BlazeFlow: a Multi-Layer Communication Middleware for Real-Time Distributed IoT Applications

摘要

The Internet of Things landscape has grown steadily over the last decade, fueling digital transformation. Data is now regarded as a precious asset. As a result, many companies have shifted their operations to the cloud to avoid maintaining massive infrastructure to hold and process all the data. However, in many settings, sending all the raw data to the cloud for processing and storage is not possible due to the unreliability of wide-area connectivity, coupled with the high costs of data transfer, storage, and processing. Also, some applications exhibit stringent response-time requirements. Edge computing can mitigate some of these issues, but it also has some drawbacks. This paper presents BlazeFlow, our vision of a multilayer data flow solution that can transport data to and from any layer of the cloud-to-device continuum based on publish-subscribe abstractions. BlazeFlow handles data flows between different services deployed onto the same node, between different devices of the same layer, and between different layers (edge, fog, and cloud). This is realized by a cross-layer common bridging solution that can automatically aggregate various protocols that are appropriate for each layer/device (e.g., ROS2, MQTT, Redis, and Kafka). BlazeFlow is designed to support bandwidth-intensive and time-critical services deployed at the various layers of the system. We evaluate a preliminary design of BlazeFlow over an autonomous robotics use case, and we show that it can sustain multi-layer data lows at a high frequency and with a low latency.