Virtualization Automation – A Proper Virtual World

Hyper-V, No Warp Drive, Struggling With Impulse Power Still

November 4th, 2008

Virtualization Critical Comparison – Chapter 09

Well, I did it, I did the deep dive into Hyper-V, it was demanded, it was not an option, and when I came up for air, I felt like I was covered in dirt, well to be more specific, not dirt. The stuff was brown, but it had an odd smell, and it did not come off easy, too greasy for my liking, more like something unexpected that oozed out of the cracks of a weak impulse propulsion engine design prototype. Which I am not sure was what Microsoft was hoping for? Or was it? Microsoft released something that was better left stuck to the bottom of my pressure suit magnetic boots, rather than introduced into my virtualization lab. But in reflection this is what Microsoft intended, yes, intended. How so? Microsoft wanted to avoid looking like they were standing still, that they had not completely lost the Hypervisor market for at least another 18 or so months?

Look at Microsoft goals and objectives, from the past, they have done this before. They almost never have a winner out the door, but something that is functional, or looks to be anyways. Something that can be built up and refined, and polished into a mirror finish of Brass. Not Gold or Silver, at least not yet. Hyper-V gets you from Earth to Saturn and back again. But if you need to leave the Alpha quadrant, such as VMware has already done? For get it. Of course, along the way, Microsoft gets just enough of a following to make the solution look passable, if not viable at some scale or degree. Look at WordPerfect and Lotus, they dominated their respective markets, and Microsoft took them down hard, over time.

The gloves are off, that is obvious, and Microsoft has VMware in the phasers targeting array, and is locked on. The interesting thing is that VMware is a moving target, at warp speed, so Microsoft is having trouble getting a kill. Microsoft is still not strong on the feature set, no matter how many their sales teams decry the faults of VMware or Microsoft otherwise discounts the gaps in Hyper-V to cool-aid sipping CIOs of firms world wide, VMware is still the best of breed, still avoiding assimilation or destruction. And I am not just referring to VMware VMotion, or what Microsoft has tried over and over to rename as transparent migration. Lets be real, quick-migration is not quick, if any reboot is required, and in Hyper-V, that is what is required, be it planned or as a recovery on a cluster, it is not HA (High Availability) but AHA (Almost High Availability).

Let us dig a bit deeper. If some of that brown stuff gets into your suit vends, sorry about that, but when you decided to read about Hyper-V, you should have expected the brown stuff would get all over the place. The incomplete features that hurt Hyper-V are as follows:

• HA (High Availability), well sort of, Microsoft Cluster running VMs (Virtual Machines) as a resource is not HA, it is really a band-aid, no matter how you look at it.
• QM (Quick Migration), a reboot is a reboot is a reboot. This was discussed above, so no need to itemize again here.
• One VM (Virtual Machine) per LUN? Are You Kidding? What bozo thought this up? Did I say band-aid again? Microsoft does not have a shared IO model that is really a shared IO model. VMware VMFS is not perfect, but it did set the expectation of standard high. Even in the Microsoft Clustering model, Microsoft says that disks should be duplicated not shared, the old X-Cluster versus Y-Cluster argument. Well, even if shared in a Y-Cluster, there are scaling problems, for anyone with 1000s of VMs. Like 1000s of LUNs?
• Networking? It is horrible in Hyper-V, and Microsoft says, well, we depend on vendors to develop new or better drivers. Bull. Microsoft has all the leverage, the design is weak. For example, HP is still working on getting teaming to work right with Hyper-V, to which HP PSP 8.2 will support teaming in Hyper-V. Dell and IBM are silent so far on the issue, stating that they rely on the NIC drivers as provided. But simple teaming to avoid Single Point of Failure (SPOF) is not the issue I am yelling about here, it is the fact that Hyper-V does not do load-balancing or even an active/active pool of NICs, similar to VMware Bond, for me, that just yells weak design out of the space dock.
• You have to have VMs with IDE for boot? But can have SCSI for other VM disks? Dumb. I think Microsoft did this to make transportation between VMware ESX and Hyper-V and back again, which will be a key trick for enterprise scale organization, as painful as possible.
• Microsoft Clustering is horrible. Yes, horrible. When it works it is great, when it is ill or sick, you often will find it easier to take a node offline and shoot it with a phaser, because recovering a node that is whacked is one step short of a digital miracle. How do I know this? I have been supporting and designing solutions around Microsoft Clustering since 1998!
• The entire Hyper-V solution is dependent on MOM, SCVMM and other Microsoft tools, so Hyper-V is not free, in fact, if you need to scale Hyper-V, it is not inexpensive. Anyone look at the cost of SCVMM and MOM? They are expensive and going up in price, with each new version, are they not?

The single most disappointing issue, well more of an architectural concept to be fair, with Hyper-V, for me, is the SPOF (Single Point of Failure) issues. With VMware ESX, to eliminate SPOF, I just double my components, and of course when needed, add an additional VMware ESX servers, maybe one or two or three, etc., and I am done, yes, done. No SPOF, unless I don’t know why physical switches and/or storage processors are. I can have as many physical NICs as the given hardware can handle, and I can map the physical NICs as I see fit. The same for HBA channels to the fabric, as long as I have shared-storage, I am good to go; a Virtual Cluster is just a few clicks away… Make it so, Number One. Microsoft, I need so many other components, applications, and layers of integration, it takes a Xenomorphic degree to keep things straight.

VMware VirtualCenter (VC) server? Sure if you need it, but not required, to achieve about 80% of the benefits of the virtualization architecture we need to avoid SPOF. VC components and features are integrated into one interface, where as with Hyper-V, you really need to use three (3) different tools just to match VC for configuration tasks, lets see, 1) the Cluster Administrator, 2) SCVMM, 3) Hyper-V MMC plug-in, because? Why Cluster Administrator and SCVMM don’t talk or play nice, you end up in Hyper-V MMC, to figure out what the heck has gone sideways. I will not itemize the other features in VC, which would only kick more of the brown stuff into the Microsoft fans suit cooling system, while they are reading this. Now I ask you? Does Hyper-V look that appealing, or is this a case of the old adage…You get what you pay for?

True, I am not the biggest fan of VMware, and at times I hammer VMware in this blog, to get things right, to improve, to be a better steward of the virtualization cutting edge, to boldly go where no, man, cough, virtualization company has gone before, sure. But compared to Hyper-V? This time I am going to praise VMware, a bit. VMware has a better solution; it may not be perfect, from my perspective, but compared to Hyper-V? VMware has achieved warp drive, where Hyper-V is still stuck on impulse drive. Maybe Microsoft should hire a few Vulcan technical experts? Hey, I wonder, VMware, Vulcan? Is it just me? Or is there a connection here? How many years did it take Star Fleet to get out from under the guidance of Vulcan? Vulcan developed 2nd and 3rd generation warp systems under the support of the Vulcan Science Directorate, I mean VMware developed VI 3 and 4, while Microsoft is still trying to get out of the space dock without training wheels. It was 3, maybe 4 years at least, right? Well Microsoft has already had 3 years already, and the training wheels are still welded to the side of the hull of Hyper-V, at best it is embarrassing for Microsoft, at worst, again for Microsoft, Vulcan, cough, I mean VMware has escaped the Borg, yet again!

Entry Filed under: A Proper Virtual World