Significant increases in WWW traffic motivates the need for high performance Web servers constructed from networked low cost computers. Load distribution within a distributed Web server is required to sustain high Web traffic demands. We propose an approach based on the Web servers themselves for Web server load distribution. Although it might seem counter-intuitive, we introduce an effective form of "static scheduling", which we call SSCR, in which each network has an associated preferable server predetermined by a static scheduler at the server site using access history. When the associated server is overloaded, the server redirects accesses to other servers temporarily. Our approach exploits temporal, spatial and geographical localities associated with Web accesses. Results from trace file simulations show improvements in load distribution and low overhead in comparison to some recently proposed approaches that use aggressive DNS TTL-based schemes. Since SSCR exploits standard features of the DNS and redirection has been supported by Web servers and browsers already, SSCR has the potential to be quickly implemented and deployed.
展开▼
