Cloud Workflow Scheduling

In the past few decades, cloud computing has been a widely usedtechnique in new distributed systems. The workflow system in cloudcomputing environment shows greater computing power. It provides betterorganizational capacity for cloud computing customers to design their owncloud resources. Though, comparing to the traditional workflow system, theruntime environment for cloud workflow makes a significant difference.Typical problems such as scheduling problem and resources organizationproblem need to be reconsidered. Among them, the resource schedulingproblem proves to be a valuable question to be solved for both customers andcloud service providers: on one hand, customers pay for cloud services. Theywant to spend less money on resources but achieve better performance fortheir workflows; on the other hand, the cloud service providers manage allcloud resources. They aim at making full use of their property and gain moremoney. At the same time, they also need to make the resources’ priceattractive to get customers’ attention. Above all, this paper will focus on thescheduling problem.In recent years, researches in the field of workflow scheduling problemhave been carried out. Algorithms such as the list heuristics and geneticalgorithms have good performances. However, some of these algorithms donot take the cost of running workflow into consideration. The othersresearches focus on decreasing the cost. But they only consider the cloudresources with consistent performance and price. A wider research aspectwould be a cloud computing environment with dynamic performances andprices. That’s exactly what this paper’s focus.Two algorithms are carried out in this paper. The first algorithm is based on the history price record for cloud resources. The dynamic planningalgorithm plays an important role here in searching the lowest cost for aworkflow within a deadline. The second algorithm is based on the researchabout cloud instance price model, which proves that the price and time thatprice changes can be predicted by certain statistical models. The algorithmaims at finding a scheduling plan whose cost is below customer’s expectationat a certain probability. At the same time, workflow under this plan shouldhave less execution time. The algorithm adopts Monte Carlo method tosimulate the runtime environment of cloud resources. Information about thedynamic effects of cloud computing is taken into consideration.As a result, both algorithms show effective ability in searching for ascheduling plan under certain restricts. Compare to the typical schedulingalgorithms, the first algorithm carried out scheduling plans with lower costfor workflows with more serial tasks than parallel tasks. The secondalgorithm considers dynamic environment and has better scheduling resultsthan other algorithms in the changing environment.