Research Of The Job Scheduling Algorithm On Cloud Platform Of Smart Grid

With the development of smart grid, the national interconnected power system hasbecome a trend. Numbers of advanced data collection and monitoring equipments such asphasor measurement unit (PMU), smart meters are being widespread applied, so thepower system is evolving into a computing systems, which is gathering big data andinformation. In order to satisfy the urgent needs of the smart grid for massive datastorage and massively parallel computing, a new computing model has been mentionedby scholars who are engaged in research in terms of power. This model is cloudcomputing which has gradually matured in IT circles.Given the integrity of the WAN ofpower system, scholars come up with the concept of Power Private Cloud,which canintegrate the widespread internal storage and computing resources of power system.Thetopology of Smart Grid cloud platform is similar to the distribution structure of powerresources.so there is a cluster node widespread problem of heterogeneous resources.Thedifferent capabilities of node distributed in power cloud and the different submitted timeof the job both lead to prominent task synchronization problems.According to the case mentioned upon,this paper ameliorates the job schedulingalgorithm working in heterogeneous environment. In this method some tasks of the jobwill be executed firstly, then calculate and evaluate the average execution time of thewhole job. At the same time, the node in the cluster that performs the same operation oftask execution are divided into fast and slow node node by the time it was consumed.And the Speculative execution task can only occur in the fast nodes. Furthermore, thealgorithm will as locally as possible assign the task to the nearest node away to the inputsplit. This paper compared the algorithm capability scheduling algorithm, Fair-Sharescheduling algorithm, LATE algorithm, and PWSW algorithm, analyzing the advantagesand disadvantages of them.We use three programs which represents the different resourcerespectively, memory, CPU, network to test the algorithm. The classic routine we usedare WordCount, CPUActivity, URLGet. We conduct3experiments,Each of themincludes6tests.Results show that This algorithm can reduce the time consumption, Theincidence of speculative execution is also abated, meantime reduces the consumption ofresources, and improve the task completion rate. Therefore, this method to a certainextent is suitable for the clusters that possess many nodes and complextopology,especially appropriate to the structure like Power Private Cloud.