Message buffering in epidemic data dissemination

Abstract

In reliable group communication, epidemic or probabilistic protocols gained popularity due to their scalability to large number of peers and robustness against network failures. Reliability properties of these protocols are ensured via probabilistic guarantees. A key issue to consider when offering reliability is the buffer space used by individual peers of the group. Our aim is to optimize the buffer space while providing reliability in epidemic data dissemination protocols. We introduce a novel randomized model and compare it with a hash-based approach for buffer management. The effect of short and Iona, term buffering of peers and the buffer size on delivery latency and reliability are considered. We compute the performance measures through simulations of large-scale application scenarios.