CAREER: Towards a New Generation of Multimedia Storage Systems

高效的存储组织和访问对于大规模多媒体系统的出现至关重要。多媒体应用，从视频点播(VOD)到医疗远程呈现或大规模科学计算，涉及到受一系列实时限制的大量信息的存储和检索。当前的存储系统不足以满足多媒体系统的需求。事实上，最近在向数千名客户群体提供VOD方面的各种试验已经将存储访问确立为成功的关键障碍。因此，对能够有效支持大规模多媒体系统的新一代高性能存储技术的需求日益增长。这项研究的主要目标是为多媒体存储系统开发新的体系结构和资源管理算法，从而使性能提高几个数量级。这里追求的主要论点是，存储系统可以利用多媒体应用程序的固有的、非随机的访问特性来实现显著的性能改进。这项研究工作将对一般的多媒体系统的研究，特别是对其存储技术的研究有几个重要的贡献。它将寻求表征广泛的多媒体应用程序的存储访问模式，并研究利用这些特征来提高存储和访问效率的技术。拟议的研究将考虑整个存储层次结构，研究整个资源池以及这些资源如何相互作用，并制定将应用程序和工作量的限制和特点纳入资源管理技术的方法。这种介质感知、工作负载感知和应用感知的存储组织和访问机制可以显著提高存储服务器的性能。***

Efficient storage organization and access is of central importance to the advent of large scale multimedia systems. Multimedia applications, ranging from video-on-demand (VOD) to medical tele-presense or large scale scientific computations, involve storage and retrieval of large volumes of information subject to a spectrum of real-time constraints. Current storage systems are inadequate to address the needs of multimedia systems. Indeed, various recent experiments in delivering VOD to populations of several thousand customers have established storage access as a key barrier to success. There is thus a growing need for a new generation of high-performance storage technologies that can efficiently support large scale multimedia systems. The main goal of the proposed research is to develop novel architectures and resource management algorithms for multimedia storage systems that will result in orders of magnitude performance improvements. The main thesis pursued here is that a storage system can exploit intrinsic, non-random, access characteristics of multimedia applications to accomplish dramatic performance improvements. This research effort will have several important contributions to the study of multimedia systems in general and to their storage technologies in particular. It will seek to characterize storage access patterns of the broad spectrum of multimedia applications and study techniques to take advantage of these characteristics in improving storage and access efficiency. The proposed study will consider the entire storage hierarchy, study the entire pool of resources and how these resources interact, as well as devise methods for incorporating application and workload constraints and characteristics into resource management techniques. Such media-aware, workload-aware, and application-aware storage organization and access mechanisms can significantly improve performance of a storage server. ***