The Research Of Energy Saving Data Placement And Task Scheduling Algorithms In Distributed Systems

With the continuous development of information technology in recent years,more and more data is collected which forms the massive data. The demand ofstoring and processing the massive data is growing in the same way, data center isundoubtedly an effective and efficient way to handle this. A research indicates thatthe power consumed by American data centers occupies1.5%of the powerconsumed by the whole nation, and it’s rising rapidly. So it is important to researchhow to save energy in data centers. This paper studied how to place data in nodes andhow to schedule the tasks to achieve the energy conservation goals.In this paper we only focused on the task scheduling which on the storage nodesin batch case. Then we analyzed two different energy-saving requirements anddefined two energy-saving task scheduling problems for each of them. For the twoproblems, we proved them were both NP-Hard problems and gave twoapproximation algorithms to solve them. In order to improve the performance of thetwo algorithms, we proposed a kind of data replica strategy and data placementstrategy.In order to validate our algorithms, we used CloudSim to conduct the simulationexperiments. The experiments showed that our data-placement algorithm and taskscheduling algorithm were effective. In the first situation, given the limited power ofthe cluster, the optimization goal is to minimize the processing time to save energy.And the algorithm proposed by us can save35.7%processing time. In the secondsituation, given the deadline of processing the tasks, the optimization goal is toactivate less nodes to save energy. And the algorithm proposed has the tasks finishedwith68.1%of the cluster nodes. For the purpose of comparing the simulationcondition and real condition, we conducted emulation experiments. We used a PC tosimulate every node in the cluster and used PM1000+to measure the power of thePC, then we can get the energy consumption of the entire cluster. The emulationexperiments showed that there was hardly any difference between real environmentand simulation environment. So the algorithm proposed in the paper will performexcellent in real condition.