物件導向資料庫中快取區的設計與實作

Abstract

隨著網際網路的日益發達，主從式、分散式、以及對等式(peer-to-peer)等架構已經普遍被應用在資料庫系統的設計上。在這些架構下，為了減少資料在網路上傳遞所造成的負擔，並且充分發揮系統整體的效能，快取區(Cache)是一個常被使用的方法。 隨著快取區的使用，系統的效能雖然會獲得顯著的提升，但是資料庫中協同控制(Concurrency Control) 的複雜度也隨之提高。在這一篇論文中，我們將整理並且討論快取區一致性的問題(Cache Consistency Problem)，以及相關的演算法。並且設計一個適合ＣＡＣＴＵＳ物件資料庫使用的快取區管理模組與策略。 此外，在論文中會提出一個根據每個客戶端存取頻率而動態調整的新演算法，並且設計出一個實驗平台來針對新演算法的效能做分析，經由實驗證明，我們的演算法在大多數情況下會有較佳的效能表現。

As the Internet progress with the time, architectures such as client-server、distributed and peer-to-peer have been widely used in the design of DBMS. Under these architectures, cache is often used to decrease the overhead of transferring data on the network, and promote the system performance. With the use of cache, the complication of concurrency control comes with the promotion of system performance. In the thesis, we would like to discuss the cache consistency problem, and the cache consistency algorithm. The design and implementation of cache manager in CACTUS is also an important issue. Besides, we present a new cache consistency algorithm which adjust itself by data access pattern in the client. We also implement a simulation model to compare the algorithms.