In this paper we will implement a new version of master-worker architecture that improves the previous ones. The commonMaster-Worker paradigm consists of two entities: a master and multiple workers. The master is responsible for decomposing the problem into small tasks and managing them until all tasks are completed. Therefore, the master should endures heavy load either communication or computation. This bottleneck in the master process typically occurs when the number of workers increases because the master process will not be able to keep all workers equally busy. The paper presents a novel technique for hierarchically nesting the basic master-worker scheme. This technique resolves the said problem by presenting a hierarchical scheme and reduces the communicational messages due to the usage of the Linda model. The obtained results for large matrix multiplication case study on a real cluster show the effectiveness of our model.

Keywords