If I am reading my tea leaves correctly, the Next Big Industry Push is going to be desktop virtualization or VDI. Over the holiday season, I have been boning up on the literature and checking my sources and it seems that VDI is a front burner effort at many development houses.
How we will actually virtualize a desktop is a topic still somewhat up for grabs. The path of least resistance seems to be to use a (non-)standard hypervisor on a server to present desktops as guested applications — just as some are doing with virtual servers.
Still, I see VDI as a potential game changer.
Think about it. If a desktop is just an application container, and if a desktop can be launched directly from storage, rather than using the conventional server virtualization meme (host it live and wait for someone to use it), we could start decommisioning a lot of “virtual server hosts” — perhaps a good deal faster than we have physical hosts.
Frankly, I am not a big fan of server hypervisor computing. I see it, at best, as an interim step to what we really need to get to: containerized applications that launch on demand. From where I’m sitting, hypervisors are pigs that have a horrendous impact on I/O unless you deploy a lot of extra value add software and hardware. Doing so creates a huge cost burden that no one is talking about. Most folks I know have found out the hard way that hypervisor-based virtualization creates more cost and complexity than anyone ever thought. The blush is slowly leaving the hypervisor rose.
Back to the point, desktops are a potentially huge virtualization target: 400M and counting are already deployed. Many users, including myself, are surrounded with multiple PCs doing multiple tasks. To be able to call a desktop image up on demand, do some work, then park it back on disk would save beaucoup bucks in hardware, software, management and administration. Being able to call a desktop to virtually any client, including a smart phone, thin client, old x86 box, your TV, laptop, etc. would give meaning to the marketecture term: mobility. Plus, realizing the vision of VDI would give a shot in the arm to end user recovery in disaster situations.
One of the problems I have been detecting in the literature is the same issue we saw in server virtualization: the gotchas of hidden cost. In addition to software and operating system licensing fee structures, the industry will need to sort out the enabling software for virtualizing desktops themselves. But the really big cost is — you guessed it — storage.
I have read with amusement and disdain the papers by leading storage hardware vendors stating that, from a storage cost perspective, you can stand up 5000 desktop images on their array for a fraction of what a physical desktop costs. They usually tout $100 desktops versus $400 physical PCs. Big cost savings there, right?
Not really. The reason why they use 5000 virtual PCs to achieve their groundbreaking cost reductions is that it is the only way they can amortize their box of disk drives and value add software. NOBODY IS GOING TO VIRTUALIZE 5000 DESKTOPS!!!
Sorry for shouting. I guess there are some companies out there with that many PCs. However, the truth is that VDI will likely advance incrementally — tens or maybe a couple of hundred PCs at a time. This is especially true at the outset of the trend, with early adopters and experimenters.
So, the truth of most of these claims that 5000 desktop images can be stored for a quarter of the price of an equal number of physical PCs is quite self serving — and, more importantly, falls apart in the real world.
The question shouldn’t be how many VDIs can I stack up to amortize my storage investment, it should be, how much cost can I take out of distributed desktops if I virtualize a few dozen? The benchmarks of the EMCs, NetApps, et al never go there.
But DataCore Software does. In case you missed their announcement today, here’s the news — at least the preliminary bits that were released today.
My friends at DataCore have just completed what I regard as the first meaningful benchmark on the cost of desktop virtualization from a storage hosting perspective. I like what Ziya, George, Bettye and the gang have done — a lot! In fact, I thought it was important enough to fly down there with a camera and shoot a video interview on the subject. They have posted the videos to their site but I will be placing them at the C-4 Summit in Cyberspace by end of week.
Here is an extract of what I just wrote about it for Storage Magazine in the Netherlands…
Late 2010 saw a succession of “proof of concept” papers coming from the storage brand name vendors touting cost models for VDI that were a fraction of the acquisition price of physical PCs. One three-letter vendor boasted that a 5000 PC environment could be effectively hosted on its storage array for roughly $50 US per box.
Once you get past the wow factor, however, you quickly realize that the storage vendor’s benchmark is rather self-serving. The benchmark ignores the cost of desktop OS and application software licensing, hypervisor licensing, server hardware, network enhancements, and storage cabling requirements to focus narrowly on the cost per VDI using the vendor’s storage gear. Moreover, a business will need to stack up a full 5000 virtual PCs in order to amortize the cost of the storage rig and achieve the cost-per-virtual-desktop advanced by the vendor.
Truth be told, virtually no one is going to virtualize 5000 desktops all at once – not even the large insurance companies or government research labs that actually have desktops in those numbers. It is more likely that companies will tread cautiously when pursuing a desktop virtualization strategy, virtualizing only a handful of machines at a time so that the real cost and efficacy of the strategy can be clarified. Until that happens, buying a huge EMC, IBM, HDS, etc. storage rig to support 5000 virtual PCs at sub-$100 each will not return its investment. In fact, companies pursuing this course will likely find out the hard way that the easiest path to doubling or tripling desktop computing costs is to virtualize their desktops on expensive infrastructure.
I wish everyone would read CTO Ziya Aral of DataCore Software’s wonderful benchmark on VDI and storage. His goals were not to amortize a specific storage rig (DataCore sells storage virtualization that works just as well with Joe’s JBODs as it does with VMAX or USP). He wanted simply to understand the capacity and performance requirements for desktop virtualization – especially in the 100 to 500 virtual machine range that will be much more common in the real world.
DataCore Software demonstrated pretty persuasively, and without trying terribly hard, that you could stand up that range of VDIs on a virtualized storage platform for about $35 per desktop. They also discovered that the low cost could be maintained as you grew infrastructure using storage and servers configured as part of a star topology – stars later serving as a building block for scaling. That price included all of the redundancy and failover capabilities touted as “enterprise class” hosting by the brand name storage rig vendors – leveraging only the secret sauce of DataCore read/writable snap shots created from virtualizing underlying storage.
There were many non-intuitively-obvious findings in DataCore’s research that I will only summarize here. For one, a Microsoft Windows 7 machine will boot in less than 198K of memory – so there is no real advantage to throwing a ton of RAM at the system. For another, most desktops require very little physical storage: so, when you begin doing VDI, you need to stop thinking as though you are deploying a physical PC with parameters dictated by “boundary conditions” (i.e., configuring according to what the most data intensive application or user might require in terms of storage capacity).
Bottom line: DataCore’s benchmark proves that it doesn’t require an overpriced “enterprise class” storage rig to virtualize desktops in a cost-effective way. All that it really requires is common sense and storage virtualization software, preferably from DataCore Software, which [will shortly release a new and revamped version of its flagship storage virtualization wares].
We are off to a great start in 2011. Maybe DataCore’s research will cajole the hardware guys to get real about their cost estimation around VDI. For now, a sub-$35 price tag is pretty compelling.