The current MS virtualization offering is Virtual Server 2005 R2 SP1. While many find it a useful (and free) product, it doesn’t have the enterprise features of a product like VMware ESX. Microsoft aims to change that. How successful they will be, only time will tell. Vmware has a huge head start and market share which will make it even harder for Microsoft.
The new hypervisor will support:
*32-bit and 64-bit VMs on the same physical server
*Large memory support within the VM (>32GB), with memory over-commit and page sharing.
*Multi-core VMs (2/4/8)
*Pass-through disk access for improved I/O throughput
*Hardware sharing model for disk, networking, input, video
*Supports clustering
*Robust networking support (VLAN, NAT, quarantine)
*MMC 3.0 interface, no more IIS (yess!!!!)
*Requires AMD-V and Intel-VT extensions plus DEP enabled processors
*Remote display uses a version of RDP that does not require RDP to be enabled in the guest OS.
*No USB support (Booo)
*Automation and management via WMI
*Snapshot and backup capability of live VMs
Recently Microsoft announced they are deferring some features for a future release. Given the probable mid-2008 initial release, we might be looking at 2009 before they implement:
*Hot add: memory, processors, storage, networking
*Live migration between physical hosts (yes VMware can do this today)
Now you might be thinking that Microsoft is not in the driver business, and now they will need to write specialized drivers for the hypervisor, like ESX requires. Interestingly, Microsoft is using a different architecture.Within the hypervisor MS is implementing a virtual shared bus called "VMBus", which is synthetic hardware. Hardware is *not* emulated. Let that sink in for a minute. No more generic "NEC" NICs in your VM, or a generic SCSI driver.
Instead, the VMbus exposes the physical hardware to the guest OS, where you use the normal hardware drivers. The VMbus mediates access to the physical hardware. So Microsoft will not need to write drivers, but will rely on regular physical hardware drivers installed in the guest. For example, if you have a 10Mbps 3COM NIC in your server, the guest will see a 10Mbps 3COM device. Now I’m still wondering how they handle sharing among the VMs since a driver would normally assume it has exclusive control of the device.
Microsoft insists this model is being highly tuned to provide performance similar to the ESX model where the drivers are in the hypervisor. Until we get closer to a ship date and can run benchmarks, the jury will be out regarding performance and how well the VMbus concept works. Given the high availability feature such as live migration and hot adding of resources is deferred into the future, I think Microsoft may have a hard time gaining market share against VMware in the enterprise space. It’s clearly a major improvement over Virtual Server 2005 R2 SP1, though.
