| Computer cluster technology is considered to be one of the most effective methods to solve the growing problem of network. Within a heterogeneous cluster, scheduler distributes request from different users according to handling ability of each node, which also means each node is in relatively balanced load state. However, the time and resources consumed in every requesting process is completely different when users access to cluster, causing load tilt ultimately.Linux Virtual Server (LVS) is one of the solutions to constructing of clusters in Enterprises. Because of its static scheduling algorithm, the operation condition of node could not be reflected immediately, which also makes load tilt of the whole clusters under high load more obviously.The solution to my own engineering problem has been presented in this thesis, namely critical feedback. We need deploying dynamic feedback module of processing ability in each node. After that, scheduler could modify kernel weights and change requiring number assigned to node according to the feedback. It is proved that throughput rate of the improved cluster system is obviously increased and the longest waiting time needed for request is reduced significantly.Research has mainly been divided into five parts in this thesis:1. Introduction to engineering background: The system structure of the server cluster and load balance technology, negative feedback theory, Linux virtual server (LVS), as well as three schemes of LVS load balancing2. Analysis of the causes of load tilt: Realization of core of the LVS- IP Virtual Service (IPVS) and introduction to the existing ten kinds of scheduling algorithm,3. Solvent of the problem: Introduction to negative feedback theory and proposition of the critical feedback method. 4. Evaluation of the result: Construction of a small LVS heterogeneous cluster in laboratory and simulation test by ab tools of Apache. |
PDF Full Text Request