In the eSkel project Edinburgh Skeleton Library, motivated by our observations on previous attempts to implement these ideas, we have begun to define a generic set of skeletons as a library of C functions on top of MPI. The first part of the talk will focus on the library eSkel, presenting the main fundamental concepts which form the basis of the library.
The use of a particular skeleton of the eSkel library carries with it considerable information about implied scheduling dependencies. I will present in the second part of the talk how we exploit these information in the Enhance project (http://groups.inf.ed.ac.uk/enhance), by modelling these skeletons with stochastic process algebras (PEPA). This modelling process is automated, and since grid technology provides facilities for dynamic monitoring of resource performance, our approach supports adaptive rescheduling of applications.