分散式程式設計的偵錯方法

Abstract

對很多程式設計者而言，發展分散式的程式 (distributed programs) 是 比循序程式 (sequential programs) 來的費力。現在，利用循序程式的 偵錯方法所開發出來的偵錯工具很多，但是分散式程式的偵錯方法大部份 只在理論階段作研究，實際應用不多。所以，本論文希望能夠提供一種偵 錯方法，幫助使用者實際去偵測出錯誤來。在這一篇論文裡，我們所要解 決的問題是 LS1* S2*...*LSn*CS12*CS21 *...*CSn-1,n*CSn,n-1 != K 。其中，LSi 表示行程 (process) Pi 的狀態、CSi,j 表示 Pi 到 Pj 管 道(channel) 的狀態、K 表示一個常數而 * 必須符合結合律與消去律的 運算元，例如：+ 加法運算、* 乘法運算等等。所須要的時間複雜度只 要 O(m)，其中 m 表示所要傳送的總訊息個數。 It is more difficult to develop distributed programs than sequential ones for many programmers. Nowadays, debugging tools which exploiting the debugging method of sequential programs are numerous. However, most debugging methods of distributed programs focus on research of theory, not on debugging software. In this thesis, we intend to supply a distributed debugging method to help users to debug. In this thesis, we want to solve a problem domain is LS1*LS2*...*LSn* CS12*...* CS21...*CSn-1,n*CSn,n-1 != K, where LSi presents process Pi's local state、CSi,j presents Pi to Pj channel's state、K is a constant and * must satisfy the operator of associative and eliminative laws, e.g. the addition operator +、the multiplate operator *,etc. The time complexity of this debugging method is only O(m), m is the total number of messages which deliver between processes.