Storage

05:36 PM
Commentary
Commentary
Commentary
50%
50%
Repost This

Calculating The True Cost Of VDI

There are a lot of reasons to go with VDI, but economics isn’t one of them; VDI is too expensive to justify on costs alone. Here’s my tally.

In recent column, storage expert George Crump postulated that a VDI installation for an organization with 250 users could be done for less than $150 per desktop--not including whatever the user actually uses for hardware connecting to the VDI backend. I'd been doing some research on the subject, and was astonished at the $150-per-user price.

The problem, according to the story, is that storage is the bottleneck. More precisely, the bottleneck is storage bandwidth, not capacity. The assessment isn't all that surprising coming from a storage expert--when what you sell are hammers, every problem will call mostly for nails.

Before I get into offering my own price analysis, let me say that I agree with a number of points of George's article. First, DRAM is really important. If you want to see storage bandwidth become the bottleneck, just oversubscribe the server RAM, and--voila--you've got a storage bandwidth problem. Second, George says early on that the economics of VDI is a challenge. I agree with that, too. There are a lot of reasons to consider VDI or some other means of hosting apps that typically run on the desktop, but economics alone is not a good enough reason; the numbers just don't support it.

The first question you might ask is, $150 per desktop for what? We know the systems sitting on the end users' desks aren't part of that number, but certainly all the back-end software and hardware are, right? Probably not. The simplest way to purchase VDI is Citrix's VDI-in-a-Box product. It comes in 10 packs, with a list price of $1,950, which includes the first year of support. That's $160 for the license and $35 per year for the support. So much for a $150-per-desktop system. The VDI software will cost more than that.

Assuming you want to run Windows 7 Pro, you can tack on another $150 per user. And, no, you can't count the licenses you may have purchased with workstations. Microsoft will not allow you to transfer them or use them simultaneously. If the Redmond Mob audits you--and they're aggressive about doing so--you'll end up with a big fat bill for software.

You can already see why the economics of VDI are a tough sell. Assuming about $200 for the terminal devices on the users' desktops, we're already up to $545 in the first year, and we haven't even gotten ourselves any servers yet.

In terms of sizing servers, CPU performance is probably not going to be the limiting factor. If you bring up the Microsoft Task Manager on your own system, you'll see that CPU use rarely pops over about 20%. Accepting that at face value, you can figure five users per core, so a server with 16 cores should accommodate 80 users or so--provided you have enough memory. A typical Windows 7 desktop computer will come with 6GB to 8GB of DRAM. Mine has 6GB. I'm running Outlook, Firefox, Word and PowerPoint, and Task Manager tells me that 42% of my memory is in use.

[ Join us at Interop Las Vegas for access to 125+ IT sessions and 300+ exhibiting companies. Register today! ]

So, let's say we can get away with giving each user 4GB in a virtual machine on our VDI server. If we stick with the 80 users and don't oversubscribe the server's memory, we're looking at 320GB. That's a pretty hefty server. If we drop the memory allocation by half, we'll have a cheaper server, but we'll also ensure that we will be exercising that storage system and probably ensure that our users will notice pokey response.

The price on that four-socket, 320GB server will be about $20,000 or about $250 per user without any redundancy. If your philosophy leads you to N+1 configurations, you'd need four servers for 250 users or $320 per user.

Now the storage system. Obviously, there will a lot of duplicated files in a 250-user system. So, clearly, you don't need 120GB per user. Let's put the number at 20GB per user. That's a 5TB system you'll need, with some room to grow, so let's put the raw capacity at 10TB. Such a system will run you about $15,000, or another $60 per user for a system with limited performance and about double that for one with good performance.

Totaling it all up, we're looking at $380 to $440 for hardware costs, $345 for software and $200 for something on the user's desk. In round numbers, it totals $900 to $1,000, or about what it costs for a very nice desktop or a pretty good laptop.

