Abstract
Web prefetching is one of the effective solutions to reducing the user perceived latency, but if not properly managed, it may consume too much network bandwidth and system resources. The side effect can partly be avoided by prefetching control. Because existing cache occupants may have to be ejected to give room for the prefetched items, the cache hit ratio is not only affected by the remained cache occupants, but also by the evicted items from the cache and the prefetched items. This paper discusses three adaptive prefetching control models based on prefetch-cache interaction. The cost function of prefetching is analyzed and further formulas for the threshold to determine its value dynamically in terms of system load, network condition and cache hit ratio are derived. Preloading the items, whose access probabilities exceed a certain threshold can minimize the cost function. Simulations show that the prefetching control models can manage prefetching effectively based on prefetch-cache interaction, and reduce the excessive consumption of the system and network resources.