Analysis of Windowing and Peering Schemes for Cache Coherency in Mobile Devices

Abstract

A major factor in determining the effectiveness of caching in wireless networks is the cache coherency scheme which maintains consistency between mobile stations (MSs) and the server. Since the wireless channel is inherently a broadcast medium, an appropriate cache coherency scheme is one in which the server broadcasts cache invalidation reports (IRs) that contain data update information. However, in a wireless environment, since MSs may connect to the network only intermittently (e.g., to save power), IRs may be missed. This would cause the MS's cache to become invalid and in turn the cache would have to be purged resulting in higher query-delay and lower throughput. One approach to improving the cache coherency for mobile devices is the Time Stamp (TS) method [2] which uses a windowing scheme. In this scheme, the IR in a particular interval contains the IRs for a number of previous intervals determined by the window size. Another orthogonal approach to improving cache coherency is the peering scheme [10] where an MS can query neighboring peers to retrieve IRs that it may have missed while it was disconnected. In this paper, we present a unified mathematical model based on Discrete Markov Models (DMMs) to study the effectiveness of these orthogonal schemes both individually as well as their relative importance when they are implemented together. The results show that both schemes are comparable for the most part. Since they are orthogonal, they can be combined in ways that is tailored for the particular environment to achieve significant improvement in performance.