一個有效率之網路類別載入機制

Abstract

目前的網路類別載入機制大部份都是透過客戶端的瀏覽器來實行，而最廣泛的應用就是Java Applet。其必須要先藉由瀏覽器把程式執行所需的類別物件透過網路抓回來，接著才能載入執行。為了要加快載入的速度，瀏覽器使用了快取的技巧，但快取內資料能否有效同步是我們常關切的問題。另外，在載入程式執行之前，也得想辦法去減少網路緩衝延遲的等待時間。這些都是網路類別載入效率上的重要考量因素。 在本篇論文中，我們提出了一個有效率的網路類別載入機制，可以有效改善上述的問題。藉由使用這套機制，可以將現有方式的一些長處結合起來，像是rsync同步演算法、依索求之類別載入方式和壓縮檔形式之類別載入方式等等。我們也提出了測試上的數據，顯示出這套機制在多數的網路情況下，對於改善快取資料同步和網路緩衝延遲問題，都能獲得好的效率。

Most of the current network class loading mechanisms were accomplished through the web browser in the client side. One wide-ranging example is the Java applet. Before such programs can be loaded and executed, we have to fetch all needed classes through the network by the web browser. In order to speedup loading processes, the web browser uses some caching techniques. The first thing we usually care that is how effective and efficient the data synchronization can do. Additionally, we also have to minimize the network latency delay in loading processes. These two are the most important performance issues in network class loading. In this thesis, we propose an efficient network class loading mechanism, which can efficiently improve the issue we just mentioned. By using this mechanism, it is able to combine the strengths of the existing ones, such as the rsync synchronization algorithm, on-demand class loading strategies, archive-based class loading strategies, etc. We also present some experimental results that show how it can perform well in a variety of network conditions.