Research On Multiple DAGs Worklfow Energy Saving Scheduling And Energy Measurement In Mobile Cloud Computing

With the rapid development of mobile Internet, the basic idea of the cloudcomputing is introduced to the mobile networking area–mobile cloud computing(MCC). The huge amount of mobile devices, explosive growth of computation tasksand the inherent limitation of mobile devices’ battery power have brought brand newchallenges to MCC. First, the large amount of global electric power is consumed bydata centers. Many researchers are working on reducing data centers’ energyconsumption in recent years. Meanwhile, due to the limited battery power of mobileclients, the introduction of MCC may bring in energy efficiency issue, which hasattracted wide attention in both academic and industrial areas. Large amount ofresearch has shown that the computation energy cost of the tasks on mobile clientscould be significantly reduced by offloading computation tasks to the cloud. However,in some specific circumstances, it might be impractical for mobile users to uploadcomputation tasks to the powerful data centers. Thus, the concept of mobile privatecloud is introduced, in which other mobile clients surrounding the target client couldbe utilized to help handling the tasks generated by the client, the cooperation amongthe mobile clients could be used to solve the multiple relation parallel task clusters inthe scenario of mobile cloud computing.Since Directed Acyclic Graph (DAG) is usually employed to reflect the relationamong multiple tasks in a parallel system in the literature, it is widely utilized toaddress the workflow scheduling issue composed by multiple parallel tasks. However,the existing related work mainly based on single DAG workflow scenario, which hasthe following problems:(1) They cannot achieve tradeoff between performance andenergy-efficiency in relatively more complex scenarios (such as multiple DAGs).(2)Some of the algorithms could bring significant performance improvement, but maycause large amount of energy wastage.(3) Limited by the application scenario, theycannot adapt for the brand new mobile private cloud computing scenario. To address the problems above, we propose MREO (Multiple RelationshipEnergy Optimization) and EAMRS (Energy Aware Multiple Relation Schedule), twoenergy-efficient scheduling algorithms that aim at multiple relation parallel taskclusters in the scenario of data center based traditional cloud computing and mobileprivate cloud, separately. MREO first analyzes the characteristics ofcomputation-intensive and communication-intensive tasks and reduces the number ofdata center processors by integrating the independent tasks to save data centers’energy cost, then leverages backtracking and branch-and-bound algorithms todynamically choose the best task scheduling path, and the algorithm complexity issignificantly reduced. EAMRS leverages the relaxation time between tasks, reducesthe number of mobile devices used in a fixed task cluster, and achieves considerablylarge energy conservation of mobile cloud cluster by task duplication. The energymodel employed in our dissertation both considers utilization rate and tail energycaused by transmission, which significantly improves the accuracy of energyestimation and the effectiveness of the algorithm. The simulation-based evaluationproves that the two algorithms could both efficiently reduce the aggregate energy costof multiple task clusters in the scenario of data center based cloud and mobile privatecloud, while guaranteeing task completion time and client performancesimultaneously. Finally, we also estimate the efficiency of our algorithms byleveraging different energy measuring equipment (e.g. PowerMonitor).