Using Dynamic Feedback to Optimise Load Balancing Decisions

The goal of a network load balancer is to distribute a workload evenly amongst a cluster of back-end servers. At present there are a number of standard request allocation schemes to distribute requests but none of these can guarantee an even load distribution. In order to improve the request allocation decisions made, I present an implementation of a dynamic feedback system as an extension for an existing open source load balancing package, Linux Virtual Server. The feedback system reports server load to the load balancer and updates the load balancer’s server weighting tables to compensate for uneven server loads; the goal of the feedback process is to restore the server nodes to an equalised state. In order to test the effectiveness of the dynamic feedback system, a simulation cluster was established and benchmarked in various configurations, with and without dynamic feedback in place. To simulate an unbalanced server load, one of the servers was given an additional CPU-intensive task to run. The results of the experimentation show that the feedback system does optimise the request allocation; CPU load of the servers is brought back to an even distribution with the dynamic feedback process running.