The Research And Implementation Of Self-Buildload Equilibrium Solution

Nowadays, the Internet expands rapidly and network is deeply in all aspects of people's lives. Website for government departments, enterprises and even personal almost become important accessory. Because web sites require specialized technical personnel and hardware, self-build system's appearance is inevitable in network times. However, such a broad user base will cause the system's concurrent pressure unprecedented. Load balancing cluster technology is a kind of mature solution to improve the high concurrent web application system performance. The load balancing algorithm used extensively, but they are not designed for a certain application system service. So, for self-build system with this kind of load balancing algorithm directly cannot achieve the best effects.This study originated from the actual self-build project. In the online tests, the load balancing server cluster with Least-Connection Scheduling bearing more than 200 concurrent requests appear bad. This paper has done a deep research for operation features of building system. This paper puts forward the operation period as the major factors of analysis method. The system operation cycle is classed as "network I/O busy period", "operation busy period", "network bandwidth busy period" and "general period". The runtime server resources consuming is much different from each other. So when heavier load period request is too many and was assigned to a same server, its performance declines. This paper use Weighted Least-Connection scheduling for self-build to complete cluster load balancing for self-build system, and proposes an improved algorithm called "period combination weighting algorithm". It considers both weights and the number of connections.This paper makes a deep comparison between this solution and minimum connection scheduling solution in experimental environment. Under completely random user requests, this solution is better than minimum connection scheduling solution under high concurrent user requests. While "computing busy period" and "network bandwidth busy period" request ratio is higher (more than 1/2), this solution is much better under high concurrent user requests.