Steady Status Study of Distributed Data Caching in Ad Hoc Networks

Abstract

There has been extensive research on distributed data caching in ad hoc networks. However, most of them focus on how to reduce the average delay of requests and improve the packet delivery ratio, not much work has been done to study the steady-state status achieved by the distributed caching algorithms. Information related to steady-state status includes the convergence time of the distributed caching algorithms, the final data replica placement in the network, and the performance comparison of distributed caching algorithms with optimal centralized caching solution. Previous theoretical results show that to minimize the average access cost (or average search cost) in the network, the optimal number of replica of each data object is proportional to the square root (or two-third) of the data's access frequency. In this work, we empirically show that the optimal number of replicas of each data depends on not only the access frequencies of the data, but also the storage capacity of each node. We evaluate the steady-states of both cooperative distributed data caching technique and a selfish caching in ad hoc networks, and compare them to that of the optimal caching solution obtained using integer linear programming (ILP). Via ns2 simulations, we gain some insights about the steady-states of distributed data caching.