Two-level workload characterization of online auctions

Online auctions are rapidly becoming one of the significant forms of electronic commerce for buying and selling goods and services. A good understanding of the workload of auction sites should provide insights about their activities and help in improving the quality of the service provided to their users. This paper presents a site level and a user level workload characterization of a real online auction site using data collected by automated agents. The main contributions of this paper are as follows: (i) a detailed workload characterization of a real auction site; (ii) an analysis of the presence of heavy tailed distributions in this workload; (iii) an analysis of the bidding activity during closing minutes of auctions; and (iv) an analysis of the arrival rate process of bidders and bids within clusters based on different attributes. These results can be used to devise dynamic pricing and promotion models to improve revenue throughput of online auction sites.