A parallel algorithm manager for networked workstations

Difficulties in computer solution of large problems often result from the requirements of the algorithm exceeding the space or time capacity of the computer chosen. A natural resolution of this difficulty is to parallelize the algorithm which can promise either reduc-tion in space or time, or both. However, several obstacles confront the scientist in order to successfully parallelize the algorithm. Time must be invested to study parallel architectures and to learn a parallel language. The cost of a special purpose parallel computer can easily exceed the available budget. One alternative is to use an existing network of workstations as a parallel computer. Although the cost has been eliminated, the task of learning network programming has been introduced. There are software packages which bundle network programming effectively, but provide varying degrees of guidance in the development and management of parallel program applications. To ease the effort required, we introduce a parallel algorithm manager which is developed using a message passing paradigm. An object-oriented design approach has been taken to facilitate distributed parallel program-ming. Base manager and worker classes are inherited by a user to form the instantiable classes to drive the parallel program. An example of a branch and bound algorithm and its resulting distributed form are presented. Copyright Kluwer Academic Publishers 1997