Dynamicity-Aware Edge Computing: Predicting Worker Availability and Utilizing Reputation Scores

Loading...
Thumbnail Image
Date
Authors
Keyword
Edge Computing , Resource Allocation
Abstract
Democratizing the edge by harnessing the underutilized computational resources of end devices, also referred to as Extreme Edge Devices (EEDs), can foster a broad spectrum of data-intensive and/or delay-sensitive applications. However, EEDs are user-owned devices characterized by a highly dynamic nature. In this thesis, we propose a framework that accounts for such dynamicity and mitigates the associated risks. In particular, we address the risk of intermittent availability of EEDs and the risk of continuous changes in their available capabilities that can lead to incongruities between their perceived and actual performance, which can profoundly impact the Quality of Service (QoS). To resolve the intermittent availability issue, we propose the Dynamic Worker Availability Prediction (DWAP) scheme. DWAP predicts the availability of EEDs (i.e., workers) and adapts to the highly dynamic nature of the computing environment at the extreme edge. DWAP employs the Continuous-Time Markov Chain (CTMC) model to forecast the availability of workers in the upcoming time-step. It does so while continuously fine-tuning the model parameters to incorporate newly available data. Towards that end, we use a dataset that consists of real-world Google cluster workload traces. In addition, we propose the DWAP-Reputation Enhanced (DWAP-RE) scheme to account for the reliability issues triggered by possible discrepancies between the perceived and actual capabilities of workers. DWAP-RE is the first scheme that uses a comprehensive reputation scoring system to assess the reliability of workers based on past performance. It then makes reliability-aware resource allocation decisions by incorporating the workers’ reputation scores into the decision-making process. Extensive evaluations show that DWAP significantly outperforms a representative of state-of-the-art prediction schemes by up to 74% and 59% in terms of the Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), respectively. Additionally, DWAP yields 97% and 48% reduction in task drop rate compared to prominent availability-oblivious and availability-aware resource allocation schemes, respectively. Moreover, DWAP-RE outperforms prominent reliability-oblivious and reliability-aware resource allocation schemes by up to 43% and 16%, and 42% and 22% in terms of execution time and satisfaction ratio, respectively. Furthermore, it outperforms prominent reliability-oblivious resource allocation schemes by 97% in terms of task drop rate.
External DOI