資料網格(Data Grid)使得分散在不同區域上的計算及儲存資源(同質或異質)，可以達到分享、選擇、以及相互的溝通的應用。尤其是需要分析大量且密集資料的科學實驗，諸如高能物理、生物資訊的運用、以及氣象的模擬等，透過應用資料網格的都獲得良好的問題解決方式。在資料網格(Data Grid)環境中，資料集被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(Co-allocation Architecture)，使得同時從多重站台平行下載資料變成可能，且發展出數種協同配置的策略被使用來解決傳輸時本地端與伺服端網路傳輸率會變動的問題。例如將欲傳輸的檔案切割成數個均等的檔案大小，或是將檔案切割成相同大小置於工作佇列，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，前述中各傳輸伺服器每次所被分配到的每一個檔案區塊大小是一樣的，將使傳送最後一個檔案區塊時，發生速度快的伺服器通常需要耗用較多的等侯時間在等侯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，造成網路資源的浪費，因此，如何減少各伺服器間完成傳輸時間的差異，且避免傳送相同檔案區塊所形成的網路資源浪費，將成為重要的工作。在這個研究中，我們提出了一個動態預測調整的方式稱為預測性遞迴調整的協同配置(Anticipative Recursively-Adjusting Mechanism)，來改善資料網格(Data Grid)中資料傳輸的效能。我們的方法有效地減少了快速伺服器與慢速伺服器間資料傳輸完成之閒置時間，也減少了整體資料傳輸之完成時間，且避免發生重覆傳送相同檔案區塊的情況。 More and more applications emphasize analysis of a huge amount of data and depend on the transmission of them. Data Grids 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 grids consist of scattered computing and storage resources located in different countries/regions yet accessible to users. The co-allocation architecture was developed to enable the parallel download of datasets/servers from selected replica servers, and the bandwidth performance is the main factor that affects the internet transfer between the client and the server. In previous works, exist a drawback of idle time or degrade the network performance for transferring the same block. Hence, it is important to reduce the difference of finished time between each selected replica server, and avert the bandwidth traffic congestion from transferring the same block in the link among the servers and clients, and manage changeful network performance during the term of transferring as well. In this thesis, we proposed Anticipative Recursively-Adjusting Mechanism scheme to adjust the workload of each selected replica server, which handles unwarned variant network performances of the selected replica servers. The algorithm is based on finished rate of the previous assigned transfer size to anticipate the bandwidth status on the next section to adjust the workload, and further, to reduce file transfer time in a grid environment. Our approach is useful in unstable gird environments and it not only reduces the wasted idle time from waiting the slowest server but also decreases the completion time for file transfers.
