Server Traces:

To evaluate the performance benefits of prefetching we use trace-driven simulation. We have gathered server traces from several Web servers from a variety of environments that include universities, research institutions, and Internet providers both from Europe and the States. All traces total more than four million of requests. Specifically the traces are from:

• Parallab (www.ii.uib.no): A Supercomputing Center associated with the University of Bergen, Norway. These traces represent heavy traffic, especially from all over Norway.
• FORTH (www.ics.forth.gr): Traces from the the Institute of Computer Science, FORTH, one of the largest Research Organization in Greece.
• Rochester (www.cs.rochester.edu): Traces from the Computer Science Department of the University of Rochester, NY, USA.
• NASA: Traces from the Web Server at NASA's Kennedy Space Center.
• FORTHnet (www.forthnet.gr): Traces from the Web server of the first and largest internet provider in Greece.

We believe that it is very important to use traces from a variety of sources, since different server traces display access patterns from different client bases.

The characteristics of the traces from our servers are summarized in the following table:

We preprocessed the original traces and removed requests to ``cgi'' scripts. We have also removed requests from local (within the same domain) clients, since local users tend to reload their pages frequently (e.g. while changing them), thereby creating an artificial popularity for some pages.