Performance evaluation of load balanced web proxies

A large university usually has many units that may locate far apart, and there are many proxy servers installed to give web access services to those units individually. However, some proxies may face the performance bottleneck when many users access the Internet simultaneously since such servers have limited capacity and could not handle high workload whereas the workload cannot be transferred to other proxies. This paper proposes an efficient load balancing technique called Shoot-Lowest Load which simulates the cache size at each proxy by counting the number and the size of log records generated by Squid proxy software in order to approximate the current workload at each proxy. We evaluate our work and compare it with the other two methods: Super Proxy and Time Round-Robin. Two metrics measured are the number of requests and the request size. In addition, the real proxy traffic is simulated as the testing dataset. The experimental results showed that our proposed method gave the best load balancing since it can closely estimate the current workload of each server so that a request will be sent to the proxy server with the lowest load.