Optimizing Web server load distribution with evolutionary computation

Domain Name System (DNS) is a distributed system database that maps names to network address location, providing information critical to the operation of most Internet applications and services. Traditionally, DNS assumes equal queue size for each web server in a round-robin setting. However, the overall performance of such techniques may be poor due to imbalance in the workload among distributed web servers. The non-uniformity of the load from different client domains and the high variability of real request workload on the Internet introduces an additional degree of complexity to the load distribution issue. These characteristics make existing policy architectures not applicable for spreading the workload of web server clusters. This motivates research in entirely new load distribution policies that require some system state information.; In this research, DNS with proximity scheduling using an evolutionary computation has been proposed. This kind of policy distributes HTTP requests from clients by clustering, ranking and learning classifier system (LCS). LCS, particularly XCS, is a significant technique from an evolutionary computation paradigm for learning to act in a dynamic environment. The proposed policy promises to assign the appropriate web sever to the client. This means the proposed policy also provides quality of service (QOS) to the client. The research uses round-robin DNS (RR-DNS), a model to compare with the proposed policy. The Results indicate that the proposed policy performs better than RR-DNS.