Dynamic Placement of Continuous Data Processing with Real Time Sensory Readout on IOT Devices

This thesis explores the potential benefits of offloading program logic from central computing units in a multi-sensory network onto the distributed microprocessors that previously only handled the sensory readout. We developed a new framework that developers can use to build a sensory network processing program to test this hypothesis. Developers can programmatically divide and assign the program to a set of devices. The framework then distributes and runs the program automatically on these devices. The thesis is evaluated using a concrete sensory network scenario and comparing different degrees of distribution of the program onto a fixed set of devices in a case study. The findings are that moving central computations partially to the microprocessors does not decrease throughput and latency significantly. The overall count of network connections is more important than moving computations, with three or more active network connections on the microprocessors leading to a significant decrease in throughput and latency.