在非精確運算及多核心處理器平台下的即時模式轉換可排程性分析

Abstract

在即時系統中，如何讓工作在時限內完成，是過去常見的研究目 標，但現今的系統會需要因應不同的需求轉換成不同的操作模式，如 在飛機系統中包含陸地、飛行等模式；如先進駕駛輔助系統中當駕駛 在高速公路或一般市區將使用不同模式，而如何能正確且有效地切換 模式是本研究目標。 並且，在現今軟體中非精確運算在影像和AI等領域被廣泛地運用， 而在本篇論文中考慮此運算模型，藉由此模型，系統可以更有效地利 用資源進行排程，並且可以縮短任務的反應時間，增進使用者體驗， 而本研究針對傳統運算模型及非精確運算模型分別提出不同的解決方 案。 為了達成上述目標，我們延伸傳統地可排程性分析的框架，提出在 多模式系統中出新的可排程性分析，在最後的實驗部分，我們藉由模 擬的方式與其他研究的分析方法進行比較，在結果中，我們的方法可 以提升15至30%的可排程工作量，並在非精確運算模型中，能夠確實 地增加可完成的工作數。

Most studies of real-time scheduling have focused on meeting deadlines of a given task set. However, many applications need to change operating modes depending on system states. For example, the flight control system of the aircraft depends on different functionalities and has Ground, Flight and Flare modes. Advanced Driver Assistance Systems can have Highway, Country Road, and City modes depending on different environments. Our motivations are that how to correctly and effectively change modes. Additionally, the imprecise computational model is widely applied to the image processing, AI system, etc. In this work, we consider this model which makes the system more flexible to schedule. Furthermore, this model can reduce the average response time, and improve User Experience (UX). We will focus on the traditional workload model and imprecise computational model to propose solutions. In order to address this problem, we extend the traditional schedulability analysis and develop a new analysis to the multi-mode system. In the experiment, we compare our schedulability analysis with other work by simulation. The result shows that our analysis can increase schedulability by 15 to 30%. Moreover, our work can actually increase the completed tasks during mode change on the imprecise computational model.