A load balancing method for message passing application in distributed computing systems
Nuha Al-Ameedi, Ahmed Badri Muslim Fanfakh · AIP conference proceedings · 2022
The message passing interface (MPI) application is a parallel application that uses different communication routines mixed with computation to solve the problem. When executing this application over a heterogeneous platform will lead to undesired idle times, according to the synchronization between nodes. In other words, this may cause an imbalance of computation in the system. However, the increase in the slack time between the nodes wasting the computing powers of the computing nodes. In this paper, a dynamic load balancing algorithm is applied to parallel Jacobi method is proposed. It reduces the slack times depending on some gathered information from a parallel application. The obtained results show that it improved the execution time of an application by 29% compared to the original version of the parallel application.