Result: External Memory Algorithms using a Coarse Grained Paradigm

We present a simple framework that allows for the use of algorithms in external memory settings that originally were designed for coarse grained parallel architectures. This framework is an extension of the model PRO as described by Gebremedhin et al. (2002). Compared to the commonly used IO model it is trading a slight (but practically not important) restriction on the internal versus external memory size for an independence of the latency of the underlying hardware. Thereby the performance of an algorithm that is described in this model will be bound to only two parameters, namely computing time and bandwidth requirements. To prove the usefulness of this setting we also describe an extension to SSCRAP, our C++ environment for the development of algorithms on coarse grained architectures, that allows for easy execution of programs in an external memory setting. Our environment is well suited for regular as well as irregular problems and scales from low end PCs to high end clusters and mainframe technology. It allows running algorithms designed on a high level of abstraction in one of the known coarse grained parallel models without modification in an external memory setting. The first tests presented here in this paper show a very efficient behavior in the context of out-of-core computation (mapping memory to disk files).