Distributed and parallel systems and HOOD4

Abstract

For distributed and parallel computing the new version of HOOD, HOOD4 [1], brings a significant advantage: it decouples the logical design from the partitioning required to map software onto a net of processors. The HOOD Run-Time Support System (HRTS), introduced for HOOD4, will support an engineer to distribute the software. With the HOOD4 approach a software engineer can concentrate on the required functionality and has not to care about a certain hardware configuration which may have to be changed when he has finished the implementation. HOOD4 tackles the problem of software distribution in threefold manner: Firstly, it introduces clearly defined planes in a design where a cut can easily be done without impacting the logic of the software system. Secondly, the HRTS provides the means needed to establish the communication channels between the physically separated partitions. Thirdly, it allows to provide timing information from which a performance prediction can be derived. This allows to evaluate the performance of a hardware and software configuration already during the design phase. Based on the results of the performance prediction the optimum hardware configuration can be evaluated in advance. The HOOD4 concept for support of distributed systems was defined during the SOFTPAR project [2] by customising approaches for migration of software [3,4] and for real-time processing [5]. During the SOFTPAR project an exercise with this concept will be done for a high performance parallel C++ application [6,7] using tools of the project and a PowerPC network [8] and a workstation cluster.