一個利用分散式資料庫建置高效能網格運算架構的方法

Abstract

格網運算能夠集合許多儲存及計算設備的計算能力，並提供一致的存取方法。對格網架構的發展者而言，目標在一個穩定及安全的方法下，建立一個能夠有效率分配與協調分散及異質性資源的機制。相對的，對格網程式發展者而言，最主要的挑戰為針對網格架構所提供的功能作最適當的運用。通常的情況下，程式設計者需要將問題切割成較小的工作，並要計畫如何對資料作適當的操作及傳輸。若程式執行對記憶體的需求超過一般個人電腦的能力，這樣的分割勢必是需要的，但是相對的會讓程式本身更為複雜。本文論述了藉由資料庫特有的技術，能夠減輕程式設計者在開發程式時對於記憶體管理的負擔。簡單而言，我們使用個別的關連式資料庫當作計算節點，利用資料庫的儲存及計算能力，將各節點串連組合成一個分散式計算平台。除此之外，我們定義了一個通用的資料表綱要，能夠儲存複雜的資料結構，並能夠有彈性地轉換成其他易於計算的表格狀資料結構。我們認為結合這些想法，能夠建構一個有效簡化格網應用程式發展的平台，有助於格網運算的發展。

Grid Computing promises to provide uniform, non-expensive access to computing power through aggregation and utilization of potentially unlimited number of storage and computing devices. For Grid infrastructure developers, this goal amounts to creating effective mechanisms that can allocate and coordinate distributed, heterogeneous resources in a robust and secure manner. For Grid application developers, on the other hand, the main challenge is to make best use of the facilities provided by the infrastructure. Typically, a developer needs to divide a problem into smaller pieces, and plan for appropriate data manipulation and transfer among them. Such divide-and-conquer effort is essential when required memory space is beyond the capabilities of individual machines, but complicated when the infrastructure provides only low-level facilities. This thesis describes database-specific techniques that can relieve developers from complicated memory management. Simply speaking, we use individual relational databases as computational nodes for their storage and computation capabilities, and connect them together into a distributed computing platform. In addition, we define a generic schema capable of storing complex data structures, and mechanisms that allow flexible translation between the schema and other computation-friendly tabular structures. We argue that together these constructs form an attractive platform that can greatly simplify Grid application development, thus contribute to the general Grid Computing community in a useful way.