2015-01-012024-05-18https://scholars.lib.ntu.edu.tw/handle/123456789/698493摘要：本計畫的主題是研究多媒體雲端技術處理的所需的系統架構。現今多數的雲端技術，大多是為一般性的應用所研發，因此，需要犧牲效能以兼顧普遍性與一般性，但多媒體資料通常需要大量的儲存空間與運算，這些效能的犧牲可能會造成無法接受的服務等待時間。因此針對多媒體資訊處理的需求，提出新的系統架構設計是很重要的，例如：可靠且有效率的多媒體服務甚至是雲端儲存空間的隱私權議題等等。類似這樣的典範轉移(paradigm shift)現象，也曾出現在過去設計處理器的歷史中。早期的中央處理器(CPU)也是為一般性的計算所設計，但是用於多媒體應用程式效能便很差，因此英特爾(Intel)公司研發MMX跟SSE這兩種擴充方式，但仍然不夠有效率，因此而有了圖形處理器(GPU)的誕生。我們相信在雲端運算的技術發展上也將有類似的趨勢，當雲端技術趨近成熟，下一步便是有效率且可靠的多媒體雲，同樣地，這可能是一般雲端技術的擴充，也可能像圖形處理器一樣是一個專門的多媒體雲。這個計畫的目的便是去探索將多媒體應用於雲端的各種可能。我們將特別研究(1)高度平行化的視訊編碼、(2)保有隱私權的多媒體搜尋、(3)支援3D渲染的雲端架構這三項主題。針對上述的多媒體應用，我們將分析他們的特性並提出對應的雲端架構設計，研究的主題包括硬體架構(如混和不同處理器的雲端及運算節點設定)與軟體架構(如編碼方法、保有隱私權的索引方式、新的程式範例、負載平衡演算法等等)。我們相信這些技術能幫助產業界面對這些典範轉移，並在多媒體雲端應用上發揮影響力。<br> Abstract: This project studies the architecture and framework for multimedia-aware clouds. Most of today`s cloud technologies are developed for generic applications. Thus, they often have to trade efficiency for generality. Unfortunately, because multimedia applications are often involved with sheer volume of data and tremendous amount of computation, such inefficiency could lead to unacceptable service latency. Thus, it is important to propose new architecture and framework designs for multimedia-aware clouds to provide reliable and efficient multimedia services and further consider the vital privacy issues for data storage in the cloud. A similar paradigm shift had happened in the history of processor design. In early days, CPUs were designed for executing generic operations. However, such generic architectures were inefficient for many multimedia applications. Thus, by analyzing characteristics of multimedia applications, Intel took efforts to develop multimedia extensions, MMX and SSE, specifically designed for speeding up multimedia applications. When extensions to generic CPUs were still insufficient, dedicated processors for executing multimedia applications had been developed, such as nVidia`s GeForce and Intel`s Larrabee processors. We believe that there will be a similar trend for cloud computing. When the cloud technologies become mature, architecture designs for efficient and reliable multimedia clouds will be the next step. Similarly, they could be extensions to generic clouds or even dedicated clouds for particular applications. This project aims to explore possibilities along this direction for several important multimedia applications. In particular, we will study (1) highly parallel video coding, (2) privacy preserving media search, and (3) 3D rendering clouds. For each of these applications, we will analyze its characteristics and propose cloud architecture designs. The research topics involve both hardware architecture (such as hybrid clouds with heterogeneous processors and configurations of computing nodes) and software framework (such as coding methods, privacy-preserving indexing, novel programming paradigms, load balance algorithms, etc.). We believe the developed technologies will help the industry respond to the paradigm shift and leverage the great opportunities in multimedia-aware cloud applications.雲端多媒體平行化視訊壓縮隱私權保護3D渲染CloudMultimediaParallel Video CodingPrivacy Preserving3D Rendering