The exponential increase in performance of next generation clusters and supercomputers is leading to challenges in handling tens to hundreds of thousands of cores and accelerators such as general purpose graphics processing units (GPGPU). Additionally large scale simulations produce ever larger amounts of data which makes data management and analysis problematic. In this seminar an overview of upcoming techniques for dealing with these challenges is presented and time is reserved for discussion around the topics.

Partitioned Global Address Space (PGAS) is a class of emerging programming languages designed for parallel computing; major languages of this class are Unified Parallel C, Co-array Fortran, Chapel and X10. These provide an interesting solution to the programming challenge of extreme scale computing. Hybrid computing, combining message-passing and shared-memory parallelization techniques, is already on current generation an important tool for improving scalability, and there is a clear evidence that it will be the main programming model for future systems.

We will also discuss the challenges in utilizing and programming accelerated systems, such as those containing GPGPUs.

Additionally, one theme will be data intensive computing, namely how to store, manage and and analyse the huge amount of data produced in extreme scale simulations and experiments.