Latency-dependant fitness in evolutionary multithreaded Web agents

Melania Degeratu, Gautam Pant and Filippo Menczer

Back to index

Summary

This paper discusses the InfoSpiders algorithm and focuses on the cost in energy associated with a dynamic network download time. The intuition behind this approach is that users would like to see the results of their query quickly, so penalizing an agent for utilizing high latency network connections reduces the average time it takes to download a document over the entire population. This method also leads to the possibility of a less-than-optimal relevant set of pages retrieved, since the most relevant pages may be sitting on a server with an extremely high download time, and agents will die before they can reach the most relevant pages. Experimental results show the speedup resulting from the use of the variable latency cost.

Methods

The InfoSpiders is a genetic algorithm that maintains a population of agents that use local selection to make reproductive decisions. The constant latency calculation is first presented. Then, the variable latency cost deducted from the agent's energy after download is calculated as:

cost const = (INIT_POP * THETA) / (2 * MAX_PAGES)

cost latency (p) = 2 * cost const * (time(p) / TIMEOUT)

Keywords

InfoSpiders, variable latency cost, energy, genetic algorithms, local selection

Rating

7

Bibtex Entry

@inproceedings = {degeratu01,

author = "Melania Degeratu and Gautam Pant and Filippo Menczer",

title = "Latency-dependant fitness in evolutionary multithreaded Web agents",

note = "Submitted to ECOMAS 2001",

url = "http://dollar.biz.uiowa.edu/~fil/Papers/ecomas.pdf"

}

Back to index