網格計算服務(grid-computing service)是分散式計算架構的應用形式之一,網格計算服務具有標準化的通訊協定與安全的認證機制,能讓內部的資源管理系統(Resource Management System, RMS)透過網際網路,將散落在各區域空間的線上(on-line)計算資源連結整合,形成一部具有強大且高速運算能力的虛擬運算群集,提供使用者大量且高複雜的計算服務,對於未來資訊社會的發展,網格計算服務將扮演一項重要利器。網格計算服務原始概念是一種公共財(Public goods)的形式,服務對象為大量複雜計算需求的用戶群。但觀察網格計算服務的發展歷史即可得知,其隱含著兩項弊病,其一為常見的搭便車問題(Free Rider Problem),即資源欲共用但不欲分享心態,導致網格計算服務供應商難以永續經營;其二,不良的資源指派方式會干擾網格計算的整體服務品質(Quality of Service, QoS),造成消費者對網格計算服務的信任障礙。現今多數資源規劃模型都無法反應這兩項現實問題。 本研究提出經濟資源佈局配置模型,其包含虛擬支付評估(virtual payment assessment)概念,根據使用者付費(pay-per-use)的原則,將網格計算服務概念轉化成需付費的準公共財(Quasi-public goods),並導入可靠性服務層級協議(Service Level Agreement, SLA),確保使用者對網格計算服務的信任感。此經濟資源佈局配置模型,目的在於使資源租用時間成本最小化,並滿足特定的服務可靠度限制,故此類資源配置規劃問題屬於組合最佳化問題(Combinatorial Optimization Problems, COP),具有NP-complete特性。本研究接續提出柏拉圖群集(Pareto-set cluster)概念,搭配先進的柔性演算法(Soft Computing),針對網格計算資源的租用成本與服務時間,進行穩定且具有經濟效益的資源佈局配置規劃,以提供網格計算服務資源管理系統一套可選擇的資源配置決策參考集合。立基於可靠性網格計算服務架構下之經濟資源佈局配置模型,或可裨利網格計算服務永續運作與未來發展。
A grid-computing service, united by numerous distributed and heterogeneous resources, supplies various advanced and cumbersome problems in terms of high-performance computing. Based on reciprocal transactions of a Grid Bank [24], this dissertation presents an economics-driven resource allocation model to determine the grid-computing service reliability for the service level agreement and to evaluate grid-computing service expenditure for the free rider problem. In terms of the probability of completing the task, this paper initially converts the grid system into a multi-state unreliable network and then estimates the service reliability in a tree topology using a simulation method (i.e., cellular automata Monte-Carlo simulation, CA-MCS) and in star topology using an analytic method (i.e., universal generating function methodology, UGFM). This paper also proposes virtual payment assessment to appraise the rental-time cost for each resource’s contribution. In order to determine the best resource allocation for a given rental-time cost and guaranteed reliability, this paper presents two revised meta-heuristic algorithms (i.e., GA and PSO), wherein Elite-selected and Reborn (ER) mechanisms improve the optimization effectiveness and a Pareto-set Cluster evolves the Pareto frontier. Accordingly, the economics-driven resource model saves total rental-time cost and ensures that the grid-computing service is reliable.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。