VM Junkie

March 10, 2009

The Client Hypervisor: revolutionizing desktop delivery

Filed under: view, vmware — ermac318 @ 6:56 am

I apologize for the lack of content in the past couple weeks, there was a death in the family and I was off the grid for a good week.

I wanted to write today about something that Brian Madden touched on recently: the Client Hypervisor (whether it’s from VMware CVP, Citrix, or MS) will be an even more disruptive technology than VDI is today. This will be the enabler that really changes the way people deploy desktops. Today, there are lots of customers (I’m on a train heading to one right now) deploying VMware View and similar technologies, and they work great where appropriate. This customer is deploying is a city gov’t and it deploying it in a library and fire stations.

But going forward, what we need to make VDI truly universal in utility is the CVP. This entry will owe a lot to Intel’s Dynamic Virtual Client group, who has done a lot of testing in this area.

Some background: Until very very recently, Intel has not been a presence in the Thin Client market. The only Thin Client that has an Intel processor in it lately is the new Dell one, and that was a tough win for them vs. AMD or VIA. You will see more Atom-based Thin Clients in the future, but even then, Intel has a vested interest in Thin Clients not replacing a real desktop in every person’s cube. That’s one less “rich client” that they sell, and profit wise I think they’d rather sell 100 desktops than one giant 4-way Xeon box. So naturally, if you talk to Intel their TCO and ROI analysis will say that Blade PCs, VDI, and Terminal Services have a higher cost than a “well maintained” desktop environment using new desktops with vPro enabled. VMware’s statistics, of course, say the opposite. 97% of statistics are bulls@#$.

What Intel really has nailed down, though, is the different ways to deliver a desktop and their strengths and weaknesses. Today we have several methods:

  1. Umnanaged Distributed Desktops
  2. Managed Distributed Desktops (vPro, streamed apps, etc)
  3. Streamed OS/Remote OS Boot (PXE/iSCSI boot)
  4. Remotely Delivered Terminal Server Session
  5. Remotely Delivered Virtual Desktop

The top 3 there all have execution taking place on the local PC, whereas the last two are Server-Based Computing (SBC) systems. Intel likes the top 3, and in fact their lab results show Streamed OS solutions (with Citrix Provisioning Server or HP Image Manager) perform very well and give you lots of the benefits of a VDI solution without the back end infrastructure.

However, the other option which falls inbetween #3 and #4 is what Intel is calling Virtual Containers. This is the Client Hypervisor, in VMware’s terms. Intel is partnering with both VMware and Citrix, so you will see a push from both companies on this front.

Virtual Containers are great because they give you the advantages of #2 and #3 – the ability to do local-execution and use local resources if available, while not requiring a remote display protocol of some kind, plus the advantages of #5 – a single virtual image to manage and deploy. It also drops the networking requirement of #3, #4, and #5, because Virtual Containers can theoretically handle being “offline” as long as you stream the entire VM down and cache it.

I think the big deal about Virtual Containers is that they can take advantage of resources as necessary. If you have a local rich client: great! Lets use its processing power and execute the VM locally. Got a thin client? No problem! We’ll use PCoIP to stream down your desktop that’s running on a server. What VMware needs to do it make this a decision that the administrator never has to worry about. The Client Hypervisor itself needs to not just be a virtualization engine but also a remote connection engine – so that when you plug this software in to your thin client or rich client device, it does whatever is appropriate. This will be the real killer-app in the VDI space.


Create a free website or blog at WordPress.com.

%d bloggers like this: