在嵌入式多核心平台上設計並實作具移植性的高效能核心間溝通函式庫

Designing and Implementing a Portable, Efficient Inter-core Communication Scheme for Embedded Multi-core Platforms

周威廷(Wei-Ting Chou)

指導教授 ： 洪士灝

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.01442

摘要

嵌入式系統的發展在最近幾年慢慢趨向多核心化,隨之而來的問題是軟體的開發、整合和移植越來越困難,而其中重要的問題之一是太多樣化的核心間溝通模式。對於軟體開發人員而言,若有一個標準的、高移植性、具延展性且有效率的核心間溝通機制,可大幅降低多核心程式開發的時程及難度。為此我們提出一套方案叫做MSG的核心間溝通函式庫,採用分層化的設計,將平台與不相依的部份從函式庫中隔分開,因此只要重做少部分的函式就能移植到新平台上。我們加入了緩衝區池以及服務處理器的設計,讓MSG在系統核心增多時,記憶體使用量不致於爆增。我們利用工研院PAC Duo平台來實驗與評估MSG的設計,在該平台上我們展現如何利用MSG的小量訊息傳輸方案以及有系統的最佳化策略來提高溝通效能。另外我們也在三個不同架構的多核心平台上展示了MSG的移植性,分別是IBM Cell,PAC Duo以及x86。最後希望我們的經驗對於現在或未來在核心間溝通函式庫的設計及開發上有實質的幫助。

關鍵字

嵌入式系統 ； 異質多核心 ； 效能 ； 移植性 ； 延展性 ； 核心間溝通 ； 協定 ； 訊息傳遞

並列摘要

The multi-core architecture is the trend of current embedded system design in recent years, but the software development, integration, and migration have become more and more difficult. The major reason of this problem is the diversified inter-core communication mechanisms. A unified, portable, scalable, and efficient inter-core communication scheme should help reduce these difficulties significantly. We proposed a scheme called MSG, which aimed to achieve those requirements. It employs a layered design to separate the platform dependent and platform independent part in our library to minimize the porting efforts. Our Buffer Pool and Service Processor designs aim to limit the growth of memory usage on many-core systems. We experimented and evaluated our design methodology with case study on the ITRI PAC Duo platform. With the design of small messages communication schemes and systematic optimization strategies, we showed how optimizations could be done to improve the performance of the MSG library. Furthermore, we illustrated the portability on three different kind of multi-core platforms: IBM Cell, PAC Duo, and x86. Hopefully, our experiences help the design and development of communication libraries for exiting and future multi-core platforms.

並列關鍵字

embedded systems ； heterogeneous multi-core ； efficient ； portable ； scalable ； inter-core communication ； protocol ； message passing

參考文獻

[4] Texas Instruments. www.ti.com.

[6] Adnan Agbaria, Dong-In Kang, and Karandeep Singh. Lmpi: Mpi for hetero-

geneous embedded distributed systems. In Parallel and Distributed Systems,

July 2006.

chronous noc architecture providing low latency service and its multi-level de-