Comments on: Hyper-V Server Released

By: slowe

slowe — Wed, 13 May 2009 22:19:08 +0000

Matt, I’ve expressed my concerns about the hypervisor’s reliance upon the parent partition/management OS/dom0 and the drivers that reside there, but I keep getting told that the indirect I/O architecture used by XenServer and Hyper-V is the one true path to virtualized enlightenment. I’m just a simple guy, so I don’t know the truth, but it seems to me that your point is rather salient–how can this solution be better if a crash of the parent partition/management OS/dom0 also takes down all the VMs?

XenServer and Hyper-V gurus, feel free to correct me if I have misrepresented something.

By: Matt

Matt — Wed, 13 May 2009 15:32:49 +0000

I have been doing a lot of research lately and been working with hypervisors for some time now. Scott I wanted to pose this question, do you think Hyper-V and even XenServer would be considered some sort of Hybrid Hypervisor? I agree about the parent partition, if that dies so do your VM’s. By contrast if the SC dies in ESX my VM’s will still run. The other thing of note is how Hyper-V actually loads up. Windows Server 2008 kernel must first load, then the Hyper-V kernel somehow “inserts” itself below 2008, making that instance of 2008 the Parent Partition. It seems to me that because you are still loading up a kernel that is not dedicated to virtualization first, then inserting a kernel below that layer… well it just doesn’t seem clean to me. You have a kernel that is not dedicated to running VM’s. It requires hardware assisted virtualization. Thoughts?

By: slowe

slowe — Fri, 03 Oct 2008 02:16:24 +0000

Stu,

Strictly speaking, you are absolutely correct–thank you for pointing that out. A “Type 1″ hypervisor does not incur the overhead of a hosted hypervisor such as that used by Virtual Server or VMware Server.

I guess my comments lean more toward the absolute dependence of the hypervisor upon the parent partition. This is an architectural decision embraced by Hyper-V and Xen but eschewed by VMware. We could get into a deep religious debate over which architecture is better (placing the I/O stack in the parent partition or embedding them inside the hypervisor), but it just seems logical to me that a hypervisor which is dependent upon an OS–even if it is a stripped down OS–is somehow less “bare metal” than a hypervisor that isn’t. Does that make sense? Like I said, I’m not trying to stir up the religious debate here, rather just calling it like I see it.

Thanks for your comments, Stu.

By: Stu Fox

Stu Fox — Fri, 03 Oct 2008 00:44:00 +0000

Just to be clear, that comparison is of Hyper-V, not Hyper-V Server. Hyper-V is the version that is Windows Server 2008, and Hyper-V Server is the standalone product.

And to answer your point Scott, it is indeed a bare metal (or type 1) hypervisor that runs on the bare metal. It does not incur any of the overheard of a “hosted” or type 2 hypervisor because of this. The fact that the parent partition (itself a VM) runs Windows Server 2008 does not change the bare-metalness of the hypervisor.

Disclaimer: I work for Microsoft NZ, but this is not an official response on behalf of my employer.

By: Jon B

Jon B — Thu, 02 Oct 2008 16:56:10 +0000

Scott and others, check out this comparison of the install of Bare Metal Hyper-V and ESXi side by side.

http://www.vmware.com/technology/whyvmware/resources/esxi-hyper-v-installation.html

…and no I don’t work for VmWare. =)