網路色情的氾濫對青少年造成的負面影響日益受到重視。即使專家學者一再強調家長陪同學齡兒童上網的重要性，青少年瀏覽色情網站的比例依舊居高不下。網頁過濾技術因此成為解決這個問題的選項之一。色情網頁的過濾基於高計算成本的網路應用層內容分析，雲端計算的興起提供了降低計算成本的可能性。然而，基於雲端的過濾架構造成不可忽視的傳輸成本。此外雲端過濾的實現亦需克服相容性的問題，以及對不同的佈署環境及使用者提供合適服務的能力。本論文藉由分析可能的雲端過濾佈署方式、網路設備能力、以及網頁流量的組成，提出適合基於公有雲提供過濾服務的網頁過濾策略。此過濾策略藉由於雲端處理TCP會議（Session）降低設備複雜度，並以中間人（Man-in-the-middle）方式，避免將網頁回應內容由雲端回傳至客戶端，因此能降低架構於雲端的過濾服務的傳輸成本，亦可相容於現有網頁內容提供架構，如通透式代理伺服器（Transparent Proxy）及內容傳遞網路（Content Delivery Network）。基於此過濾策略以及現有網頁過濾及內容分析方法，本論文設計一個基於雲端之網頁內容過濾服務的系統架構並進行實做。此系統可對不同佈署情境採用不同的過濾方法，並對使用者提供不同的過濾效果。分析結果顯示本論文提出的過濾策略可降低於公有雲提供過濾服務的傳輸成本，且當阻擋率越低改善的幅度越高。實驗結果驗證與現有網路環境的形容性及效能。本論文藉由設計一基於雲端之網頁內容過濾服務，以期能讓過濾服務更廣泛被佈署，降低色情網頁對青少年的危害。 Pornography on the web brings non-ignorable influence on minors. Moreover, according to statistics, the problem is hard to solve by simply asking parents and teachers to watch their children go on line. Web filtering, on the other hand, is a potential choice to mitigate the problem. The web filter for pornography relies on expensive application layer analysis, and the rise of Cloud Computing provides a direction for proposing a Cloud-based web filtering architecture to cut the cost. Nevertheless, the architecture of web filtering on Cloud causes significant delivery overhead. Besides, the compatibility problems, different deployed scenarios and requirements of users are also needed to consider for achieving web filtering on Cloud. The web filtering strategies for public Cloud scenarios are proposed in the thesis. The strategies maintain TCP session on Cloud to reduce the complexity of network equipment. It uses the “Man-in-the-middle” way instead of TCP hijacking, thus reduces the delivery overhead on Cloud, and is compatible to current web content providing services, such as transparent proxy and content delivery network. Moreover, system design of Cloud-based web filtering service is then proposed based on the proposed strategies. The system offers a flexible range of deployment options, and can provide different service level for costumers. Analysis shows that the proposed web filtering strategies can reduce delivery overhead in public Cloud scenarios, and the improve rate is inversely proportional to the blocking rate. Experiments show that the proposed system is compatible to the current network environment.
