海量规模下高性能对象存储服务技术的研究

发布时间：2018-03-16 14:35

本文选题：对象存储　切入点：基准测试　出处：《上海交通大学》2013年硕士论文　论文类型：学位论文

【摘要】：随着对象存储技术的不断深入人心，目前已经涌现出一大批功能相近接口相似的对象存储系统。然而，由于对象存储系统实现的多样性以及每个系统本身的复杂性，外加缺乏模拟对象存储应用的通用负载工具，导致人们不仅难以在众多的系统中选出最适合当前应用场景的解决方案，同时也无法有效对这些解决方案进行系统调优以及性能优化。为了应对上述问题，本文提出了对象存储的应用模型，并在此基础上设计并实现了对象存储性能基准测试工具：COSBench。COSBench具有很好的可扩展性、可伸缩性、灵活性，以及易用性，从而使得人们能够方便地将COSBench适配至异构的存储系统，并模拟出各类不同的测试场景，包括那些极端的使用模式。目前，COSBench用于支持Intel内部多个对象存储系统的性能评估以及系统调优工作。此外，COSBench也成功地得到了10多个企业客户的使用以及认可。在未来，COSBench将转为开源软件，从而使得更多的人能够接触到这款工具。在系统调优以及性能优化方面，本文以Openstack Swift为案例，探讨了该系统在4种不同使用场景下的性能瓶颈以及硬件资源使用模式，并在此基础上提出了相应的系统调优方案以及性能优化方案。此外，本文还对Swift系统的源代码进行了剖析、对实验经验进行了总结，并对除性能外的其他重要系统属性进行了相关的介绍以及说明。结合本文的成果，人们可以针对不同的使用场景以及性能瓶颈来更加合理地部署Swift系统，从而能够在给定的性能要求下，尽可能地降低构建系统所需的成本。于此同时，，本文的实验数据也可帮助人们更好地挖掘系统潜在的性能优化机遇，并更有针对性地改进现有的系统架构以及算法设计，从而能够在给定的预算条件下，尽可能地提高系统的总体性能。
[Abstract]:With the deepening of the object storage technology, has emerged a large number of similar functional interface similar to the object storage system. However, due to the complexity of implementation of the object storage system and the diversity of each system itself, and general lack of simulation object storage load application tool, resulting in many people not only to choose the solution system for the current application scenarios, but also not effective against these solutions for system optimization and performance optimization.
In order to deal with these problems, this paper puts forward the applied model of object storage, and on the basis of the design and implementation of the object storage performance benchmark tools: COSBench.COSBench has good scalability, scalability, flexibility, and ease of use, so that people can easily be adapted to heterogeneous storage system COSBench, and simulate various test scenarios, including those with extreme mode. At present, COSBench was used to evaluate the performance of Intel internal support multiple object storage system and system optimization. In addition, COSBench also successfully obtained more than 10 corporate customers and recognition. In the future, COSBench will be converted to open source software thus, more people can come into contact with this tool.
In the aspect of system optimization and performance optimization, this paper takes Openstack Swift as a case, discusses the system in 4 different scenarios of the performance bottleneck and hardware resource usage patterns, and put forward the corresponding scheme of system tuning and performance optimization scheme. In addition, the source code of the Swift system. Analysis on the experimental experience were summarized, and the other important system properties besides high-performance are introduced and illustrated. According to the results, people can according to different usage scenarios and performance bottlenecks to more reasonable deployment of Swift system, which can in the given performance requirements, as far as possible to reduce construction for the cost of the system. At the same time, the experimental data can also help people to better exploit the potential opportunities for system performance optimization, and more targeted to the Some system architecture and algorithms are designed to improve the overall performance of the system as much as possible in a given budget.

【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333

【相似文献】