Performance management in the cloud

Researchers from the University of New South Wales report that performance can vary by 20x. http://www.itnews.com.au/News/153451,stress-tests-rain-on-amazons-cloud.aspx

It would be good to see the details on the tests. But at a high level the article makes sense. All clouds are based on an over-subscription model. There are more virtual servers on host machines than equivalent hardware. The theory is that most hardware is utilized at the 10% level and there is excess capacity available. There is no guarantee that all the equivalent hardware that you want is available when you want it in any virtualized environment including the Amazon, Microsoft or Google clouds unless they are being managed.

One question is whether the 20x difference is significant, e.g., if response time varies from .5 seconds to 10 seconds. This may not be important to the business and the application that is placed in the cloud.

The article also says that that the clouds are not offering the tools necessary to manage service levels. In many cases, I believe that when one buys infrastructure as a service, e.g., Amazon EC2 and RackSpace, one also has to deploy application management tools in the cloud to manage availability, capacity and performance service levels. This is the same requirement that one has if one deploys real metal or virtual machines in a private environment. It will be interesting to see what Microsoft and Google offer for management tools for their web services in the cloud offerings.