Tumbler:在SVR4/MP上提供檢查點與重啟始的功能

Abstract

在最近幾年, 執行時間長且包含大量計算的應用程式有越來越多的趨勢, 而這些應用程式在執行時可能會被某些外在因素所中斷, 譬如系統發生錯 誤, 定期系統維修或是當機, 為了避免這些情形所造成時間與計算資源的 浪費, 我們設計而且製作了一個檢查點與重啟始的工具, 叫作 Tumbler. 在製作檢查點與重啟始的功能時, 我們利用了UNIX SVR4/MP所提供的行程 檔案系統, 這是一個用來存取行程影像和控制行程執行的檔案系統介面, 利用行程檔案系統可減少對系統核心的修改; 本篇論文即是藉著擴增行程 檔案系統以及修改少許系統核心程式, 來完成檢查點與重啟始的功能 , 並使其具有使用者通透性. In recent years, more and more long-running, computation- intensive application programs are running on computing systems. These programs may be interrupted by many events, such as system failures and preventive maintenance. In order to reduce the loss of work in progress resulted from those events, we design and implement a transparent checkpoint and restart facility called Tumbler. We develop this facility by using the features of the process file system which provides access to the image of each active process in the system. Therefore, by expanding the process file system and slightly modifying the kernel, we have implemented a complete checkpoint and restart facility in the user level.