云存储系统中海量时效数据的组织模式及关键技术研究

Current massive storage systems need to address the main problems of weak availability in data, low performance in systems and delay constraints in values. These problems cause the challenges of scalability and performance improvements in storage systems. In order to address these problems and offer the delay-constrained quality of storage system services, this proposal proposes the methodology of approximate processing in storage systems for delay constraints and key techniques for performance improvements. The basic idea is to explore and exploit the inherent principles of approximate processing in storage systems, i.e., potential semantic correlation, to divide the datasets and group correlated data, thus obtaining the suitable tradeoff among the data amounts, computation complexity and result accuracy and improving entire system performance. The research work will study the principle and model of approximate processing in storage systems and dynamic namespace schemes. The semantic-aware feature vectors and storage system organization model for near real-time operations can support the processing of the multi-source, heterogeneous and timing-contrained data and provide high-quality storage services. Furthermore, the in-memory based model for combining computation and storage can leverage the advantages of computation models and storage structures to support correlation-aware rapid identification. Based on previous achievements, this work will use real cloud storage platforms to examine and analyze the system performance.

当前云存储系统面临的主要问题包括：海量数据的弱可用、存储系统的低效能和数据价值的强时效，使得存储系统的可扩展性和性能优化面临挑战。为解决这些问题、提高时效约束的存储服务质量，提出面向时效约束的近似处理的系统组织模式、方法论和性能优化的关键技术。基本思想是通过研究近似处理的系统内在机理，即隐含的语义关联特征，对数据集进行语义划分，使得对语义分组的数据管理操作在处理数据量、计算复杂度和结果准确性方面达到量效均衡，进而优化存储系统的整体性能。研究工作注重研究近似处理的系统内在机理和演化规律、以及动态命名规则，并通过构建语义感知的特征向量和支持近实时操作的存储系统组织模式来融合多源、异构、时效的数据，提供高质量的存储服务。面向内存级的存算联动模型能够进一步融合计算模式和存储结构的特点，支持关联特征的快速感知和识别。研究工作具有一定的前期基础，将在真实的云存储系统平台上进行系统实现和测试分析。

项目注重在云存储系统中面向海量时效数据开展分析和研究工作，在存储器件、内存系统和优化机制等多方面取得重要研究进展，发表了多篇学术论文，这主要包括海量时效数据的高性能内存组织模式、基于数据语义关联关系提升新型内存可靠性的方法、面向新型内存硬件感知的加密技术、云存储系统中基于近似处理的能耗有效的方法和带宽有效的云存储系统优化和安全保证机制等。项目发表学术论文38篇，其中中国计算机学会CCF推荐的A类国际会议论文11篇，A类国际期刊论文10篇，中文A类期刊论文2篇。出版英文著作1部。已经申请和授权国家发明专利7项，其中授权发明专利1项，申请发明专利6项。研究成果获得IEEE HPCC 2021的最佳论文奖，2020年教育部自然科学一等奖，2018年湖北省科技进步一等奖。应邀在国内学术会议上作特邀报告2次，在国际学术会议/论坛上作特邀报告2次。相关成果具有重要的实际意义和应用价值。