Research On Energy-oriented Cloud Computing Task Scheduling Strategy

With the rapid development of information technology, the energyconsumption brought about by information technology industry has become oneof the issues of great concern. With the development of cloud computing, relatedapplications is increasing every year, due to the growth of cloud computing datacenter servers and supporting equipment scale, rapidly rising energy consumptionhas become an important factor affecting corporate profits, study about how tomanage the data center resources and tasks optimally to reduce energyconsumption and pollution has important significance to enterprises andenvironmental protection.Cloud computing data centers typically contain a server cluster, theseservers run a lot of applications concurrently, in this situation can consolidate thedatacenter applications and use a smaller number of servers running the task, theresource of the server can be fully utilized without resource contention, whichcan reduce costs, save energy purposes, that is the content of this paper to study.As different tasks have different demands on the CPU, memory, and othercomputing resources, in order to make the data center server resources are fullyutilized, we first need to predict the task demands of different computingresources for solving this problem. In this paper, we first proposed programresource consumption prediction model based on neural network, with thisprediction model to predict the computing resource consumption of cloudcomputing tasks. The prediction model uses various factors that affectingprogram run resource consumption as the neural network input, and programshistorical data collection as a neural network training and testing samples, topredict the program run time, CPU utilization, memory usage, disk usage as thenetwork output, to achieve the program’s performance and resources usageforecast.According to the prediction results of Cloud computing tasks resourcesconsumption, data centers can consolidate the server resources and tasks,optimize task schedule strategy. In order to reduce the running host and make thehardware resources can be fully utilized, and able to avoid resource contention, we transform task allocation problem into a multi-dimensional and multi-knapsack problem to be solved, because the task allocation problem is NP-complete problem, we design using hybrid genetic algorithm to solve thisproblem, with minimize energy consumption as the objective function, to obtainthe lowest power consumption in task allocation optimal solution, in order toachieve lower energy consumption and costs.