Content-based routing in software-defined networks

Abstract

Content-based routing, as provided by publish/subscribe systems, has emerged as a universal paradigm for interactions between loosely coupled application components, i.e., content publishers and subscribers, where published content is filtered and forwarded by content filters to interested subscribers. Over the past few decades, content-based publish/subscribe has been primarily implemented as an overlay network of software brokers. Even though these systems have proven to efficiently support content-based routing between a large number of distributed application components, such broker-based routing and content filtering in software results in performance (w.r.t. end-to-end latency, throughput rates, etc.) that is far behind the performance of network layer implementations of communication protocols. As a result, the goal of this thesis is to develop methods that enable content-based filtering and routing at line-rate in the network layer by exploiting the capabilities of Software-Defined Networking (SDN). In particular, this thesis focuses on realizing a high performance SDN-based publish/subscribe middleware, called PLEROMA, while addressing major obstacles raised by data (forwarding) plane and control plane limitations of software-defined networks. More specifically, the following contributions are made in this thesis. Our first contribution is to provide methods to fulfill the functional requirements of the content-based publish/subscribe paradigm on the network layer in order to enable line-rate filtering and forwarding of published content in the data plane. We propose methods to establish paths between publishers and their relevant subscribers by installing content filters directly on hardware switches in the data plane. While the developed methods result in a publish/subscribe middleware whose performance (w.r.t. end-to-end latency, throughput rates, etc.) is significantly better than state-of-the-art solutions, a network layer implementation faces some serious challenges due to inherent limitations of software-defined networks. In fact, our next three contributions focus on addressing the problems associated with expressive filtering of content in the network layer, i.e., on hardware switches in the data plane, in the presence of hardware limitations. In particular, we address limitations w.r.t. limited flow table size and limited number of bits available for filter representation in hardware switches that curtail the expressiveness of content filters. Our contributions include various methods that use the knowledge of workload in the system to mitigate the adverse effects of these data plane limitations, thus improving bandwidth efficiency in the system. Not just the data plane, but also the control plane can have its own limitations (w.r.t. scalability in the presence of dynamically changing subscription requests) which can pose as a significant bottleneck for content-based routing on software-defined networks. As a result, our final contribution is to provide methods that enable concurrent and consistent control distribution, thus paving the way for a scalable and distributed control plane solution to high dynamics in an SDN-based publish/subscribe system.