When do scalability claims not really matter?

Feb 10, 2010

Author:

Share This Post On

Greetings –

By way of introduction my name is Mike Coleman and I’m a member of the Technical Marketing staff here at VMware. Part of my role is helping our customers and partners understand our value proposition vis-à-vis other desktop virtualization offerings on the market.

Just a bit on my background – early in my career I spent over a decade as a senior level IT engineer working on some pretty big installations. From IT I moved to Microsoft and worked as a product manager on desktop operating systems including Windows 98 (I apologize), Windows 2000, and Windows XP. In January of 2008, after a brief stint at Intel helping to deliver virtualization solutions to the reseller channel, I joined VMware in the role I have now.

But enough about me – onto the topic at hand . . . .

Recently Citrix posted some documents around architecting and deploying desktop virtualization implementations, and I wanted to take a few moments to give you my perspective.

First, I think it’s great that Citrix is getting around to publishing some real scalability numbers for XenDesktop – I’m not sure they’ve gone far enough in their efforts, but I think it’s great that they’re doing something.

As most of our customers know VMware has put out a number of different reference architectures for VMware View, and our take has always been that reference architectures need to provide a comprehensive picture on what it takes to deploy an enterprise implementation of our solution. This includes not only detailing the requisite software and hardware components, but how do those components scale in a real world implementation.

Unfortunately for Citrix customers the documents released fall short in a lot of areas.

First, their “reference architecture” (titled “XenDesktop Reference Architecture”) merely outlines the multitude of components their solution requires. There is no indication to customers on how these components scale – which is the type of information companies need if they want to being to understand how to build and deploy a desktop virtualization implementation.

As mentioned before, we have always given prescriptive guidance not only on the components necessary for a successful View implementation, but also how customers can expect those components to scale.

While the “reference architecture” didn’t elaborate on scalability, Citrix did publish a couple of documents that dealt with sizing various XenDesktop features.

One of the documents entitled “Delivering 5,000 Desktops with XenDesktop 4” discusses how Provisioning Server can theoretically (I say theoretically because they actually only tested to just over 3,000 desktops) scale to deliver 5,000 desktops from a single physical server.

It certainly sounds impressive, but there appears to be some confusion inside of Citrix as some of their press materials refer to these as “virtual desktops” while other sources more accurately refer them as streamed desktops. To be clear this document is not about managed virtual desktops – it refers to PC’s being booted across the network to fat client PC’s (or diskless workstations) and operating the same way a traditional PC operates today (along with all the headaches of a traditional PC).

I will say one thing that’s interesting to note is that this solution isn’t virtualized on XenServer – it is, in fact, virtualized on VMware ESX 3.5 – finally something both Citrix and VMware can agree on.

Lastly – there is a document on single server scalability where Citrix touts being able to host 16 (actually 16.25 – but since you can’t host .25 desktops we’ll just call it 16) virtual desktops per CPU core. This is something to be proud of, we were certainly proud when we published the VMware, Cisco, EMC reference architecture (VCE RA) several months ago and demonstrated the ability to deliver 16 virtual desktops per CPU core. The thing that makes the VCE RA even more impressive is that we were able to deliver that density in only 48GB of RAM whereas the Citrix architecture uses 72GB of RAM.

If you would like to read the VCE RA you can check it out here: http://www.vmware.com/go/vce-ra-brief

In the end scaling a desktop virtualization solution isn’t about how many VMs can you cram onto one server. A true scalability discussion needs to include all the components needed for an end-to-end desktop virtualization implementation – desktop brokering, secure access, provisioning services, etc. This is the approach we’ve taken when creating our reference architectures, and it’s one that resonates with our customers

In closing, our leadership in scalability – along with our advantages in licensing costs allow us to deliver virtual desktops at a price that can come in at ½ of what a similar Citrix solution would cost. Couple that with the fact that VMware View is less complex, while delivering a complete desktop virtualization solution and you can see why more and more customers are choosing VMware View every day.

I plan to come back over the next few weeks and discuss in-depth our advantages in terms of cost, complexity, and completeness of our solution (ok that doesn’t sound great, but I’m a sucker for alliterations) – so stay tuned.

468 ad