Research On ASAS Cluster And Its Intelligent Control Mechanism

Cluster system has good performance at high reliability, high availability, scalability, load balancing and price/performance ratio, it is the main aspect to develop high performance server. ASAS(Automatic Self-Allocating Server) is invented under the situation. ASAS cluster system divides and shifts the job of load balancing from special load balancer to back-end application servers, the only information used for scheduling in the front-end dispatcher is capacity number sent by back-end servers, the main scheduling job is actually transferred to back-end servers.At the beginning of this paper, the basic concepts, fundamental principles, and architecture is introduced, intelligent control techniques are also presented. The essential principles of cluster are followed by interpretation of ASAS cluster in detail, especially its automatic self-allocating policy, system architecture and performance.The function relationship between resource usage in back-end servers of ASAS and capacity number required is nonlinear, and there is no clear mathematic model, so traditional control method can't satisfy the requirements. As Fuzzy control does not need precise mathematic model of the controlled object. So we apply fuzzy control method to deal with the fuzzy variables used in resource status, moreover the required expert experience is introduced in, so we designed a fuzzy capacity control algorithm(FCCA) to solve capacity control problem in back-end server, inputs of the algorithm are cpu utilized ratio, memory usage and request response time, while output of the algorithm is link number(capacity). FCCA works well. With fuzzy controller, load is balanced, resource usage is more reasonable, system parallel processing ability is enhanced, and QoS is also guaranteed. But the most important is, it realized the core idea of ASAS cluster, which is automatic self-allocating.Fuzzy logic control represents quantitative values using linguistic concepts and solves the fuzzy inference problem, while artificial neural network (ANN) has very strong learning ability. Having considered the characteristic of capacity controlling problem in cluster, this paper adopted ANN and combined with fuzzy control technique. a fuzzy neural network controller(FNNC) for back-end server is designed in ASAS cluster system. Because FNNC inherits advantages from both fuzzy control technique and neural network control technique, so it achieves better performance.