Research On Cloud Computing Workloads Characteristics And Co-located Interference Analysis Methods

Cloud computing technology is widely used in current data center service platforms due to its high scalability,on-demand services and extremely low cost.More and more enterprises and individuals choose to use cloud computing platform to run and manage their applications.As people's demands for service types become more and more diverse,the workloads running in the cloud environment also presents more and more complex and heterogeneous.In order to improve the overall resource utilization of the cloud environment,cloud data centers pay more and more attention to the mixed utilization of physical hosts.Therefore,how to deploy the workloads in a more reasonable way so as to ensure the normal operation of the workloads while improving the utilization of resources effectively has become an important topic at present.The main research contents of this paper are as follows:(1)Through the resource characteristic analysis of the trace dataset released by alibaba in September 2017,select features,filtrate several sets of feature vectors that had a great impact on the workloads to describe the workload characteristics.Using unsupervised machine learning Kmeans algorithm to run the workloads in ali-Cloud to cluster,in accordance with the shortcomings of determining the clustering number K by artificial of Kmeans algorithm,puts forward the application of contour coefficient method to measure the performance of clustering number,and concluded that several kinds of workload types with the same characteristics,and several workload mode combinations which are easy to be scheduled on the same machine,they provide an important reference for the job scheduling and resource allocation in the cloud.(2)Use the Docker container to simulate a real production environment,selecting several typical workloads in the application scenarios,and use the co-located deployment strategy to perform the workloads we have selected,the hardware event monitoring metrics of workloads' microarchitecture layer is analyzed,get the certain workloads co-located patterns on the degree of interference hardware events,and makes the recommended(not recommended)co-located workload patterns,thus reducing the frequency hardware events occur,improve system performance of the servers.(3)According to the workloads characteristics and the results of interference analysis of co-located workloads,the scheduling policies at different levels of the cloud platform are analyzed in detail.Combined with the scheduling strategy of Resource Layer and Hardware Event Layer,a new workloads scheduling Model RHDLM(Resource Layer-Hardware Event Layer-Double Layer Model)for cloud data center was proposed,and the Model was compared with the traditional scheduling scheme through experiments,further verifying the feasibility and advantages of the model proposed in this paper.