The dynamic job assignment problem in grid computing: A comparative study of two optimization methods

The advent of the Software Defined Networking (SDN) and Network Function Virtualization (NFV) technologies in the field of telecommunications has led to a tremendous reduction of costs as well as an increase in terms of infrastructure flexibility. Thus, equipments that play a pivotal role in telecommunications have merely been replaced by computers and their functions have accordingly been softwarized. To further reduce costs, some telecommunications carriers are now considering using decentralized grid computing-based networks to route a part of their traffic. The user-PC computer system (UPC) system is a technology for pooling and making available the unused computing resources (CPU) of the personal computers (PC) of members belonging to the same organization. For instance, in the drug discovery, economic forecasting, seismic analysis fields that deal with computationally intensive problem solving, Grid Computing has become an essential tool for building an affordable computing environment. We have previously designed an efficient static job scheduling genetic algorithm considering CPU core utilization for the UPC system. It finds an optimal assignment that minimizes the processing time of the available set of jobs in a static scenario, i.e., jobs are all available in the beginning and all worker PCs are idle. However, in reality, jobs join and leave the system dynamically and therefore, when new jobs join the system, some workers may be busy. In this paper, we propose an extension of our previously proposed static job scheduling genetic algorithm, that considers dynamic job scheduling in the UPC system. We conducted experiments using six worker PCs and up to 51 jobs. A comparative study of the results obtained using our dynamic scheduling genetic algorithm showed an improvement of about 13.67% in Makespan compared to that of the Dynamic Randomized Multi-Start Local Search (DRMSLS)-based method.