Robustness of a Neighbor Selection Markov Chain in Prefetching Tiled Web Data

Abstract

The service speed of tiled-web data such as a map can be improved by prefetching future tiles while the current one is being displayed. Traditional prefetching techniques examine the transition probabilities among the tiles to predict the next tile to be requested. However, when the tile space is very huge, and a large portion of it is accessed with even distribution, it is very costly to monitor all those tiles. A technique that captures the regularity in the tile request pattern by using an NSMC (Neighbor Selection Markov Chain) has been suggested. The required regularity to use the technique is that the next tile to be requested is dependent on previous k movements (or requests) in the tile space. Maps show such regularity in a sense. Electronic books show a strong such regularity. The NSMC captures that regularity and predicts the client’s next movement. However, Since the real-life movements are rarely strictly regular, we need to show that NSMC is robust enough such that with random movements occurred frequently, it still captures the regularity and predicts the future movement with a very high accuracy.