經由格網計算推動金融服務競爭力

Abstract

金融服務近幾年來面臨快速的資訊科技發展所帶來的的挑戰，在金融服務業務上，如交易、避險與風險管理等等，經常需要同時兼顧快速反應與大量費時的計算模擬，其競爭力的關鍵經常僅在於有效資訊提供與市場交易的秒差優勢。要達成如此的競爭優勢，需要的不僅是購買高效能的計算與儲存設施，還需要包括客製化的商業邏輯與重要(Mission Critical)分析功能，甚至需要將分析的演算法調整到效能能達到毫秒之等級。在客製化的商業邏輯部分目前流行的架構以網頁服務(Web service)與服務導向架構(Service Oriented Architecture, SOA)為主，XML為基本資料交換架構，始能極易與不同應用資訊平台間進行資訊整合、融合與視覺化。在重要分析功能方面，形成一個競爭瓶頸，可以高效能計算解決，但高速計算資源極為昂貴且其計算循環(Compute Cycle)有限。隨著高速光纖的興起與普及，使利用分散式的大型計算資源串連來擴充計算的循環變成更為有效，格網計算(Grid Computing)是近年來最重要的工具。本研究即以格網計算來加速與擴充計算循環，其中探討現有較常用且需要大型計算資源的演算法其在格網計算上的加速方法與效益；並且以蒙地卡羅模擬法(Monte Carlo Simulation)之選擇權評價與風險管理值為範例，分別以大型叢集電腦、無碟遠距啟動形成的小型計算格網為比較基準，與超大型的雲端式計算格網為測試平台，比較分析其間差異。最後，與具有前端客制化商業邏輯能力且能動態串流大型時間序列資料的中介軟體RBNB連結，並依此對格網計算對於金融服務競爭力的提升提出建議。

Securities trading is one of the few business activities where a few seconds processing delay can cost a company big fortune. The growing competitive in the market exacerbates the situation and pushes further towards instantaneous trading even in split second. The key lies on the performance of the underlying information system. Following the computing evolution in financial services, it was a centralized process to begin with and gradually decentralized into a distribution of actual application logic across service networks. Financial services have tradition of doing most of its heavy lifting financial analysis in overnight batch cycles. However, in securities trading it cannot satisfy the need due to its ad hoc nature and requirement of immediate response. A new computing paradigm, Grid computing, aiming at virtualizing scale-up distributed computing resources, is well suited to the challenge posed by the capital markets practices. It is also no doubt that Grid computing has been gaining popularity to serve as a production environment for finance services in this couple of years. In this study the core computing competence for financial services is examined. How the underlying algorithm for financial analysis can take advantage of Grids is presented. One of the most popular practiced algorithms Monte Carlo Simulation is used in our cases study for option pricing and risk management. The various grid platforms are carefully chosen to demonstrate the performance issue for financial services, which include small diskless remote boot Linux (DBRL) clusters, large scale computer cluster, and a densely distributed at-home style PC grid, which resembles the clouding computing. The Service Oriented Architecture (SOA) based on Ring Buffer Network Bus (RBNB) is also used with web services to demonstrate the streaming data with grids.