Performance Analysis of Real Time Streaming Systems for Smart Buildings

Abstract

The Internet of Things (IoT) extends traditional cyber-physical systems by linking sensor-based edge devices to additional network-accessible services and resources. In most current IoT deployments, sensor data is streamed from edge devices to servers for storage and analysis. Analytical pipelines translate this raw sensor data into actionable information in real-time. Higher sensor densities lead to increased data volumes at higher frequencies, which increases the utilization rate of the overall system. This can lead to an increase in unacceptable response latencies for IoT analytical systems. In this paper, we compare the impact of alternative stream processing topologies for ingesting and analyzing IoT sensor data in real-time. We use real building sensor data from a commercial, LEED certified smart building on the University of Washington campus with our real-time IoT platform Namatad. We first characterize and analyze the latency impact of how the data streams are ingested and routed to analytical pipelines that predict occupancy at different levels of granularity. We then develop a queuing-theoretic analytical performance model for each of the four IoT streaming topologies. Our results show that as IoT systems continue to scale in density, server-side topology is critical to meet real-time requirements for analytical pipelines. Keywords: IoT, QoS, Real-Time Analytics, sensors, topology, queueing theory