It is therefore desirable to figure out how to automatically construct predictive models of application performance. Tessellation’s performance isolation should permit us to build highly accurate models. However, we
must decide when and how these models will be built.
One solution is to profile the applications in advance.
This solution could be viable for application distribution
models like the iTunes Application Store or the Android
Market. The application distributor can profile the applications for the limited platforms it supports and provide
predictive models when the application is downloaded.
Another option leverages the Cloud by requesting every user to record performance, resource, and platform
statistics. This approach is currently being used by Microsoft Research to generate performance models of Microsoft applications for developers