一個非同步低耦合度之動態格網運算系統

Abstract

諸如基因排序以及蛋白質解析等複雜的解碼工作需要大量的運算量以及繁複眾多的執行步驟，業界多半利用格網運算(Grid Computing)的方式將繁重的運算工作交予後端的分散式叢集電腦群來快速完成，這一類的格網系統通常仰賴著昂貴的硬體設備或是特殊、特定的軟體，使得格網運算這個名詞數年以來一直狹隘地隸屬於高速科學計算的領域。縱使在科學領域已被廣泛使用，昂貴的整體擁有成本(Total Cost of Ownership)讓一般使用者或甚至中小型企業對於採用格網概念望之卻步。龐大的潛在電腦運算資源依然未能有效被利用。使用者持續尋找著更多的運算資源來解決他們的問題。在這些挑戰之上，格網本身尚暴露著技術層面的瑕疵。業界的方案無法有效解決諸如此類單一切入點故障(Single-point of Failure)的架構性問題、動態擴充格網體積的延展性問題(Scalability)、或是支援跨平台特性的普及性問題。 本篇論文提出一個低成本的純軟體跨平台格網方案，有效解決單一切入點故障、動態擴充格網體積等問題，並提供簡化格網應用程式開發的工具組以及程式設計介面(API)，大幅縮短研發人員粹取企業內部運算資源的時間，將生產力最佳化。研發人員將能專注於設計與開發，而不再需要週而復始地在企業中辛苦獵取閒置但是隱形的運算資源。

Complex jobs such as bio-genetic sequencing and protein modeling requires massive quantity of calculation and execution procedures. Today, industry applies Grid Computing technologies to delegate the intensive computational work to a farm of cluster computers in order to accelerate computing speed. This category of grid computing rely on sumptuous hardware or distinctive, specific software, thus restraining grid computing to constricted domains such as high-speed scientific computation. Despite the widespread acceptance of grid concept, high TCO(Total Cost of Ownership) intimidated the general public or even SMEs(Small-Medium Enterprises) from adopting grid technologies. Vast amount of potential computing capacity still remains untapped. Users are continually searching for more computing resources to assist solve problems. On top of these challenges, Grid itself suffers certain technical imperfections. Commercial solutions are incapable of solving single-point-of-failure issues, incapable of dynamically expanding the volume of grid network and is certainly having a difficult time migrating grid infrastructure to a universe of different electronic devices existing today. This research proposes a low-cost, pure software-based, cross-platform grid framework, eliminating the mishap of single-point of failure, allowing dynamic grid expansion. The framework also provides utility tools and Application Programming Interfaces(APIs) that simplifies the process of grid application development, thus optimizes overall productivity. Developers must focus on design and development rather than hunting for resources hidden within the enterprise.