Request distribution with pre-learning for distributed SSL reverse proxies

As network data security becoming more and more universalized, distributed Secure Sockets Layer (SSL) reverse proxies are often used in Web systems to offload CPU exhausting SSL operations from Web servers and improve the execution performance of the SSL protocol. The distribution strategy of user requests to the SSL reverse proxies is a significant factor affecting the system's performance in processing SSL operations. Aiming at improving the quality of request distribution decisions, this paper proposes a new approach for SSL reverse proxy load estimation, i.e. the family of algorithms called Load Estimation with Pre-Learning (LEPL), which estimates load using pre-learned machine learning models. Using LEPL, high accuracy of load estimation can be achieved, so that better request distribution decisions can be made. Our experimental results show that by using pre-learning, the SSL reverse proxy system's average response time can be shortened by about 30% - 50%.