The next generation of storage

该提案通过调查我认为是存储系统的未来来解决计算基础设施的快速发展和存储系统的非常缓慢的发展之间的紧张关系。 我们必须开始了解如何构建和部署真正满足用户需求的存储系统，例如，可用性、便利性、灵活性和安全性，以及系统管理员的需求，例如，正确性、效率、可靠性和虚拟化感知。 我提出的架构工作在文件抽象级别（与许多系统相反，包括我以前的一些系统）在较低级别的块抽象级别上工作），将文件的存储与我们用来标识它们的命名系统分离，并为操纵命名空间提供丰富的抽象，以帮助解决查找，组织，以及管理存储系统中的文件。 文件作为不可变对象存储在单个实例存储中，并且可以复制到云中。 编辑文件时，将创建文件的新版本。 单实例存储中的对象将被复制以提供更高的可用性，并将自动版本化以提供从用户错误或恶意软件中的恢复。 在单实例存储的上面，我们使用名称空间数据库提供了灵活的名称空间组合。查询空间通过URL命名，可以共享，除了包含标准卷或文件系统之外，还可以是合成的，即定义为对所有文件的查询，例如“会计团队的所有Excel文件”或“包含对2011年建议预算的引用的所有Excel、Word和Powerpoint文件”。 合成文件也可用于跨单个虚拟机聚合数据，例如“所有病毒扫描日志文件”或“所有浏览器配置文件”，从而为系统管理员提供强大的工具，以确保遵守组织的安全和隐私政策。

This proposal addresses the tension between the rapid evolution of computing infrastructure and the very slow evolution of storage systems by investigating what I believe is the future of storage systems. We must begin to understand how to construct and deploy storage systems that actually meet the needs of users, e.g., availability, convenience, flexibility, and security, as well as the needs of system administrators, e.g., correctness, efficiency, manageability, and virtualization-awareness. The architecture that I propose works at the file abstraction level (in contrast to many systems, including a few of my previous ones) that work at the lower-level block abstraction level), separates the storage of files from the naming systems that we use to identify them, and provides a rich abstraction for manipulating namespaces to help address the complexities of finding, organizing, and managing the files that populate our storage systems. Files are stored as immutable objects in a single instance store and may be replicated into the Cloud. When files are edited a new version of the file will be created. Objects in the single instance store will be replicated to provide increased availability, and will be versioned automatically to provide recovery from user mistakes or malware. Above of the single instance store we provide flexible namespace composition using a namespace database. Namespaces are named via a URL, can be shared, and in addition to encompassing standard volumes or file systems, can be synthetic, that is, defined as queries over all files, such as "All Excel files from the accounting team'', or "All Excel, Word, and Powerpoint files that contain references to the 2011 Proposed Budget''. Synthetic files can also be used to aggregate data across individual virtual machines, such as, "all virus-scan log files'' or "all browser configuration files'', thereby providing system administrators powerful tools to ensure that organizational policies on security and privacy are being followed.