Saving Energy with Relaxed Latency Constraints: A Study on Data Compression and Communication

TL;DR

This study explores how relaxing latency constraints in edge computing can significantly reduce energy consumption by optimizing data compression and processing, challenging traditional strict latency targets.

Contribution

The paper introduces a model analyzing the tradeoff among latency, reliability, and energy in wireless devices considering compression and processing speeds.

Findings

Energy costs grow exponentially with reduced latency.

Relaxing latency constraints can lead to substantial energy savings.

Application-specific latency budgets are more effective than rigid targets.

Abstract

With the advent of edge computing, data generated by end devices can be pre-processed before transmission, possibly saving transmission time and energy. On the other hand, data processing itself incurs latency and energy consumption, depending on the complexity of the computing operations and the speed of the processor. The energy-latency-reliability profile resulting from the concatenation of pre-processing operations (specifically, data compression) and data transmission is particularly relevant in wireless communication services, whose requirements may change dramatically with the application domain. In this paper, we study this multi-dimensional optimization problem, introducing a simple model to investigate the tradeoff among end-to-end latency, reliability, and energy consumption when considering compression and communication operations in a constrained wireless device. We then study the Pareto fronts of the energy-latency trade-off, considering data compression ratio and device processing speed as key design variables. Our results show that the energy costs grows exponentially with the reduction of the end-to-end latency, so that considerable energy saving can be obtained by slightly relaxing the latency requirements of applications. These findings challenge conventional rigid communication latency targets, advocating instead for application-specific end-to-end latency budgets that account for computational and transmission overhead.