Creating a Framework for Developing High-performance Web Servers over ATM

During the past two years, the volume of traffic on the World Wide Web (Web) has grown dramatically. Traffic increases are due largely to the proliferation of inexpensive and ubiquitous Web browsers (such as NCSA Mosaic, Netscape Navigator, and Internet Explorer). Likewise, Web protocols and browsers are increasingly being applied to specialized computationally expensive tasks, such as imaging servers (e.g., Siemens MED and Kodak Picture Net) and database search engines (e.g., AltaVista and Lexus Nexus). To keep pace with increasing demand, it is essential to develop high-performance Web servers. The central thesis of this proposal is that to achieve optimal performance, Web servers must be able to dynamically adapt to various conditions, such as machine load and network congestion, the type of incoming requests, and the rate of requests. Therefore, we propose to develop an adaptive Web server framework and use it to empirically determine (1) the performance impact of different server design and implementation strategies, (2) the scalability of servers under high-load conditions, and (3) the pros and cons of different design alternatives.