模擬結果顯示,在任務延遲容忍度 150ms 到 200ms、伺服器 4 台且使用者裝 置 60 台時,將任務劃分並卸載至邊緣運算伺服器進行平行運算,基於權重和叢 集運算量之叢集排序卸載策略的阻擋率會比任務無劃分卸載策略約低 16 倍,而 基於權重之叢集排序卸載策略的阻擋率會比任務無劃分卸載策略約低 13.7 倍, 皆可看出顯著效益。 此外本篇論文提出的策略在特定條件下 (例如在任務延遲約束較嚴格、使用 者數量較多或是伺服器採用不同資源分配方案時) 各自表現出不同的優勢。;Applications like virtual reality (VR), remote surgery, and smart factories are growing rapidly. Multi-access Edge Computing (MEC) effectively meets these applications’ low latency demands by placing computation servers closer to devices. However, MEC’s limited computational resources create challenges in maximizing resource utilization and optimizing task offloading strategies.
This thesis examines task partitioning strategies by breaking down tasks into subtasks and using Directed Acyclic Graphs (DAG) to model their dependencies. Subtasks are grouped into clusters for offloading, leveraging their dependencies. Offloading is prioritized based on cluster weights to improve resource allocation to edge computing servers.
Two cluster ordering strategies are proposed: one based on weight, and the other on weight and cluster cycles. The study investigates the benefits of these strategies, designing a multi-edge server resource allocation scheme to enhance task processing and meet delay constraints.
Simulation results indicate that with a latency tolerance of 150ms to 200ms, four servers, and sixty user equipment, the Weight and Cluster Cycles-Based Cluster Sorting Offloading Strategy reduces the blocking rate by approximately 16 times, and the Weight-Based Cluster Sorting Offloading Strategy by approximately 13.7 times, compared to non-partitioned offloading.
Additionally, these strategies show different advantages under specific conditions, such as stricter latency constraints, more users, or varied resource allocation schemes.