IT@Intel Client Blog

RE: Virtualizing the Client PC – A Proof of Concept

John Dunlop — Tue, 19 Aug 2008 17:19:41 GMT

Mike - Sorry for the delayed response. I was fortunate to have had an extended summer vacation.

Traditional thin client models (e.g. terminal services, VHD) are not really viable for most segments within Intel since we are 70/80% mobile platforms and derive significant productivity from mobile usages given our culture of distributed/virtual teams. Reliable, robust external connectivity is not ubiquitous enough or cheap enough yet to rely exclusively on network-dependent desktop sessions. In addition, thin client architecture would be a hugely expensive transition for us in that our data center and network infrastructures are designed for distributed client computing. Finally, since we already need to invest in centralized manageability and service delivery capabilities to reap many of the benefits you describe, it makes more sense for us to begin our transition to the holy grail of virtual containers through application and OS streaming as these are universally applicable across Intel usages. VHD would be a niche model for us and we can derive the same benefits from app streaming for mobile usages and OS+app streaming for connected desktops. Ultimately we are looking to improvements in OS streaming technology, hardware virtualization, and other strategies to move us to a virtual container compute model even for mobile usages.

The security considerations I mentioned were not in regards to data storage; rather the concerns were that the usage we were testing envisioned a managed guest on an unmanaged (perhaps user owned) host. The idea of an unmanaged host raised concerns among our info sec folks that we are still exploring with them.

Usability challenges came down to performance and the fact that some people did not like the "fusion" of host/guest desktops (they preferred a separate window on the host containing the guest desktop). In addition, the need to patch everything twice (host and guest) presented some loss of productivity.

We're looking forward to continuing work in this space to address the concerns raised by this PoC. Even since the conclusion of our testing at the end of last year, we've seen big advancements in the products & capabilities we tested in terms of performance, security, manageability, etc. We also expect that the acquisitions and convergence of the products in this space will result in simplification and improvements on all of the challenges we've identified.

John

RE: Decentralized or centralized client computing?

Jon — Wed, 28 May 2008 19:51:13 GMT

I don't see the shift going back to completely dumb terminals at the user end or even dumb terminals with a local hard drive.

There will always be applications that you will want complete control over. Banking is the big one but purchased media is another. I think the end user wants to retain control over things they pay for.

Software as a Service and networked applications always carry with them the worry that you may not be able to access what you have been promised. Take Google Video for example. They were selling videos with the assurance that you would always have them available and one day they decided that it was not in their corporate best interest to continue offering them. If your personal consumer data or corporate infrastructure are controlled by another company you will always be dependent upon them in some way which makes many people nervous.

RE: Virtual Hosted Desktops: Too much of a good thing?

nongaw — Thu, 28 Feb 2008 16:47:21 GMT

I recently helped with some testing to evaluate VHD and how it faired on a few different server configurations. The results were interesting. We found exactly what you mention regarding the appropriate workloads. Some worked well, some did not.

The really interesting thing for me was the fledgling nature of this type of environment. Virtualization as a technology and industry has not yet evolved in the area of widely accepted test methodologies. Given the new possibilities you mention, a lot of people are coming up with their own way to evaluate configurations. Look up vConsolidate, VMMark, and the pending SPEC Virtualization Benchmark. They are a great start at test and benchmark solutions, but they are effectively testing the same thing; Consolidation of workloads onto a single server. There are many more ways to implement a virtualized environment where these tests do not evaluate. More maturity in benchmarking and simply having tools to evaluate configurations is necessary.

Defining a workload for the VHD testing I was involved in was simply our definition of a light vs. heavy user load. It would be interesting to see what Credit Suisse's workloads were and if they truly mimic actual user use cases (I have no reason to believe they wouldn't.) We checked with other industry leading companies in this area and they effectively did the same. As example, VMWare also appears to have drawn a line in the sand for testing their VDI solution in this report: http://www.vmware.com/pdf/vdi_sizing_vi3.pdf

My last comment is to reiterate your 'warning' about finding an appropriate use model. Virtualized environments are not new, but their popularity is certainly refreshed. As is with most newly popular things, the buzz may exceed delivery in some areas.

RE: Could Streaming Apps run faster than traditionally installed?

josh.hilliker — Thu, 14 Feb 2008 15:13:48 GMT

great post cathy. do you have any video's on the testing you can share?

RE: Streaming Impact on the Server and Network

Jason A. Davidson — Fri, 25 Jan 2008 18:37:41 GMT

Great job Cathy & team, you have a fantastic report here! This study really illustrates that each of these emerging compute models have varied results on specific workloads.

A quote from the whitepaper "Virtual Hosted Desktop is better suited for standardized applications with relatively static screens, such as data entry. Streaming more efficiently supports multimedia and other
dynamic applications, including unified communications."

Jason A. Davidson
http://www.intel.com/go/emerge