The bottom line is you can justify VDI for a number of reasons--easier support, better security, perhaps better availability or a more appropriate system for task workers--but you can't justify it on hardware and software savings. The numbers won't work. If you have a large turnover in workers, or the need to frequently reconfigure desktops, or you have many remote workers, or you want to use VDI as part of a BYOD plan, then by all means take a close look. But do your homework on costs--they'll add up.

Comment  | 
Print  | 
More Insights
Comments
Newest First  |  Oldest First  |  Threaded View
pvelikin
50%
50%
pvelikin,
User Rank: Apprentice
4/23/2013 | 9:35:27 PM
re: Calculating The True Cost Of VDI
First of thanks for recommending that people try VeloBit vBoost, Art! Perhaps you could try it as well -- we offer web-downloadable free trial version of our software and strongly encourage people to see the results for themselves before they buy. We are very confident you would be delighted with the performance.

I wanted to address several of your comments:

1.Overall cost per desktop

You are right to point out that the overall cost per desktop should account for the software, which will drive the price up. But even in your calculations, you conclude that hardware accounts for almost 50% of the overall cost. And for many organizations this cost is unacceptably high. Hardware CAPEX has been a major obstacle to VDI adoption for some time.

VeloBit can bring CAPEX under control. We do it in a non-disruptive and transparent way and enable end users to leverage their existing infrastructure. And, a dramatic reduction to the component that drives 50% of the overall VDI cost will surely made VDI adoption much more attractive.

2.Persistent vs. non-persistent desktops

VeloBit will be releasing the results of similar tests with persistent desktop. We are confident you will be impressed with these performance results.

3.Is LoginVSI a Gǣreal worldGǥ test?

LoginVSI is the de-facto industry standard benchmarking tool to test the performance and scalability of Virtual Desktop Infrastructure (VDI). It does incorporate randomness to help it simulate Gǣreal worldGǥ environments. In the test, the tool simulated a knowledge worker by opening and using up to 5 apps simultaneously. The test also includes idle time to simulate real-world users. The virtual desktop users log in in a staggered manner so that the same workloads are not running simultaneously. In the test, virtual desktop users have a random delay after they log in and the application input generated by virtual desktop users is also randomized.

4.RAM allocation for desktops and page faults

In your article, you argue that 4 GB of RAM needs to be allocated for each VM in order to properly handle page faults. If you have 256 GB of RAM, at 4 GB of RAM per VM, you still have a maximum of 64 desktops. Of course, you need to leave RAM for the OS, etc, so the maximum number of desktops you can support with 256 GB RAM is actually lower. This is fairly close to the baseline George used (55 desktops) and is common for systems that are not powered by VeloBit vBoost.

When you deploy VeloBit vBoost and put the RAM into VeloBit rather than the VMs, you create a Gǣglobal VM cacheGǥ rather than a local VM cache. Since the VMs share data, the global cache is superior; it eliminates the redundancies (to the tune of a 80%+ hit ratio). This precisely hits on the unique value proposition that VeloBit offers: VeloBit vBoost leverages a new caching algorithm, which is ideally suited for random content such as VDI data blocks. Data stored in cache is compressed using our unique similarity detection technique. The result is a cache that effectively holds much more data than its physical size, dramatically improves performance, and requires less RAM for every desktop. The net result is that you can pack 5x more desktops on the same server and storage.
awittmann941
50%
50%
awittmann941,
User Rank: Apprentice
4/19/2013 | 6:39:31 PM
re: Calculating The True Cost Of VDI
Ideally, you'd have mobile apps for mobile devices. But if you're budget constrained and have already spent piles of money to get apps on windows, you might want to get some more mileage out of that by using VDI. Not ideal, but certainly practical.
awittmann941
50%
50%
awittmann941,
User Rank: Apprentice
4/19/2013 | 6:37:49 PM
re: Calculating The True Cost Of VDI
I took a quick read through the report. I'd be interested to see how the product you were paid to test works in a real world application. You said that you were working to show how VDI could be made to work cost effectively outside of applications like the call center, which is great - but you tested with non-persistent desktops. You'd only use non-persistent desktops in an environment like call centers, so your test isn't indicative of a real world application.

I also don't know anything about the software you used to exercise the VMs, but unless there's some randomization to its function, you'd have another idealized situation where every single user is doing exactly the same thing at the same time. Again, not real world.

That said, the idea of applying smart caching to VDI specifically is a good one and that's what VeloBit (the product you're hawking) appears to do. No matter what, you need a 3 - 4 GB virtual machine for each user - Windows7 requires it. If you attempt to oversubscribe the DRAM of the server (yours had 256 GB, so 64 to 85 users should have been possible without oversubscription), you'll end up with VMs that are page faulting like crazy, or you'll end up with a hosted operating system that's page faulting like crazy, or both depending on how you configure things. By caching the identical pages of non-persistent VMs, you can indeed support a lot more users. Maybe 256 is reasonable, maybe not.

Given the highly idealized test environment here, it's probably a fair bet that in real life the results will be something like half as good as you predict here. That probably still makes the VeloBit product worth a try. If it get boosts the users per server from say, 64 to 128, that's still pretty impressive.

The notion that this at its root is a storage issue is wrong in my mind. You're basically trying to solve an acceleration/speed problem by putting a bigger gas tank on the car. DRAM is the issue here. Storage only becomes important if you've done something silly like ignored the amount of DRAM you'll need. It looks like VeloBit may do some good in making DRAM go further - it's certainly feasible.

You won't get desktops for $150 - the licensing costs won't permit it. And you should have said that you were talking about a product's performance that you'd been paid to test. That lack of disclosure is concerning.
GeorgeCrump
50%
50%
GeorgeCrump,
User Rank: Apprentice
4/19/2013 | 3:58:00 PM
re: Calculating The True Cost Of VDI
Hi Art, Thanks for reading my column. You make some fair points but we disagree on the storage calculation. As the lab test that we did that I link to below will show we were able to support far more desktops with far less memory than what you showed above. This drives down the cost of the server. We also were able to do so with far less storage. We validated our configuration with LoginVSI, so I believe it to be very realistic and as I suggest in that lab report we could have gone even denser on the desktop density.

As for the licensing prices, that was based that on some conversations with companies that are currently rolling out VDI and the pricing that they negotiated with Mircosoft/Citrix. Not on list prices, since based on my experience and research, companies with a little negotiating muscle are not paying anything close to that. In fairness I should have stated that and indicated that a companies software licensing costs will vary.

Here is the lab report, note it was a sponsored activity and we were brought into validate their findings. I am confident that we disclosed that properly and that we verified those findings:

http://www.storage-switzerland...
lgarey@techweb.com
50%
50%
lgarey@techweb.com,
User Rank: Apprentice
4/18/2013 | 6:04:45 PM
re: Calculating The True Cost Of VDI
I've been seeing a lot of VDI coverage lately, none of it focusing on what is to me the big question - do you REALLY want to drag the Windows desktop paradigm into the mobile era?
More Blogs from Commentary
Infrastructure Challenge: Build Your Community
Network Computing provides the platform; help us make it your community.
Edge Devices Are The Brains Of The Network
In any type of network, the edge is where all the action takes place. Think of the edge as the brains of the network, while the core is just the dumb muscle.
Fight Software Piracy With SaaS
SaaS makes application deployment easy and effective. It could eliminate software piracy once and for all.
SDN: Waiting For The Trickle-Down Effect
Like server virtualization and 10 Gigabit Ethernet, SDN will eventually become a technology that small and midsized enterprises can use. But it's going to require some new packaging.
IT Certification Exam Success In 4 Steps
There are no shortcuts to obtaining passing scores, but focusing on key fundamentals of proper study and preparation will help you master the art of certification.
Hot Topics
1
Scale-Out Storage Has Limits
George Crump, President, Storage Switzerland,  4/21/2014
White Papers
Register for Network Computing Newsletters
Cartoon
Current Issue
Video
Slideshows
Twitter Feed