在資料網格環境中，資料集被重複製為複本且分送到多重的站台。由於資料集的檔案通常很大，如何有效率的存取及傳輸成為重大的課題。因此先前有學者發展出協同配置的架構，使得同時從多重站台平行下載資料變成可能，目前發展出數種協同配置法用來解決傳輸時本地端與伺服端網路傳輸率不斷變動的問題。例如將欲傳輸的檔案切割成數個均等的檔案大小，或是將檔案動態的切割置於工作佇列，透過連線品質較佳者傳送佇列中末完成傳輸的檔案區塊，來解決網路變動的問題。無論各個下載連線的效率為何，當伺服端傳送最後一個檔案區塊時，發生速度快的伺服器閒置的等待最慢的伺服器完成最後一個檔案區塊的傳送，或是因為不同伺服器傳送相同的檔案區塊，造成網路頻寬資源的浪費，因此，若能在一群候選伺服器中找到最大的頻寬資源，有效分配工作減少各伺服器間完成傳輸時間的差異，將成為最重要的工作。近年，在全球各地的學者先後提出頗具新意的資料網格平行檔案傳輸策略；本研究中，匯集8種各具代表性的平行檔案傳輸方法，融合各方法的優點改善其缺點，採用TCP頻寬估計模型與突發模式等新策略，藉此強化預測性遞迴調整的協同配置法，進一步提高大量資料集於資料網格中的傳輸效能。我們的方法能有效地找出一群快速伺服器並分配較多的工作量提高其資源利用率，動態計算出檔案切割量，有效減少各伺服器間的相互等待時間。藉由各項實驗證明其傳輸的高效能，並具有網路自適應性與高度容錯性，有效因應不同的網格環境。 Data grid enable the sharing, selection, and connection of a wide variety of geographically distributed computational and storage resources for content that the large-scale data-intensive application needs, such as high-energy physics, bioinformatics, and virtual astrophysical observatories. Data grid consists of scattered computing and storage resources located in different regions yet accessible to users. Co-allocation architectures can be used to enable parallel transfers of data file from multiple replicas in data grids which are stored at different grid sites. Schemes based on co-allocation models have been proposed and used to exploit the different transfer rates among various client-server network links and to adapt to dynamic rate fluctuations by dividing data into fragments. These schemes show that the more fragments used the more performance. In fact, some schemes can be applied to specific situations; however, most situations are not common actually. For example, how many blocks in a data set should be cut? For this issue, we proposed the anticipative recursively adjusting mechanism (ARAM) in a previous research work. Its best feature is performance tuning through alpha value adjustment. It relies on special features to adapt to various network situations in data grid environments. In this thesis, the TCP Bandwidth Estimation Model (TCPBEM) is used to evaluate dynamic link states by detecting TCP throughputs and packet lost rates between grid nodes. We integrated the model into ARAM, calling the result the anticipative recursively adjusting mechanism plus (ARAM+); it can be more reliable and reasonable than its predecessor. We also designed a Burst Mode (BM) that increases ARAM+ transfer rates. This approach not only adapts to the worst network links, but also speeds up overall performance.
