異構多核系統（Heterogeneous multicore system）是現代計算系統中的重要 架構之一，由不同類型的核心組成，這些核心在性能、架構、功耗和功能上存 在差異。例如，這種系統可能包括高性能的核心，適合處理需要大量計算的任 務，以及低功耗的核心，適合處理較輕的任務或保持系統待機狀態。這種異質 性的設計帶來了資源管理的靈活性和任務調度的優勢，有助於提高系統的整體 性能和效率。然而，這也使得任務分配和管理變得更加複雜，需要精細的調度 策略和優化技術來最大化這些不同類型核心的優勢。 該研究專注於在異構多核系統中進行任務複製和映射技術的優化，以在提 升系統的容錯能力和能源效率之間達到平衡。研究首先確定最佳的任務複製數 量和運行頻率，然後根據不同類型核心的特性和系統當前的狀態，靈活地將任 務複製映射到合適的核心上，並進行核心電壓和頻率的調整。這種精細的優化 策略有助於在滿足任務需求的同時，最大化系統的性能和壽命，並盡量減少能 耗。 實驗結果顯示，該方法顯著延長了系統的壽命，減少了功率，並且能夠與 現有的方法無縫整合。該研究還探索了使用不對稱策略進一步提升系統壽命和 能源效率的可能性，這顯示了該方法在異構多核系統中的廣泛應用前景以及潛 在的突破性影響。研究的成果有助於推動異構多核系統在各種計算環境中的應 用，例如在雲計算、物聯網、人工智能等領域，為未來的高性能計算系統提供 了新的思路和方法。;Heterogeneous multicore systems are a critical architecture in modern computing systems, comprising different types of processor cores that vary in performance, architecture, power consumption, and functionality. For instance, these systems may include high-performance cores suited for handling computation-intensive tasks and low-power cores ideal for lighter tasks or maintaining system standby. This heterogeneity offers flexibility in resource management and task scheduling, aiding in enhancing overall system performance and efficiency. However, it also complicates task allocation and management, necessitating precise scheduling strategies and optimization techniques to maximize the advantages of these different types of cores. This study focuses on optimizing task replication and mapping techniques in heterogeneous multicore systems to strike a balance between enhancing system fault tolerance and energy efficiency. Initially, the study determines the proper number of task replications and operating frequencies. Then, it flexibly maps the task replications to suitable cores based on the characteristics of the different core types and the current system state, adjusting core voltages and frequencies accordingly. This fine-grained optimization approach helps to meet task demands while maximizing system performance and lifespan and minimizing energy consumption. Experimental results demonstrate that this method significantly extends system lifespan and reduces energy consumption, seamlessly integrating with existing methods. The study also demonstrates the broad applicability and potential breakthroughs of this approach in heterogeneous multicore systems. The results contribute to advancing the application of heterogeneous multicore systems in various computing environments iii such as cloud computing, the Internet of Things (IoT), and artificial intelligence (AI), providing new ideas and methods for future high-performance computing systems.
