Design and Implementation of a Highly Dependable Systems on a Resources-Limited and MMU-Less Embedded System

Abstract

In recent years, the extremely progress of hardware and software technologies, powermanagement scheme, and SoC techniques enable the proliferation of embedded systems. Thus,embedded systems can always be found in our surrounding life and place an important role inthe work, entertainment, and recreation. Nevertheless, user demands for a dependableembedded system increase when more and more jobs are executed on embedded systems. Inother words, it is not acceptable that a hardware or software failure in an embedded systemcauses the data or job to be lost.Although there have been many researches addressing the dependability issues inpersonal computers, server systems, and cluster computing, however, none of them can bedirectly applied to the embedded systems. This is because embedded systems have some quitecharacteristics that make previous scheme either useless or inefficient. These characteristicsinclude: limited memory, MMU-less processors, timing constraints, restricted power.Consequently, in this project, we will design and implement a dependable embeddedsystem. Firstly, we will analysis the hardware and software characteristics in embeddedsystems to propose a proper fault detection scheme. Notably, the fault detection scheme willidentify both hardware and software faults. Then, we will design a fault isolation scheme toprevent other irrelevant modules/components also to be infected. Furthermore, we will alsopropose scheme to realize the incremental checkpoint, asynchronous checkpoint, andcopy-on-write checkpoint on MMU-less and memory shortage embedded systems. Finally, wewill also investigate the power issue and propose a low-power fault tolerance scheme. We willdesign and implement our system on the ARM-based Linux operating systems. By our project,we will increase the dependability of embedded systems and thus provide a reliable executionenvironment for embedded users.

近幾年來，由於軟硬體能力的進步、省電機制的演進以及SoC (System-on-Chip)技術的實現，促進了嵌入式系統的蓬勃發展。於是，環顧生活的週遭，嵌入式系統已經走進人們的生活，在工作、休閒以及娛樂上扮演著重要的角色。但是，當越來越多的應用使用嵌入式系統，越來越多的工作都在嵌入式系統完成時，人們對於嵌入式系統的可靠度要求也將隨之大幅提升。換言之，不可預期的軟體或是硬體錯誤導致使用者的資料遺失或是工作白費將是一件不可被接受的事。雖然之前已經有許多研究探討在個人電腦、伺服器系統或是叢集架構上的高可靠度問題，但是，因為嵌入式系統的一些獨有特色，如果以往提出的高可靠度機制在嵌入式系統上將會面臨效率不佳、甚至可能出錯的窘境。這些特色包含：有限的記憶體與電源、不具備記憶體管理機制的處理器、嵌入式系統上的工作具有timingconstraints…等等。有鑑於此，我們提出了這個研究計畫，設計並且實作一個高可靠度的嵌入式系統。首先，我們將研究嵌入式系統上的軟硬體特色，提出適當的錯誤偵測機制，注意，我們將同時支援硬體和軟體的錯誤偵測機制；其次，我們將設計錯誤隔離的方法，避免錯誤感染到其它程式模組，同時，我們也將研究如何在不具備記憶體管理機制和記憶體不大的嵌入式系統上，實現incremental checkpoint、asynchronous checkpoint 和copy-on-write checkpoint。最後，我們也將探討電源問題，設計一個低電力消耗的高可靠度機制。我們將實作在以ARM 為核心平台的Linux 作業系統上，並藉由修改作業系統達到高可靠度的目標。經由本研究計畫，我們將大幅提升嵌入式系統的可靠性，提供使用者一個可以信賴的嵌入式系統使用環境。