Scalable and Adaptive Load Balancing on IBM PowerNP
Web and other Internet-based server farms are a critical company resource. A solution to the increased complexity of server farms and to the need to improve the server performance in terms of scalability, fault tolerance and management is to implement a load balancing technique. It consists of a front-end machine which intelligently redirects the traffic to several Real Servers. We discuss the feasibility of implementing adaptive load balancing with minimal flow disruption on the IBM PowerNP Network Processor. We focus our attention on the steady-state part of the algorithm and propose a PowerNP-tailored mapping algorithm derived from Robust Hash Mapping. We propose and show a fast algorithm solution (despite the simple arithmetical logic of the PowerNP), as well as a scalable approach (aiming at minimizing the packet processing time) and, finally, we present some initial performance results.