Replication-Based Load Balancing in Distributed Content-Based Publish/Subscribe

Abstract

In recent years, content-based publish/subscribe (pub/sub) has become a popular paradigm to decouple content producers and consumers for Internet-scale content services. Many real applications show that the content workloads frequently exhibit very skewed distribution, and incur unbalanced workloads. To balance the workloads, the literature of content-based pub/sub adopted a migration scheme (Mis) to move (a subset of) subscription filters from overloaded brokers to underloaded brokers. In this way, the publications that successfully match the moved filters are then overloaded, leading to balanced workloads. Unfortunately, the scheme cannot reduce the overall matching workloads. In the worst case, suppose that all brokers suffer from heavy workloads. cannot find available brokers to offload the heavy workloads of those overloaded brokers, and fails to balance the workloads. To overcome the issue, the contribution of this paper is to develop a set of novel load balancing algorithms, namely a similarity-based replication scheme (Sir). The novelty of is that it not only balances the workloads of brokers but also reduces the overall workloads. Based on both simulation and emulation results, the extensive experiments verify that can achieve much better performance than, in terms of 43.10% higher entropy value (i.e., more balanced workloads) and 46.39% lower workloads.