A Scalable Server Load Balancing Method Using IP Address Stealing

To deal with rapid traffic growth, service providers who offer their services through the Internet must make their service system scalable. To achieve scalability, service providers often use a load balancer on the front of their system. However, load balancers tend to be a bottleneck of the system or a single point of failure. To address this problem, we propose a novel load balancing method as a way to help achieve scalable services. To eliminate the bottleneck imposed by conventional load balancers, our method utilizes IP routing. Each server has a large number of IP addresses for providing client service. When load imbalance occurs, a lightly loaded server "steals" IP addresses from a heavily loaded server. A preliminary experiment carried out using a prototype implementation demonstrates that our method is effective and that randomized monitoring topology works better than ring topologies.