Scheduling Scheme And Performance Assessment For Timely Responsive Service On Cluster Server

Real-time cluster system, can meet the need of processing data by load balance and real-time scheduling. Real-time cluster server is newest and of hard difficulty in the cluster field, how to deploy cluster's great performance-cost ratio, high scalability and high reliability is the point of this topic.The main content and outcome are listed as follow:1) Provided a scalable cluster server architecture that is of high scalability and reliability.2) Provided three real-time scheduling schemes, Pull-Based Distributed Scheme totally deploys the scheduling functions on the end-node, which is a whole new way.3) Built up a RTCS application and do experiments on an 40-node cluster to evaluate RTCS performance in terms of average response time and system overall throughput.4) Built up a Pull-Based Distributed Scheme performance model to analyze and evaluate the real-time scheduling scheme.Prototype experiments on an 40-node Linux cluster show the this scheduling schemes outperform the popular load balance approach by cutting the average response time, with only little trade-off in overall throughput. Further experiment on larger number of sensors demonstrates the cluster server maintain the optimized performance under high intensity workload. In our next phase of work, is focus on the Linux kernel. We intend to create a connection between kernel and the upper scheduling, which will improve the performances in high load.