近年來大數據的議題在很多領域都被討論，為了有效的處理及分析如此龐大的資料，資料倉儲是一個重要的關鍵，已經有多研究表示Column-based Data Warehouse比傳統的Row-based Data Warehouse有更好的表現，所以Column-based Data Warehouse 成為了現在很多資料庫系統所使用的資料儲存架構，如SAP HANA。除此之外，在資料倉儲系統中，使用者進行查詢會產生大量的成本，View Selection問題決定哪些查詢的結果資料要預先儲存在資料倉儲之中，View Maintenance Policy則決定什麼時候要去更新這些儲存在資料倉儲內的資料。 在本研究中，我們建立了一個新的MVPP模型能夠表現出Column-based Data Warehouse中的查詢過程，並藉由修改Liu等人(2008)所提出的成本模型，建立了可以考慮到隨機性的查詢及資料更新在系統查詢回覆時間的限制之下。為了符合現實的情況，模型假設隨機的查詢到達率符合普瓦松分配，使用M/G/1模型來限制系統查詢的回覆時間，並在AMPL/MINOS的環境下建立數學模型，計算出相關的成本以及決策。除此之外，我們設計數個不同的案例，來評估及比較Column-based Data Warehouse與傳統資料儲存架構Row-based Data Warehouse的差異。 ;In recent years, the issue of Big Data has been discussed in many areas. In order to analyze such a huge amount of information, the data warehouse is an important key. Many researches show that the performance of column-based data warehouse is better than the row-based data warehouse. The column-based data warehouse becomes popular storage architecture used by database systems such as SAP HANA. In the data warehouse, the view selection problem is to select a set of views to be materialized, when minimizing the total of query processing cost and view maintenance cost. The update policy is to decide when to refresh the data in a data warehouse. In this research, we propose a new multiple view processing plan model which can present the operations in the column-based data warehouse. Modify the cost model in Liu et al. (2008) and propose a cost model which can consider the appearance of the stochastic query arrival and stochastic update, which contained a specified response time limit. For model according to the reality, we incorporate stochastic query into the model follows Poisson process and the constraints of system response time is formulated by an M/G/1 model. We use AMPL/MINOS to solve and implement the mathematical model. In addition, we also design several cases to evaluate the difference in view selection and total cost between the Column-based data warehouse and Row-based Data Warehouse.
