適應性租約: 一個在WWW快取伺服器上的同步機制

Abstract

WWW的興起為網路流量帶來驚人的成長，造成網路的壅塞與伺服器負荷的增加。快取（cache）在網際網路上被廣為使用，它可以減低網路流量，減少存取延遲與減輕伺服器的負荷，然而要保證快取內資料的正確性，快取同步技術隨之興起。目前常見的快取同步之研究有：適應性存活時間（Adaptive TTL）、每讀詢問（Poll-Each-Read）、伺服器通知（Callback）與租約（Lease）等方法。如何決定物件在快取伺服器中的存活時間是一個重要的問題，我們提出了一種適應性租約（Adaptive Lease）的方式來決定物件的存活時間，它不但可以達到強同步的效果，還可以減少傳統方法在記憶體以及頻寬使用上的浪費。此外，我們也利用實驗來瞭解快取伺服器的容量大小和系統效能之間的關係，而結果顯示兩者之間並無明顯的關係。最後我們用數學的模式來分析在相同的租約存活時間下所需要傳送的訊息（messages）數量，並且用模擬方式來驗證理論值的正確性。

Caching objects can reduce network traffic, client response time and server memory usage. However, to make sure that cached copy is up-to-date or not, cache consistency mechanism should be maintained. There are several well-known approaches to maintain cache consistency: Adaptive TTL, Poll-Each-Read, Callback and Lease, but how to choose the object’s lifetime is a trade-off problem. In this thesis, we propose an Adaptive Lease approach to determine caching object’s lifetime. This approach can not only achieve strong consistency, but also reduce server memory usage and control message transmission. Furthermore, we find out the relationship between cache size and system performance. The simulation result shows their relationship is very slightly. Finally, we estimate the control messages transmitted for Constant Lease approach by mathematical analysis and use trace-based simulation to verify our estimation.