A Rigorous Evaluation of State-of-the-Art Scheduling Algorithms for Cloud Computing

Cloud computing has recently been evolved in terms of the dynamic provision of computing resources to the users based on payment for usage on a pay-as-you-go basis. This provides feasibility to gain access to the large-scale and high-speed resources without establishing their own computing infrastructure to execute high-performance computing (HPC) applications. However, for the past several years, the efficient utilization of resources on a compute cloud has become a prime interest of the scientific community. One of the major causes behind inefficient resource utilization is the imbalance distribution of workload in a distributed computing. This paper contemplates the scheduling objectives of contemporary state-of-the-art heuristics to investigate their behavior to map HPC jobs to resources. Furthermore, the status of workload distribution in cloud computing is also critically assessed. A set of nine scheduling heuristics is validated in the CloudSim simulation environment. The potential of all the heuristics in terms of resource utilization is assessed by combining the workload balancing and machine-level load imbalance using different instances of benchmark scientific datasets (i.e., Heterogeneous Computing Scheduling Problems instances and Google Cloud Jobs dataset). The empirical assessment shows that it is not only an optimal solution to schedule the independent jobs on machines solely based on the execution time, throughput, and average resource utilization ratio; instead, the machine-level load balancing must also be considered to effectuate the usage of full capacity of computing power in a cloud system. Among all the heuristics, Resource-aware load balancing algorithm (RALBA) heuristic has outperformed, and it seems an optimal choice in terms of the tradeoff between complexity and the performance in terms of resource utilization and machine-level load balancing.