Creating a more efficient Internet of Things through predictive analysis of resources

Abstract

The greatest driver behind the stellar growth of the Internet of Things in the last couple of decades have been their minimal energy requirements. While this low energy envelope have made them near ubiquitous, it is also a huge bottleneck in real world deployments. A fine balance needs to be achieved such that these devices can perform enough meaningful computation in a reasonable amount of time to justify their deployment while consuming a reasonable amount of power to keep them running for extended periods. A common method to achieve this is through computational offloading wherein these devices move the heavy computation away to the edge and the cloud thereby increasing the network lifetime. Offloading computation from these devices also improves their performance as the computations are performed by powerful hardware at the edge and the cloud. However, this increases power consumption and network latencies from the need to transmit and receive extra synchronization data. Optimum performance can be achieved by intelligently deciding if the computation should be offloaded or performed locally - a task that is inherently difficult given the transient nature of the network and the systems. In this work, we formulate a distributed network optimization problem to achieve a balance between performance and energy consumption. Our simulation results show substantial network latency reduction and overall reduced power consumption.