A Fast Edge-Based Synchronizer for Tasks in Real-Time Artificial Intelligence Applications

Abstract

Real-time artificial intelligence (AI) applications mapped on edge computing need to perform data capture, data processing/intelligence extraction, and device actuation within some given time bounds. Synchronization across devices is an important problem that needs to be solved at different stages of an AI application. Synchronized data capture reduces the amount of time required in preprocessing data (data aggregation, data cleaning, missing data handling, etc). In the data processing phase, synchronization is key in ensuring convergence, accuracy, and speed of the distributed training process across multiple edge devices. The actuation phase in some cases requires certain actions to be performed at the same time on different devices. In this article, we develop a fast edge-based synchronization scheme that can time-align the execution of input-output tasks as well compute tasks. The primary idea of the fast synchronizer is to cluster the devices into groups that remain closely synchronized in their task executions and statically determine synchronization points using a game-theoretic solver. The cluster of devices uses a late notification protocol to select the best point among the precomputed synchronization points to reach a time-aligned task execution as quickly as possible. We evaluate the performance of our synchronization scheme using trace-driven simulations, and we compare the performance with existing distributed synchronization schemes for real-time AI application tasks. We implement our synchronization scheme and compare its testing accuracy and training time with other parameter server synchronization frameworks.