IT pro and virtualization expert Scott Lowe had an interesting discussion with Jeff Woolsey, Senior Program Manager for Hyper-V at Microsoft. Scott posted a summary of their conversation on his blog, here’s an excerpt:
What are the key architectural advantages of Hyper-V as compared to Xen or ESX?
Jeff indicated that Hyper-V and Xen are architecturally very similar. Both use a privileged VM; Microsoft calls it the parent partition, Xen calls it dom0. In both cases, I/O is routed through this privileged partition and only the privileged partition has access to the physical hardware. Microsoft believes the hypervisor should be as thin as possible; Hyper-V is only about 600K worth of code. The networking stack and the storage stack are pushed up into the parent partition to keep drivers out of the hypervisor. Jeff referred me back to his session earlier in the day, where he discussed the need for the parent partition (my summary of that session is here). ESX puts all the drivers in the hypervisor, which means that they have a harder time providing support for new hardware (the example given was 4Gbps Fibre Channel HBAs vs. 8Gbps Fibre Channel HBAs). In talking about the placement of device drivers, our discussion naturally led us to the next question.
How would you respond to the concerns about the quality of the device drivers in the parent partition affecting the stability of the hypervisor?
Jeff doesn’t buy into this argument. Unlike desktops or workstations, administrators don’t typically go willy-nilly with drivers on production servers. Drivers are generally provided by the hardware vendors. In addition, because Hyper-V requires the x64 edition of Windows Server 2008, this is even less of an issue; it’s impossible to use unsigned drivers with x64 Windows. This means that any driver that can be used with Hyper-V will be WHQL-tested. Supposedly, this will keep out potentially faulty device drivers. Jeff pointed to the exclusive use of Hyper-V to power the MSDN and TechNet web sites at Microsoft as proof. I can see his point, but I still have to wonder if another level of qualification and validation shouldn’t have been established to ensure that everything works as expected with Hyper-V. It still seems possible to me that organizations stepping outside the “Big 3? server vendors—Dell, HP, and IBM—could run into issues.
Read the rest (3 more questions) here.