【摘要】：采用多核處理器架構(gòu)技術(shù)現(xiàn)已成為提升處理器性能的主要手段,并逐步應(yīng)用到實時系統(tǒng)設(shè)計中。當前在嵌入式系統(tǒng)領(lǐng)域一個日見增長的趨勢是將多個相對獨立的不同關(guān)鍵性級別的子系統(tǒng)集成至一個共享的硬件平臺下,以此來降低成本,減輕重量和減小能耗。使用傳統(tǒng)的實時調(diào)度技術(shù)調(diào)度混合關(guān)鍵性系統(tǒng),不能有效的利用多處理器平臺提供的計算帶寬,造成不可接受的資源浪費。這主要是由于在可調(diào)度性分析時的最差執(zhí)行時間預測和運行時實際的執(zhí)行時間之間巨大誤差所造成的。所以在混合關(guān)鍵性任務(wù)模型被提出后,在實時系統(tǒng)領(lǐng)域?qū)旌详P(guān)鍵性系統(tǒng)的可調(diào)度性研究迅速成為焦點熱點問題。即使單處理器平臺上對這樣一個混合關(guān)鍵性系統(tǒng)的可調(diào)度分析也是非常有挑戰(zhàn)性的,在多處理器平臺上會更加困難。目前還沒有一個實際的操作系統(tǒng)支持混合關(guān)鍵性實時任務(wù)模型,制約了其在實時系統(tǒng)領(lǐng)域的實際應(yīng)用,因此迫切需要建立能夠檢驗不同混合關(guān)鍵性實時調(diào)度算法運行時性能的實時操作系統(tǒng)實驗平臺。本文在集成Litmus 2012-3內(nèi)核補丁的Linux 3.0.0版本內(nèi)核之基礎(chǔ)上,研究并實現(xiàn)了一個多核平臺下支持混合關(guān)鍵性任務(wù)模型的分層調(diào)度器,支持集成不同的混合關(guān)鍵性調(diào)度算法。與此同時,實現(xiàn)一種關(guān)鍵性單調(diào)優(yōu)先級分配的分層調(diào)度策略,不同關(guān)鍵性級別的任務(wù)采用不同的內(nèi)部調(diào)度器進行調(diào)度。實現(xiàn)了實時任務(wù)運行時的狀態(tài)轉(zhuǎn)換和動態(tài)搶占,運行時CPU優(yōu)先隊列高效管理,混合關(guān)鍵性實時任務(wù)釋放隊列、就緒隊列的高效管理,幽靈作業(yè)狀態(tài)動態(tài)監(jiān)測及處理等關(guān)鍵技術(shù)。設(shè)計實現(xiàn)了簡潔用戶庫,使得混合關(guān)鍵性實時任務(wù)的設(shè)計和創(chuàng)建、執(zhí)行相分離。大量的實驗測試和實時任務(wù)集運行時實驗驗證了本文提出的混合關(guān)鍵性分層調(diào)度框架正確性。本文工作有利于更多混合關(guān)鍵性實時調(diào)度算法的運行時性能分析與比較,為推動理論研究成果向?qū)嶋H系統(tǒng)的應(yīng)用起到了促進作用。
[Abstract]:Multi-core processor architecture technology has become the main means to improve processor performance, and gradually applied to real-time system design. A growing trend in the field of embedded systems is to integrate several relatively independent subsystems of different critical levels into a shared hardware platform to reduce cost, weight and energy consumption. Using the traditional real-time scheduling technology to schedule hybrid critical systems can not effectively utilize the computing bandwidth provided by multi-processor platform, resulting in an unacceptable waste of resources. This is mainly due to the huge error between the worst execution time prediction in schedulability analysis and the actual execution time at run time. Therefore, after the hybrid critical task model is proposed, the schedulability of hybrid critical systems in real-time systems becomes a hot issue. Even the schedulability analysis of such a hybrid critical system on a single processor platform is challenging and more difficult on a multiprocessor platform. At present, there is not a practical operating system supporting hybrid critical real-time task model, which restricts its practical application in the field of real-time systems. Therefore, there is an urgent need to establish a real-time operating system experimental platform that can test the runtime performance of different hybrid critical real-time scheduling algorithms. Based on the kernel of Linux 3.0.0 which integrates the patch of Litmus 2012-3 kernel, this paper studies and implements a hierarchical scheduler supporting hybrid critical task model on a multi-core platform, which supports the integration of different hybrid critical scheduling algorithms. At the same time, a hierarchical scheduling strategy for critical monotone priority allocation is implemented, and different internal schedulers are used for different critical level tasks. The key technologies such as state transition and dynamic preemption, CPU priority queue management, mixed critical real-time task release queue, ready queue management, ghost job state dynamic monitoring and processing are realized. A simple user library is designed and implemented to separate the design, creation and execution of hybrid critical real-time tasks. A large number of experimental tests and real-time task set runtime experiments verify the correctness of the hybrid critical hierarchical scheduling framework proposed in this paper. The work in this paper is conducive to the analysis and comparison of runtime performance of more hybrid critical real-time scheduling algorithms, and plays an important role in promoting the application of theoretical research results to practical systems.
【學位授予單位】：東北大學
【學位級別】：碩士
【學位授予年份】：2013
【分類號】：TP332
，

本文編號：2261844