Efficient Winner Determination Techniques for Internet Multi-Unit Auctions

In this paper we develop algorithms for efficiently processing bids in a single item, N unit ( N ≥ 1) open cry auction for a large class of winner determination rules. Existing techniques consider all previously submitted bids along with the new arrivals to update the current set of winners. We propose that at most N “potential winner bids” amongst the previously submitted bids need to be used for such updates, thus significantly reducing the computation time and memory requirements. This is crucial when a large number of auctions are being conducted simultaneously. For a commonly used greedy auction rule we show that the expected number of potential winner bids may be much less than N under reasonable probabilistic assumptions.