PPS: A Publish-Process-Subscribe Middleware for Predictive Supply Chains

IoT deployments collect sensor data in supply chains, industrial monitoring, and smart environments. Raw sensor data is sent to remote dashboards or the cloud. Raw data have limited insights compared to processed data, which enables valuable insights through predictive analytics and rule-based alerts. Existing data-sharing middleware, like publish-subscribe frameworks, lacks on-the-fly data processing. This forces users to handle raw data. Built-in data processing in middleware reduces user process management overhead and provides near real-time insights. We propose a novel Publish-Process-Subscribe (PPS) middleware that provides on-the-fly data processing abilities while preserving the loose coupling benefits of the publish-subscribe paradigm. Our middleware simplifies the process management efforts for the end users by allowing them to register processes and map them to real-time data streams. Besides, PPS dynamically allocates computing resources based on the data rate and perceived processing demands. As a result, the end-users receive processed data in near real-time without manually setting up processing nodes. Experimental results show that our PPS offers significant benefits at the cost of minimal performance overhead compared to the traditional publish-subscribe middleware. We conclude that the significant reductions in replicated computations among subscribers outweigh the overheads of our framework, rendering it an attractive option for predictive supply chains.