Adaptive Real-Time Predictive Collaborative Content Discovery and Retrieval in Mobile Disconnection Prone Networks

Abstract

[EN] Emerging mobile environments motivate the need for the development of new distributed technologies which are able to support dynamic peer to peer content sharing, decrease high operating costs, and handle intermittent disconnections. In this paper, we investigate complex challenges related to the mobile disconnection tolerant discovery of content that may be stored in mobile devices and its delivery to the requesting nodes in mobile resource-constrained heterogeneous environments. We propose a new adaptive real-time predictive multi-layer caching and forwarding approach, CafRepCache, which is collaborative, resource, latency, and content aware. CafRepCache comprises multiple multi-layer complementary real-time distributed predictive heuristics which allow it to respond and adapt to time-varying network topology, dynamically changing resources, and workloads while managing complex dynamic tradeoffs between them in real time. We extensively evaluate our work against three competitive protocols across a range of metrics over three heterogeneous real-world mobility traces in the face of vastly different workloads and content popularity patterns. We show that CafRepCache consistently maintains higher cache availability, efficiency and success ratios while keeping lower delays, packet loss rates, and caching footprint compared to the three competing protocols across three traces when dynamically varying content popularity and dynamic mobility of content publishers and subscribers. We also show that the computational cost and network overheads of CafRepCache are only marginally increased compared with the other competing protocols.