Optimized dynamic load balancing in distributed exascale computing systems

Abstract

The dynamic nature of new generation scientific problems needs undergoing review in the traditional and static management of computing resources in Exascale computing systems. Doing so will support dynamic and unpredictable requests of the scientific programs for different type of resources. To achieve this facility, it is necessary to present a dynamic load balancing model to manage the load of the system efficiently based on the requests of the programs. Currently, the distributed Exascale systems with heterogeneous resources are the best branch of distributed computing systems that should be able to support the scientific programs with dynamic and interactive requests to resources. In this thesis, distributed Exascale systems are regarded as the operational and real distributed systems, and the dynamic load balancing model for the distributed controlling of load in the nodes in distributed Exascale computing systems are presented. The dominant paradigm in this model is derived from Operation Research sciences, and the request aware approach is replaced with the command-based approach in managing the load of the system. The results of evaluation show us the significant improvement regarding the performance by using the proposed load balancing mechanism in compare with the common distributed load balancing mechanisms