High performance computing is used to run models of the real world. This is true in both ultra high performance simulations used in scientific computing to study Physics, Biology etc. as well in designing approximations of real world involving applications such as gaming, social networks etc. Maximizing realism of the world being modeled and mimicked is the key to get this right. This work attacks the problem of realism on two fronts : first a framework is developed to speed up sequential parts of the computation using a large number of available cores based on a new concept of probabilistic speed-up. Secondly a runtime solution is devised to maximize realism in immersive applications such as gaming under the constraint of responsiveness. The frameworks involve development of programming models, interfaces, APIs and run-time system to solve the above problems by managing the underlying parallelism and computation. Apart from research this effort involves developing a cross-cutting graduate level course that spans between simulations, algorithms and programming languages.
