In the posting entitled Which Client Virtualization To Choose; VDI or PC Blades? which dates back a week or two, I ploughed in to the topic of the cost of a PC Blades based approach to desktop virtualization as opposed to the cost of using Virtual Desktop Infrastructure (VDI), which is roughly the same, except that you put a hypervisor on the blades and create virtual PCs – so you need fewer blades.

This is not a trivial issue, because a fairly substantial number of companies are trying one or the other or even both of these solutions together. The simple question is “which is most cost effective?” and, as explained in the previous posting, although intuitively VDI feels as though it should be cheaper, often the VDI solution is actually more expensive – which is counter intuitive until you analyze the costs.

The extra per seat costs of VDI come from addition SAN costs which can be around $300 more, and software license costs for the VMware VDI suite (around $200), MS-Vista Enterprise Centralized Desktops (about $300) and MS-Software Assurance, which is needed for VM solutions and comes in at around $150. Considering we’re talking of per seat costs, the additional per seat expense can be as high $950 per virtual PC and I’ve assumed enterprise discounts in this. The only variable part of the $950 is the SAN costs which could be lower.

But $950 more than covers the additional hardware costs in a PC Blade which are about an extra $500 per seat. So it can be cheaper to just give everyone a PC Blade and not bother to virtualize. However, I have little doubt that virtualization is becoming the way of the world and the cost equations will inevitably change. So I expect that the more prudent approach will be to embrace both solutions.

A Solutions Architect with a major hardware vendor pinged me with an email to provide further information and the benefit of his experience. I don’t think there’s any bias in this, but you can judge for yourself. His comments are italicized:

With PC Blades you can use the old statistical aggregation model that is also used with Citrix implementations. You would typically only need to look at between 65 and 75% to provide 100% service – so that for 1,000 users this would only need 650 – 750 blades.

That’s probably true with VDI too. The point being that 1000 seats are really more like 700 seats.

When comparing VDI (based on any Hypervisor) with another model based on physical hardware do be mindful of the potential GPU Load? By this I am referring to the fact that in most normal PC’s there is a dedicated GPU of 1/3 to ½ of the CPU & the Graphics RAM of 256 to 512Mb is also typically something like 10 times faster?

(A GPU is a Graphics Processing Unit)

The main point regarding mentioning the Graphics is that with all things being equal the more demands placed on the GPU components in a VDI instance will be “passed down” the stack internally to the Hypervisor to deal with and as a consequence the VDI solution may only able to sustain loads of say 4 – 5 VDI’s per core as opposed to say 6 – 7. Depending on how the ROI was scoped in the first place even a shift of only 1 or 2 VDI’s per core may have a significant impact?

Technically there can be no argument here. The problem for the buyer is that graphics loads are very difficult to predict and they can vary according to user behavior, even with something as simple as what wallpaper capability you select and the use of screen savers. The individual user is unlikely to have a clue about the graphics overhead of their activity.

Anyway if you’re about to embark on a desktop virtualization project then hopefully this information can help you in some way.

This is a posting in the Virtualization Focus Series. Click here to see an index of such postings.