Performance Improvement in Multiprocessors Using Three Steps of Non Contiguous Migration for Online Mapping

In this paper, we have presented different concepts and parameters in the online mapping for different jobs in the network on chips. Thus, three essential steps are considered which are finding the appropriate size of sub-mesh, finding a sub-mesh place in integrating the mesh for online task allocation and finding the main place in sub-mesh for online task mapping. For this purpose, efficient previous models to select the dimensions of the sub-mesh, the processor migration methods based on the two-row boundary, limited left-right compaction, limited top-down compaction, online dynamic compaction-four corner (ODC-FC) and hybrid migrations for mesh topology are compared with the proposed algorithm to check the comparative performance. Also, the impact of different performance parameters which are average job execution time and average system utilization will be compared against the previous mechanisms to achieve the appropriate configuration of the network on chips. It is worth noting that in this article, 7 algorithms, which have achieved better performance, have been selected among the 29 ones. We have demonstrated that using hybrid migration strategies enable us to limit the number of processors migrations. Consequently, significant improvements have been achieved in the average job execution time (%36 ~ %38.1), and the average system utilization (%38.2~%48.5).