A prefetching model based on access popularity for geospatial data in a cluster-based caching system

Abstract

Geospatial tile popularity reflects the general characteristics of user preferences in tile access. However, tile access has both long-term popularity features (characterized as stable) and short-term popularity features (characterized as explosive). Specific features of tile popularity are an important theoretical basis for improving the accuracy of caching and prefetching. This article considers both long-term and short-term popularity features for tile access and presents a Markov prefetching model in a cluster-based caching system based on a Zipf distribution. First, it describes the navigation path and the transition probability path for tile access based on the global features of tile access to find a way to estimate the transition tile access probabilities based on the access pattern, which satisfies Zipf's law. Then, based on temporal and spatial local changes in tile access patterns, the basic Markov model is used to prefetch tiles with the highest probability in the follow-up state for current hot tiles and these tiles are labeled as the set of prefetched objects. Finally, based on the access probability for prefetched tiles, they are evenly distributed in a cluster-based caching system. This method takes into account both global and local space–time changes in tile access patterns. This method not only makes the set of cached objects relatively stable but also adapts to changes in access distribution. Experimental results reveal that this method has a higher prefetch hit rate and a shorter average response time for tile requests and thus can improve the efficiency and stability of cluster-based caching systems.