雙原址法容錯系統的效能分析

Abstract

分散式處理與軟體容錯技術的成熟，使得組合工作站為高效能低成本的電 腦系統，成為代替購買具容錯能力大型主機的另一種選擇。而藉著網路系 統中多重備援(backup)節點(node) 與即時檢核(check- points)，使主要 節點電腦得以達到容錯的功能，此即原址法(primary site approach)之 理論基礎。本論文即是在這樣的背景下，研究雙原址容錯之方法，並予以 適當的功能分類，個別比較其與單原址系統效能之差異；並進而改良檢核 的排程與作法，以有效的增進可用度(ava- ilability)、可靠度（ reliability）等系統效能 。從雙原址多備援模式的價值與貢獻，理論上 ，可以做為進一步研究推展多原址多備援系統的根據；實務上，我們則以 金融資訊系統為例，說明使用本文方法的分散式容錯系統較優於現行集中 式系統之處。 Distributed processing using low-cost workstations has become an attractive alternative for high-performance fault- tolerance computing systems. The primary-site approach which utilizes a primary site and multiple backup nodes under a distributed processing environment can lead to a computing system with effective fault tolerance capability. This thesis extended the single primary site model to the dual primary-sites model. Three system types were classified and investigated; new formula regarding availability and system throughput were obtained. It also showed that effective scheduling of checkpointing activities improved system availability. A banking information system were outlined to illustrate applications and advantages of the dual primary- sites approach.