A recent article in Information Week discusses how Credit Suisse has been very successful implementing virtualization in the data center and how they view the desktop as the next area of opportunity. By virtualizing desktops and bringing them into the data center, Credit Suisse hopes gain the ability to quickly re-provision desktops in response to changing user needs. But at what cost?

Depending on how many users Credit Suisse has, this means moving the processing from thousands of independent computing elements into the data center. Out in the enterprise, power and cooling is abundant and there is no issue with rack space. Perhaps they have ample space in their data centers.

What about user experience? There is something to be said about moving the processing as local to the user as possible. Some applications lend themselves to be hosted centrally and accessed via a browser or portal interface. Other applications including multimedia, unified communications and complex user interfaces are better served at the endpoint device to enable the best responsiveness and/or mobility.

Virtualization is a great technology and it definitely creates new possibilities in terms of OS and application portability. This might be the right solution for Credit Suisse, given their user needs, sets of applications and data center configurations. Who can blame them for wanting to build on their past success. However, there is a bigger picture to consider the correct balance of computing models for particular usage models.

It seems counter-intuitive to think that applications streamed over the network could run faster than the same applications installed locally. If the circumstances are right, it could happen!

Here is a Systems Manufacturing example. We ran a series of key tasks across a variety of configurations to collect performance metrics. Our script opened a work order in our ERP system, created packaging labels using a bar code generation program, looked up label part numbers for our product bill of materials and ran work order activity reports in three custom web-based applications.

Our baseline was a Pentium 4 desktop system (3.0 GHz). Our trial system was a Celeron 215 desktop system (1.33 GHz). Both had 1 GB RAM. The baseline system had applications installed locally on its hard drive. Applications and the OS were delivered to the trial system via streaming. Throughput time of our script on baseline system was 6 minutes 15 seconds. The same script executed on the trial system for 2 minutes and 45 seconds.

Two things come to mind to explain the difference. First, our script contained a good mix of local and remote processing to maximize our trial processor. Second, the nature of the computing model provides explanation. Applications are broken up into execution blocks so we only need to load and execute the portion of the application that we need. Further, since virtualization was used in conjunction with the application streaming, the virtual software layer makes things like registry settings easier and faster to access.

Several Intel IT folks (and others!) have expressed concern over the back-end implications of hosting a streamed computing solution. How many clients can be supported by a server? How will streaming affect the network? Well, we had the same questions so we constructed a lab experiment to find out.

Streaming was more efficient than we expected. We demonstrated that server utilization remained low and network utilization improved over time. We successfully executed a variety of applications including audio and video. We also encountered a few challenges.

Want to know more? Read our full report: Streaming and Virtual Hosted Desktop Study

In a large enterprise like Intel, there are many different ways that software is provisioned for specific usage models. The general business user's client build contains common applications to support day-to-day tasks. On top of that build, other software applications are installed by the user using a repository of installation kits. The current self-service model works but it could be improved by taking a page from the Software as a Service (SaaS) approach.

With SaaS, services are delivered on-demand over the internet using a consolidated backend infrastructure. Typically, SaaS application usage is metered and billed on a per-use basis. If we take some of these concepts into the enterprise, we can explore benefits of boosting productivity and lowering costs. Productivity from the user standpoint is improved by automatically providing application updates and patches. Users no longer have to download new versions from the repository and take the time to install them. Lower costs could include a new strategy to manage licenses. For example, better tracking of application usage could lead to a plan to reclaim unused licenses. Information generated by application usage ensures that the correct licensing is purchased during supplier negotiations for renewals.