Wireless cache invalidation schemes with link adaptation and downlink traffic

Abstract

Providing on-demand data access in client-server wireless networks is an important support to many interesting mobile computing applications. Caching frequently accessed data by mobile clients can conserve wireless bandwidth and battery power, at the expense of some system resources to maintain cache consistency. The basic cache consistency strategy is the use of periodic invalidation reports (IRs) broadcast by the server. Recently, IR-based approaches have been further improved by using additional updated invalidation reports (UIRs) (i.e., the IR+UIR algorithm) to reduce the long query latency. However, the performance of the IR+UIR approach in a practical system is still largely unknown. Specifically, previous results are based on two impractical simplifying assumptions: 1 ) broadcast traffic is error-free and 2) no other downlink traffic (e.g., voice) exists in the system. The first assumption is clearly unrealistic as signal propagation impairments (e.g., multipath fading) and, hence, packet reception failures are inevitable in a practical situation. The second assumption is also inapplicable in real life because mobile devices are usually multipurposed (e.g., a mobile phone equipped with a browser may be used for Web surfing while having a phone conversation). In this paper, we first study the performance of the IR+UIR approach under a realistic system model: The quality of the wireless channel is time-varying, and there are other downlink traffics in the system. Our simulation results show that query delay significantly increases as a result of broadcast error and the additional downlink traffics experience longer delay due to extended broadcast period. Exploiting link adaptation (i.e., transmission rate is adjusted dynamically according to channel quality), we then propose three schemes to tackle these two problems. Our results indicate that the proposed schemes outperform IR+UIR under a wide range of system parameters.