Distributed high performance computing in heterogeneous environments with DOTS

This paper deals with high performance computing in heterogeneous clusters using the Distributed Object-Oriented Threads System (DOTS). DOTS is a paralleliza-tion platform that enables the programmer to build dis-tributed parallel applications using a high level and easy-to-use parallel programming paradigm. It integrates a wide range of different systems (from realtime kernels to main-frames) into a single system environment for high perfor-mance computing. A special design focus of DOTS is to optimize usability aspects. Additionally, the paper discusses three advanced fea-tures of DOTS that especially support its deployment in het-erogeneous environments for high performance computing. For defining the parallel environment and for documenting and evaluating the results of program runs an integrated XML centric approach is presented. A light-weight check-pointing mechanism which is seamlessly integrated into DOTS and especially designed for heterogeneous environ-ments is described. DOTS is based on a highly customizable communication kernel. Run time measurements are pre-sented that prove the efficiency of the communication ker-nel in heterogeneous clusters. Also, performance measure-ments of a non-trivial parallel application from the field of Symbolic Computation are presented, which revealed good speedups on a heterogeneous cluster of workstations. 1