Research On Strategies Of Resource Reservation For Cloud Workflows

With the rapid development of Internet, cloud computing adopting a pay-as-you-gomanner has been widely used in the field of research, business and network security.Sophisticated applications composed of numbers of independent tasks have become animportant part of application requests on the cloud platform. As an effective way to realizethe combination of cloud service, cloud workflow can model sophisticated applicationsabstractly, configure component service flexibly and execute processes automatically. Withmore and more applications are deployed on the cloud platform, the situation that muptipleusers compete for the same resource will emerge. As an important means, advance resourcereservation guarantees the resources will be available at the needed execution time whilemeeting the Quality of Service (QoS) constraints, which could ensure the QoS and improveuser’s satisfaction. However, large proportion of resource reservation requests may result inlots of resource fragments, which will degrade system performance significantly. Therefore,resource reservation strategy for cloud workflows need to be researched and improved,including how to apply resource reservation technology to guarantee the smooth running ofworkflows, and how to ease the drawbacks bringed by resource reservation.The following work is done in this paper.First, based on the analysis of existing resource reservation strategy for cloudworkflows, an improved resource reservation strategy to solve the drawbacks is proposed.In the proposed strategy, only critical tasks in the workflow need to be reserved in advance,and advance reservation request is submitted at the time when its prior task starts to execute.According to the proposed strategy, not only the decline of system performance can beeased, but also the wasting of resources cased by reserving excess resources can be avoided.Besides, the proposed strategy argues that it is necessary to set soft deadline representingthe expected finished time and hard deadline representing the latest completion time of aworkflow. This setting reduces the frequent consultations on the completion time ofworkflow and allows more resource scheduling space which could improve the systemperformance. Based on the resource reservation strategy, we model the process of resource reservation for cloud workflows and analyze the specific implementation process.Then, by introducing a fee model based on the soft deadline and hard deadline, theproposed algorithm makes full use of the given relaxed time to gain more revenue for thecloud provider on the premise of not violating the hard deadline of task.Finally, we carry out some experiments on Cloudsim. By comparing the systemperformance of Deadline Sensitive Resource Scheduling (DSRS) algorithm and RevenueOptimization Resource Scheduling (RORS) algorithm, RORS algorithm has a progress inresource utilization, requests admission ratio and cumulative revenue compared with DSRSalgorithm.