在 Real-Time Mach 即時作業系統上製作 Imprecise Computation Model排序策略

Abstract

發展像 VOD 的多媒體應用程式是未來的趨勢。為了支援此類的應用程式, 一個支援多種即時排序策略的即時作業系統是需要的。雖然 Real-Time Mach 即時作業系統已經提供多種即時排序策略，但是當系統的負荷過重 或是有些程式是 I/O 導向時，這些策略仍無法保証所有執行中的即時程 式可以在它們的期限之前執行完畢。此篇論文中，我們設計以及製作擁有 周期性特性的 imprecise computation model 排序策略能在系統負荷過 重時支援預約服務，使得每一個服務要求都能滿足。我們修改 Real- Time Mach 即時作業系統來製作此排序策略。我們也設計以及製做一個簡 單的視訊伺服器來驗證此策略的能力。我們的實驗顯示當我們每六張圖片 中跳過一張時，我們可以省下將近百分之八的網路流量以及客戶端百分之 十二到十四的處理時間。我們的實驗也顯示我們可以根據系統的負荷動態 地跳掉一些圖片。 Developing multimedia applications such as video-on-demand is a future trend. In order to support this kind of applications, a real-time operating system supporting various real-time scheduling policies is needed. Real-Time Mach supports various real-time scheduling policies, but these policies can not guarantee that all of running threads can meet their deadlines when some running threads are I/O-bound and the system load is heavy. In this thesis, we design and implement imprecise computations model with periodic property to support service reservation that fulfill deadlines of requests under heavy system load. We modify Real-Time Mach to implement this new policy. We also design and implement a simple video server to demonstrate this policy. We show that we can save about 8% of network bandwidth and 12% to 14% of processing time for client when we skip one frame per six frames statically. We also show that we can dyamically skip the frame depending on system load.