•   When: Friday, November 02, 2018 from 10:00 AM to 12:00 PM
•   Speakers: Haoliang Wang
•   Location: ENGR 2901
•   Export to iCal

As one of the key enabling technologies for large-scale Internet-of-Things applications and Cyber-Physical Systems, Wireless Sensor Networks (WSNs) provide an efficient, affordable and ubiquitous data collection, processing and distribution infrastructure.  Emerging use cases, such as many Smart City applications, have the potential to include millions of interconnected sensor nodes.  Applications running on top of such massively scaled WSNs are often designed to interact with moving subjects, such humans, vehicles, robots, etc.  These massive network sizes and complex network to network and network to non-network interactions yield new workload and mobility models that are dramatically different from those traditionally used and pose significant challenges to analyze the performance of such massively scaled WSNs. My research aims to incorporate these new scalability and workload factors to obtain modeling and performance insights into the next generation of WSN applications.

I first examine the all-to-one communications in large scale energy harvesting WSNs (EH-WSNs) with the focus on the end-to-end performance analysis. My contribution in this area is to define a novel method using Stochastic Network Calculus to account for energy underflow and overflow during harvesting and wireless transmission operations.  I demonstrate how to obtain a number of performance-related stochastic bounds for individual nodes, as well as end-to-end multi-hop tree-topology data collection networks.   

Next, I explore the all-to-all communication primitives that enable system-wide actions such as consensus and data aggregation. I demonstrate how to analyze the performance of a widely studied all-to-all communication method, namely the Synchronous Broadcasting Gossip Protocol (SBGP).  For a static SBGP environment I derive an exact result of the convergence latency for complete graphs, along with approximations for other topologies types. I then extend this work to consider the impact of realistic node mobility scenarios, incorporating multi-agent modeling and simulation techniques. My multi-agent simulation work also involves extending current simulation tools to support Smart City applications.

Using the results of this performance analysis I identify potential communication bottlenecks and propose several novel architectures to improve convergence performance of all-to-all communications.  These include a cluster-tree-based approach, where inter- and intra-cluster communications are conducted using two heterogeneous radios, and a cluster-less heuristic for sensor nodes to efficiently utilize multiple radios to improve the performance of all-to-all communications with dynamically changing network topology.  Overall, the results of my work can used by system architects to efficiently deploy large scale WSNs.