隨著空氣污染日益的嚴重，污染程度已經開始為害到人體的健康，人們開始重視空污因子的即時濃度變化量並著手對空污因子作監控與紀錄分析。由於不斷的從空污監測站收集資料，資料庫儲存的資料量越來越大，舊有的關聯式資料庫已經無法處理過於龐大的資料，為了保持監控的流暢，必須刪除未整理過的歷史資料。然而，在作空污分析時，歷史資料是一個非常重要的指標。因此在不變動原本資料庫架構的情況下，如何將大部分的資料庫資料轉存到分散式資料庫變成一個相當重要的事情。為了實現此目標，我們空氣污染監測系統結合Hadoop叢集將舊有的資料庫資料轉移至分散式資料庫並且備份。這樣不但降低舊有資料庫的負擔且提升了服務的品質。資料轉存必須在不影響空氣污染的即時監控服務前提下進行，在這部分我們特別強調網頁服務不中斷的部分。透過最佳化轉存的模型提高約60\%的效能，並且舊有資料庫受損時備份的資料讓受損的服務可以透過分散式資料庫和MapReduce的方式讓服務可以快速的重新啟動。最後，空氣污染監測系統提供一個關於空氣中污染因子變化量的訊息，供相關單位作為環境檢測和分析的重要依據，讓人們生活在一個更為舒適的環境。 As air pollution becomes more and more serious, pollution hurts human health, people start to pay attention on real time value of air pollution factors monitoring and recording analysis. Because our system need to get data from air pollution monitoring stations usually. Among of data is growing faster, RDB (Relational database) is hard to process huge data. In order to maintain smooth monitoring, we must remove the historical data is not consolidated. But when we analysis air pollution data, historical data is an important target. So, how to dump data to NoSQL without change RDB system become an important things. In order to achieve our goal, this paper proposed an air pollution monitoring system combines Hadoop cluster to dump data from RDB to NoSQL and backup. This will not only reduce the loading of RDB and also keep the service performance. Dump data to NoSQL need to processing without affecting the real time monitoring on air pollution monitoring system. In this part we focus on without interruption web service. Improve 60\% efficiency up, through optimization of dump method and data backup service let service quickly restart by MapReduce and distributed databases when RDB impaired. And through three different types of conversion mode get the best data conversion to be our system. At last, air pollution monitoring service provide message about air pollution factors variation, as an important basis of environment detection and analysis, let people live in a more comfortable environment.
