Virtera's Daniel Beveridge hosted one of the more interesting Desktop Virtualization Breakout Sessions at VMworld 2009 called Efficient VDI Design - The Math Behind VDI. (VMworld 2009 attendees and VMworld.com subscription owners may login and view the full session.) Daniel is tasked with making VDI designs efficient for his clients. These days his clients are facing the challenge of getting past desktop virtualization product selection and moving towards an architecture that works in their environment meeting user experience expectations at a price point that is feasible.
Daniel was quick to point out that you can't think of VDI as just another OS in the data center. There are risks and challenges that are unique to VDI, but these can be quantitatively measured and managed making product selection easier and user experiences predictable. With desktop virtualization, there is a high level of workload volatility that increases the risk of unpredictable behavior and user experiences.
In this session Daniel demonstrates how to use some very simple statistic calculations - mean, standard deviation, and coefficient of variation - to compare performance and aid in the VDI design process. With these concrete measurements you can choose the products or processes that will lower the workload volatility, provide a better user experience, and reduce the cost of physical infrastructure investment requirements.
Using a specific example, Daniel compared two user profile management solutions – Citrix User Profile Manager and RTO Virtual Profiles – for a client. RTO Virtual Profiles was the solution of choice because it maintained a steadier workload state transferring data as needed rather than causing the huge workload burst that Citrix UPM did at logon and logoff time. RTO Virtual Profiles doesn't copy all the user's files at logon reducing the initial network bandwidth and storage access loads. Files are copied from storage only as they are requested by the user, and any changes are pushed back up in real-time which reduces large bursts of workload activity at logoff as well.
We couldn't agree with Daniel more. As we've said before, being able to quantitatively measure your performance can really make the difference between a VDI pilot stalling or moving into production. Users are not willing to settle for slow or unpredictable user experiences.
