近年來巡天計畫觀測技術的進步，為了探索小行星軌跡及觀察瞬變天文事件等天文現象，所儲存的天文數據資料量已達到了PB等級，利用巨量資料分析已經成為天文研究的趨勢。在本研究中的三個天文應用模組，基於位置的天體查詢(DSSIS)；變星天文事件序列模式管理(DASPQS)；小行星軌跡延伸(DADS)，是依照不同天文資料特性與關鍵的索引結構，透過分散式運算，輔以雲端環境之儲存系統，來加快對觀測資料儲存與處理的速度，供天文學家可以進行後續的分析與維護，本研究透過結合分散式運算的方法所執行的時間比在單機上減少了98％。然而使用索引結構會影響資料存取及運算的比例，針對不同的應用對資料儲存與計算行為皆有不同資源分配需求，且在分散式節點進行運算時，會有I/O負載問題的情況產生，因此，必須考慮索引結構與資料結構並配合資料區域性才會有突顯的效果，本研究運用OpenStack與Hadoop作為雲端運算平台，透過不同雲端環境的參數，解析對於運算效率的影響程度，進行資料區域性的策略探討。實驗結果表明，所提出的資料區域性策略Hybrid locality比傳統方法（Node locality）所執行運算的速度提升了13-50％，因此本研究所提出的適用性策略，能夠有效地解決分散式雲端平台上的I/O負載問題，進而提升運算的效能。;Recent advances in astronomical observation technology have led to the collection of Petabytes of data. Such massive datasets warrant a big-data approach to analysis. Numerous recent projects have involved the construction of advanced telescopes and their use to survey the sky, to obtain data on asteroid movements and transient astronomical events. Astronomical researchers use various methods to analyze the observational datasets. This study is concerned with three data access models in the field of astronomy, which are location-based queries about celestial objects, management of sequential pattern of event of variable stars and asteroid track linkages. Astronomers’ research can be accelerated using cloud computing and data indexing technologies. We provide appropriate distributed systems – DSSIS, DASPQS and DADS to deal with corresponding problems. Each methodology has associated index structure or intermediate result. The best of our experimental results reveals that the distributed approach reduced the execution time by 98% below that required when running on a single host. However, index usage substantially affects data access behavior, and especially the ratio between the storage and computation. Various applications have different costs in terms of storage space and processing time, so that it causes a significant number of non-local tasks which data are accessed across nodes through networks. Accordingly, this research considers data locality on a cloud platform (Hadoop on OpenStack) using different approaches of critical index on astronomical applications. The performance of applications is discussed with reference to specific parameters concerning cloud environments, and a strategy for data locality is proposed. Experimental results herein demonstrate that the proposed strategy, Hybrid locality reduced the execution time of these applications by up to 13-50% from that achieved using a conventional method (Node locality.) Hence, our proposed strategies for data locality provide a great performance improvement.
