ASTRON & IBM Center for Exascale Technology


Since a couple of years, the internal clock speed of computer chips is not increasing. In reality even a reduction of speed has taken place. Nevertheless are applications demanding more computing throughput. The predominant way to address this is either programming applications using parallel algorithms, or when that is not possible, developing special hardware, called accelerators. We can distinguish fixed accelerators (for a fixed task) or programmable accelerators (for a family of tasks which have similar characteristics). IBM research has developed several accelerator approaches, one being a programmable accelerator for problems which can be formulated as finite state machines (parsing, lookup, pattern detection etc.) Out initial gut feeling is that some of the algorithms used in today’s Radio-astronomy data processing might be able to profit from this type of technology, and achieve a significant increase in its performance. We will initially bring together exponents of the algorithms and the accelerator architect from IBM and explore together the possible solutions. From there on, we will identify the most promising and mostly needed application to be accelerated and continue the project in that direction. With the flattening of computer performance, the world will turn to special architectures for addressing next level of performance in workloads. This will be a major trend in computer technology and understanding and mastering this is a major skill for the future in highly skilled job creation, possible as basis for SKA design. There are also opportunities in the area of software radio acceleration (TU Twente). In addition we will also look at massive parallel processing as e.g. contemporary Graphics Cards.