Research On QoS-aware Task Scheduling Algorithms In Cloud Computing

Cloud computing is a newly emerging service mode based on Internet which isevolved from distributed computing, parallel computing and grid computing. Manywell-known IT companies such as Google, Microsoft, IBM, Amazon are conductingstudies of cloud computing, and some application platforms they developed have beenput into use.Task scheduling is the core research of cloud computing which studies theoptimal task allocation strategy that decides how to allocate the tasks among thecomputing resources of cloud system so that the tasks can get a balanced allocationor each task’s execution cost decreases to the minimum or the overall systemperformance is optimal. At present, the targets that existing scheduling algorithmsmainly includes the execution time, system load, economic principles and so on. Onthe basis of the above research work, considering the nature and characteristic ofcloud computing to provide powerful parallel capability, this paper designs two newtask scheduling algorithms from the perspective of reliability and response time ofQoS. The main works are described as follows:1、 Based on the cloud computing system architecture, we give the abstract taskscheduling system model, especially the mathematical model of this system model’seach component. This cloud computing model mainly includes three parts: users,schedulers and computing nodes. Each user issues tasks independently, whichfollows an Poisson distribution. The scheduler receives tasks from the users, thenbreaks each task down into task slices according to task scheduling strategy and sendsthem to the corresponding computing nodes to be executed. The computing nodeexecutes task slices sent to it, and each computing node can be modeled as an M/G/1queuing system.2、 On the basis of the abstract cloud computing system model, we design a newtask scheduling algorithm that targets at the reliability with the tool of game theory.Reliability means the steady-state availability that the computing nodes provide. This method takes the steady-state availability that computing nodes provide as the target,takes the task slicing strategy of each scheduler as the game theory, then determineseach scheduler’s task slicing schema. By comparison with balancing schema, underthe circumstances of the computing nodes’ calculation capacities are balanced not,different system size and different system load, our schema can make the system morereliable. At the same time, our task scheduling algorithm has the advantage of highconvergence speed and can make each scheduler get fair chance.3、 On the basis of the abstract cloud computing system model, we design a newtask scheduling algorithm that targets at the response time with the tool of theadjustable entropy function method. The task slices that the scheduler breaks a taskinto are executed in parallel on the computing nodes and this method takes themaximum of a task’s all task slices’ execution time as this task’s response time. Whendecomposing a task, each scheduler expects its response time to be the minimum, thusthe task scheduling mathematical model is built. With the adjustable entropy functionmethod, we give its solution and design the task scheduling algorithm. Experimentalresults show that this task scheduling schema is better than the balanced algorithm andits response time is much lower than the game-theoretic algorithm. In addition,compared with the balanced scheduling algorithm, game-theoretic algorithm is notnecessarily superior in parallel although its objection function is better.