A task-based adaptive TTL approach for Web server load balancing

Abstract

Web sites attracting a high client-traffic cannot simply rely on either mirrored servers or a single server to balance the client-request generated load. DNS load balancing techniques have shown their advantages in dealing with heavy Web traffic. These techniques use the time-to-live (TTL) value associated with a name-to-address translation. Unfortunately name-to-address translations are cached in intermediate name servers for a period defined by the TTL This results in all requests reaching the same Web server for this TTL period. The proposed adaptive-TTL approach (called DLB-TS - dynamic load balancing based on task size) takes into account the time taken to fetch a document while choosing the least loaded server. To alleviate the problems of client-side caching and non-cooperative intermediate name servers, a server-side redirection is proposed and implemented. Though the algorithm caused degraded performance because of server-side load balancing under a light load, it reduced the client perceived latency by at least 16% when compared to existing size-based algorithms.