SBIR Phase I: pNFS in the Cloud (pITC): A Replicated, Parallel File System for Cloud Computing

这个小型企业创新研究（SBIR）第一阶段项目将基于最近完成的NFSv4.1/pNFS标准开发一个云文件系统平台，并扩展了灵活的复制和服务局管理。该项目的智力价值在于灵活复制的新机制，该公司的意思是可变复制（可写数据的复制），基于客户端的复制和具有协商一致性保证的复制。该公司提出了pNFS复制布局，这将使服务器和客户端能够协商最适合应用程序一致性要求的管理首选项的复制策略。具有一致性的可变复制是分布式文件归档技术的一个主要进步，因为主流分布式文件系统中的实现还没有出现。该项目更广泛的目标包括推进基于标准的存储基础设施，推进开源和专有文件系统技术，以及开发易于商业化的软件基础设施。其结果非常适合在商用计算设备上部署，也适合与专用的快速存储和网络硬件集成。这项工作的目标包括提高当前存储管理抽象的灵活性和简单性，超越当前最先进的技术水平。企业级当前系统的经验表明，卷抽象可以通过分类（例如具有特定可用性和一致性保证的卷、复制策略或底层存储功能）来丰富，组织可以根据这些分类构建策略。 研究人员已经收到了来自商业和学术组织的兴趣表达，他们愿意参与早期试点，以提供测试，反馈和验证。

This Small Business Innovation Research (SBIR) Phase I project will develop a cloud file system platform based on the recently-finalized NFSv4.1/pNFS standard, with extensions for flexible replication and service-bureau management. The intellectual merit of the project lies in novel mechanisms for flexible replication, by which the company means mutable replication (replication of writable data), client-based replication, and replication with negotiated consistency guarantees. The company propose pNFS replication layouts, which will enable servers and clients negotiate replication strategies that best fit administrative preferences to application consistency requirements. Mutable replication with consistency is a major advance in the state of the art for distributed filing, as implementations in mainstream distributed file systems have not emerged. The broader goals of the project include advancement of standards-based storage infrastructure, advancement of open source as well as proprietary file system technology, and development of software infrastructure that can be readily commercialized. The result is well suited to deployment on commodity computing equipment, but also to integration with dedicated, fast storage and networking hardware. The goals of the effort include increasing the flexibility and simplicity of current storage management abstractions, exceeding the current state of the art. Experiences with the current system at enterprise scale suggest that the volume abstraction be enriched with classifications (such as volumes with specific availability and consistency guarantees, replication strategies, or underlying storage capability), from which organizations can construct policies. The investigators have received expressions of interest from commercial and academic organizations willing to participate in early stage pilots to provide testing, feedback, and validation.