| Summary: | Volunteer Edge Computing (VEC) is a promising solution for addressing the challenge of high round-trip latency in traditional cloud computing systems. Leveraging distributed computing resources reduces latency and improves performance. However, resource management in VEC is challenging, first, due to the uncertain behavior of volunteers, which frequently go offline unexpectedly, and second, since sequences of tasks can be executed on different volunteers, which requires transmitting data from one to another volunteer, which can lead to processing interruptions and network overhead. To address these challenges, we propose a trust-aware scheduling procedure that consists of two stages. First, we train a regression model based on lagged data suitable to accurately predict volunteer availability. Second, we assign tasks to volunteers using a metric based on the predicted availability from the first stage. The metric assesses the likelihood that a candidate volunteer can successfully complete a task and the likelihood that nearby nodes are available for successor tasks or as replacements if the processing is not completed. Thereby, we increase the chances of assigning tasks with dependencies to nearby resources, thus reducing long-distance communication and hence latency. We evaluate our approach in a discrete-event simulation using real data from Telecom’s base stations. The simulation results indicate significant improvements in task failures, task completion rates, delays, and average execution times when compared to the existing alternative algorithm.
|
|---|