終端-邊緣-雲端協同系統計算資源分配之啟發式最佳化演算法

Abstract

本論文旨在解決終端-邊緣-雲端協同系統中的延遲優化挑戰。儘管移動設備、邊緣伺服器和雲端伺服器的整合共構提供了高效率低延遲的計算服務，但最小化延遲對於延遲時間敏感的任務仍然是一個關鍵議題。

為了應對這一挑戰，本研究提出了一種混合計算系統，結合了序列運算和平行運算的優勢。這個系統利用序列運算的可預測性和可靠性，以及平行運算的效率，從而優化了任務完成的延遲時間。我們所提出的系統模型還包括任務分割和部分卸載策略，並構造成一個具有最小化-最大化目標函數的非線性最佳化問題。為了解決此最佳化問題，我們將使用一種基於模擬退火最佳化技術的元啟發式演算法，此演算法生成的可行解將迭代地進行精煉與優化，以達到最佳化目標。

本研究的實驗結果將通過與現有的序列運算和平行運算系統進行比較以評估所提出系統的性能。這些權衡將利用任務完成延遲時間與資源利用率等評估指標，並與其他模型的性能進行比較，例如:隨機資源分配方案與固定分割比例方案。

總體而言，本論文所提出的系統模型在模擬退火演算法的運作下，使用較短的求解時間來獲得性能更好的解，提升用戶體驗，對於終端-邊緣-雲端協同系統的延遲最佳化展現顯著的貢獻。通過結合任務分割和部分卸載策略，以及序列運算與平行運算，此系統確認具有更高的響應性和效率。尤其對於延遲時間敏感的任務，本研究採用最小化-最大化目標函數與模擬退火啟發式演算法來解決問題，並從模擬結果中驗證其在提升系統性能具有明顯效益。

This thesis aims to address the challenge of latency optimization in end-edge-cloud collaborative computing systems. While the integration of mobile devices, edge servers, and the cloud server has provided efficient and low-latency computing services, minimizing latency remains a critical issue, especially for latency-sensitive tasks.

To address this challenge, a hybrid computing system that combines the strengths of both sequential and parallel computing is proposed. This system leverages the predictability and reliability of sequential computing and the efficiency of parallel computing to optimize the delay associated with task completion. The proposed system model also includes task splitting and partial offloading strategies which is formulated as a nonlinear optimization problem with a min-max objective function. To solve this optimization problem, a metaheuristic algorithm utilizing the simulated annealing optimization technique will be employed. This metaheuristic algorithm generates initial feasible solutions that can be further refined and optimized to achieve the objective.

The experimental results of this thesis include a comparison of the proposed hybrid system with existing sequential and parallel computing systems. The performance of the proposed system will be evaluated in terms of task completion time and resource utilization. These results will be used to compare the performance of the proposed system with other models, such as a random allocation scheme and the fixed splitting ratios scheme.

In summary, the proposed system model, operating under the simulated annealing algorithm, achieves better performing solutions in a shorter resolution time, enhancing user experience and providing significant contributions to latency optimization in end-edge-cloud collaborative systems. By utilizing a combination of task splitting and partial offloading strategies, as well as sequential and parallel computing, the proposed system is shown to be more responsive and efficient, particularly for latency-sensitive tasks. Validated by the simulation results, we conclude that the use of a min-max objective function and the simulated annealing heuristic algorithm is able to improve the performance of the system.