Worker Resource Characterization Under Dynamic Usage in Multi-access Edge Computing

Multi-access Edge Computing (MEC), also known as Mobile Edge Computing, has gained significant momentum as a key facilitator of the stringent Quality of Service (QoS) requirements associated with delay-sensitive and data-intensive applications. Recently, the advantageous nature of MEC has been further enriched by leveraging the latent yet underused computational resources of Extreme Edge Devices (EEDs), such as smartphones, tablets, and autonomous vehicles. However, EEDs are typically user-owned devices, and thus have dynamic resource usage behavior since users dynamically navigate through various applications on their devices. This, along with the heterogeneity of EEDs, makes it harder to accurately estimate their computational capabilities, drastically affecting task allocation and resource utilization, thus increasing the delay. In this paper, we propose the Usage-based WOrker Resource Characterization (U-WORC) scheme to alleviate this problem and address the issues related to device heterogeneity, resource contention, and network communication delay. U-WORC presents a prediction-based approach to characterize the resources of EEDs (i.e., workers) by clustering the resource usage information and the corresponding execution time while running a benchmark task. Performance evaluation shows that U-WORC yields significant improvements that reach 91.42 % and 38.8 % in terms of characterization accuracy and task execution time, respectively, compared to a prominent scheme that does not consider resource contention and network communication delay.