Review # 3 – Locality Aware Request Distribution in Cluster - based Network Servers

Two-tier HTTP server clusters consist of a set of front-end servers receiving incoming requests that are passed off to back-end servers, which actually serve the requests. Traditional state-ofthe-art algorithms are based on Weighted Round Robin (WRR) distribution of requests to backend servers. This provides good load balancing, but lacks performance when the total working set of requests causes frequent replacements in the caches of the back-end servers. Vivek et al. propose an algorithm to “hash” requests to specific back-end servers based on the request’s contents. This method logically partitions the back-end servers according to request content, thereby improving server affinity. Additionally, the algorithm also redistributes requests between back-end servers in order to maintain a balanced load (based on the number of active TCP connections per server).