Scheduling And Resource Configuration Of Video Streaming Services

Video streaming service is one of the most popular internet services in recent years.To guarantee the timely response of the user requests,this kind of service is usually constructed on large-scale server clusters.On traditional local data center,the scheduling of video streaming services mainly bases on the data locality,instead of considering the periodic variation characteristic of the service,results in the insufficient utilization of the server cluster and the waste in device cost and power resources.On the emerging virtual machine cluster of cloud platform,single kind of resource configuration cannot bring the superiority of the using-on-demand public platform into full play,which leads to a high rental cost.Aiming at above problems,we propose a power-conserving online scheduling scheme and a cost-effective cloud resource configuration for video streaming services,to lower down the power consumption of server data center and efficiently utilize the public virtual resource.The research is performed from the following three aspects,the analysis and prediction of the video streaming service characteristics,service scheduling and power consumption,resource configuration and cost efficiency on cloud platform.1.We analyze the characteristics of the video streaming services of the video-ondemand system,then propose the task length prediction based on user behaviors,and the task volume prediction based on historical data,and the analysis of their accuracy and overhead.The prediction of the video streaming service characteristic plays an important role in the task scheduling and resource configuration work.Experiments show that the two prediction algorithms all have high accuracy and low overhead,which can serve the subsequent algorithms effectively.2.Based on the task length prediction,we propose a power-conserving online scheduling scheme for video streaming services,build a mathematical model and use real-world datasets to accomplish simulation experiment.We found that the influence of task length characteristic to the video streaming service scheduling,and the power waste of the traditional scheduling strategy.We classify the video streaming services by length,and schedule them by methods like sorting and isolating.Experiments show that the scheduling algorithm can lower down the power consumption of a data center and utilize the server cluster efficiently.3.Based on the task volume prediction,we propose a cost-effective resource con- figuration for video streaming services on cloud platform,use real-world datasets and commercial cloud platform to accomplish the experiments.This algorithm combines the periodic variation feature of the video streaming service,uses hybrid kind of instances and storage resources,and allocates the type and proportion of each resource category effectively.Experiments show that such configuration have at least a 20% greater cost saving than unoptimized ones with negligible overhead.