The High-performance Flexible Service Platform For Cloud Computing

As an infrastructure, high-performance computers are widely used in cloudcomputing core business.With the expansion of the cloud-user scale, the diversificationof the type, the differentiation of user level, the traditional service model ofhigh-performance computer cannot fully meet the growing demands, thus restricting theimprovement of quality of cloud services. Based on the model of cloud computing andaimed at high-performance computer system’s security and scalability issues, this paperdesigns a high-performance and flexible service platform for cloud computing. Thebasic architecture of this platform consists of the infrastructure layer, the platformsupport layer, and the user-developed layery, whichenable ease of use, safe and efficient,customizable service environment.To fulfill the requirements of high-performance computing by cloud users, wepropose a flexible high-performance sevice plateform, including three layers.Theinfrastructurelayer abstracts the high-performanceclusters intocomputing services,anddynamicly constructs user containers based on the user requirements and theavailableresources. The plateform support layer consists of development tools andexecution environments. Users adopt these tools to use the high-performance computingservices. User developmentlayer provides a easy-to-use fat-client developmentplateform for cloud users. Besides, it also integrates an eclipse-based graphic networkdevelopment environment, which cangreatly improve the efficiency of cloud users.Cloud users share the use of environment when traditional high-performancecomputers provides cloud computing.This brings about the security and configurabilityproblems. Based on virtualization technology, this paper applies the high-performanceuser container (HPUC) to provide users with support to build a stand-alone environment.Through the operating environment, the operating system supports the flexibleconfigured system as user wish. At the same time, environmental isolation ensures thetolerence of any HPUC fault. As HPUC uses virtualization technology in the operatingsystem level, the loss of system performance is low, and the overall performance loss isless than3%, which meets the requirement of cloud computing on the high-performanceparallel computer system.Finally, according to the load balancing needs of HPUC, we studied a technology,which achieve the load balancing of the tasks andservices. Besides, we also dynamicconstruct HPUCs by combining the task processing demand andcurrent avilableresources to maximize the service efficiency. Through the parallel load scheduling,users only need to propose the parallel tasks, without the need to care about in whichnodethe task is running. Thisachieves the goal of providing high-performancecomputing capability as a service.