Energy Aware Strategies for Scheduling Precedence Constrained Parallel Tasks in Dynamic Voltage Frequency Scaling-enabled Cloud Datacenter

Energy management has been recently a major concern for those who work on cloud computing systems as it plays an important role in support the fast growth of data and computing centers. Therefore, in computing systems (e.g., cloud data centers), the parameters related to energy and performance receive much attention when task-based parallel applications are to be scheduled. Some strategies (e.g., clustering, duplication, and dynamic voltage) as well as frequency scaling techniques are individually focused on the management of energy consumed and also optimization of performance parameters. The current study presents a five-step reliability- and energy-aware scheduling algorithm called ERADCD; in this algorithm, duplication and clustering strategies and the Dynamic Voltage Frequency scaling (DVFS) technique are applied to a cloud center processor in which frequency and voltage are scaled dynamically. The ERADCD’s goals are the minimization of energy consumption and also meeting the task scheduling limit. Through the two initial steps, the proposed algorithm reduces the execution time and the energy consumption of the processors when performing the tasks in the directed acyclic graph; this reduction is realized by intelligently combining the replication and clustering strategies. Then, at the third step, each task’s worst execution time is determined in each virtual machine at each frequency level. Then, the deadline for each task is specified and, in the fourth step, the tasks are assigned to the most suitable virtual machine by the DVFS technique. The main idea in the intelligent selection of the virtual machine by the DVFS technique is to use the long execution time of the tasks with the aim of decreasing the virtual machine’s frequency. The ERADCD algorithm can effectively balance the tradeoff between reliability and energy saving. Our performance evaluation study, based on both randomly and real-world directed acyclic graph, shows that our proposed algorithm surpasses the existing algorithms in terms of system reliability enhancement and energy saving.

کلیدواژه ها:

Green Cloud Computing ، Reliability ، Data Center ، energy consumption