這個計畫的宗旨是提出一個在動態網格上的通用型的分散式最大可能性估算 (Maximum Likelihood Evaluation, MLE) 計算架構，讓使用者可以輕易的去發展一個適用於動態網格的MLE 應用程式。最大可能性估算是科學界以及工程界所常用來做資料分析的統計學方法，但是隨著近代某些科學問題所要探勘的資料大量增加，使用單一處理器的電腦來處理MLE 計算問題已明顯不敷使用。舉例來說，下一代高能物理研究的「搜尋消失的重子」計畫就需要平行處理的方式來加速MLE 演算法的分析速度，也因此我們可以使用網格計算的技術來幫助這類型的研究。然而撰寫一個在計算網格上執行的分散式MLE 應用程式並不容易，如果再考慮到使用動態負載平衡來改善程式執行效率的話就會使程式研發更加困難。所以我們提出了一個可以在動態網格上執行的通用型分散式最大可能性估算計算架構讓人們使用。這個架構有兩個特點是：(1)它提供一個分散式 MLE 計算架構的套件、(2)它提供動態系統重組的功能以達成系統的動態負載平衡。在這個研究中我們所預期的貢獻有：(1)協助並簡化需要MLE 的科學或工程分散式應用程式的發展、(2)提供發展問題導向的分散式計算架構的經驗、(3)提供動態網格上 Peer-to-Peer 負載平衡的效能分析與實作。這個計畫將可以幫助很多需要分散式MLE 演算法來分析大量資料的基礎科學的研究，也因此它間接對國家的發展也有助益。除此之外，我們所訓練的人員將可掌握最新的動態網格技術與許多跨領域的知識如高能物理、天文學、生物資訊等。本計劃所發展的程式碼將會公開讓人使用與下載。 ; In this proposal, we present a generic, distributed maximum likelihood evaluation (MLE) computing framework. MLE is a statistics technique which can be seen in a lot of science and engineering researches. As the rate of data exploration rapidly increases in modern science, using a uniprocessor computing environment to handle MLE computation will become almost impossible for a lot of problems. For example, the missing baryon problem in high-energy physics will demand more parallelism when the next-generation detector is available. Computational grid can provide huge computing power and high-degree parallelism to fulfill the computing demands from those problems. Unfortunately, it is hard to develop a distributed MLE application on a computational grid, in particular when dynamic load balancing is considered to improve application efficiency. Therefore, we propose a generic, distributed MLE computing framework which has two features: (1) a software package to facilitate implementation of distributed MLE applications on a grid environment, and (2) a dynamic system reconfiguration strategy. We expect the following three contributions from this project: (1) it will offer a toolkit to simplify the development process of distributed MLE applications in science and engineering researches; (2) it will give people valuable experiences of developing problem-based, distributed computing frameworks; (3) it will provide efficiency analysis and implementation of a peer-to-peer, dynamic load balancing strategy. The project will benefit fundamental science research and thus indirectly contribute to national development. Furthermore, we will hire several part-time research assistants to execute this project. They will be able to master the latest dynamic grid technologies, and gain knowledge as well from different fields such as particle physics, astronomy, and bioinformatics. ; 研究期間 9708 ~ 9807
