Experimentation



next up previous
Next: Simulator Design Goals Up: Autonomous Replication in Wide-Area Previous: Deriving Network Topology

Experimentation

 

The true test of any theory is a real world implementation. Only by actually constructing and distributing a working Web push-cache system could we accurately assess its performance. Unfortunately, such an approach leaves no room for evaluating the various parameters that exist for such a system. Once distributed, any system becomes hard to change.

We decided therefore to first build an Internet simulator that would allow us to evaluate push-caching and to compare it with other popular caching methods. No simulator is perfect; simulated results are only useful insofar as they provide relative performance between competing schemes. Furthermore, in designing such a simulator it is very important to avoid building any bias into the simulator that might influence its performance. We were careful to be objective in building the simulator, gathering trace data, and setting the parameters that determine its performance.





James Gwertzman
Wed Apr 12 00:26:11 EDT 1995