Cluster Server System Scalability And Its Implementation

As widespread popularity of the network, there lies some problems. On the one hand, Internet network hardware and bandwidth have imbalancing speed in development, which is a bottleneck restricting the provement of Internet. On the other hand, the services of network are more and more strict in availability, reliability and manageability. So what we should do is to reach an agreement between the develepment of Internet and the increasing demand. We must make effort to solve these problems, while the cluster provides new methods to deal with the chanllege.Salability is an important sympol to measure cluster's performance. But, what we aim at isn't only cluster's salability. The main requirements that a cluster should satisfy are:dealing with different types of attributes, discovering clusters with arbitrary shape, ability to deal with outliers, interpretability and usability. To design a cluster with such advantage, we should get implication of salability and follow some rules, including independent principle and blanced principle.This paper compares a variety of server cluster's architecture, which include mirroring cluster, DNS cluster and scheduling cluster. These clusters base on some crucial technologies of SSI, HeartBeat and Load Blancer. Then this paper puts forwards a new architecture that combines distributed scheduler with centralized scheduler. Ater taking the new architecture, the cluster can reduce the intensity of scheduler's task and avoid bottleneck to occur. The system manages cluster's nodes through a logical structure and monitors the running status of nodes by the mechanism of heartbeat. By model of Message Passing, the nodes can dynamically join in and leave the cluster system. With the help of the mechanism of monitor, the cluster system can master the system's dynamical extension and take action to maintain the data's consistence. Finally, analysising system by building model, this paper proves the correctness of this strategy.Based on parallel programming function of Message Passing, combined with programming language C++, this paper builds a parallel programming environment. Then we realize dynamic expansion of nodes in cluster system. After that, changing the scale of cluster's problem, we give comprehensive analysis and comparison about system's performance. Acceleration, parallel efficience and running time show that the system has good salable performance.