Hybrid Resource Scheduling Algorithms in Heterogeneous Distributed Computing: a Comparative Study and Further Enhancements

In the context of heterogeneous distributed systems like modern High-Performnace Computing (HPC) that must respond to unpredictable requests of variable complexity with variable resource requirements (processing power as well as storage capacity), a classical scheduling algorithm would not be suitable. Therefore, hybrid dynamic scheduling approaches have been adopted. These later have the ability to adapt over time based on the knowledge gained from the results of previous work. Several techniques are thus used to optimize these algorithms such as resources clustering. In this paper, we propose a comparative study of some of most popular algorithms in order to highlight the situations in which each algorithm is more suitable. We evaluate their performance and evolution in a realistic setting of CloudSim tool without neglecting load-balancing, and measure these performance metrics at runtime.