全闪存Ceph的混合优化策略研究

MIXED OPTIMIZATION STRATEGY OF NVME SSD CEPH

  • 摘要: Ceph是一种经典的分布式存储系统,但随着数据量级和实验要求的不断提高,Ceph在读写延迟方面的不足制约了其进一步的发展。随着固态硬盘(Solid State Disk,SSD)等新型存储介质与Ceph的结合,Ceph整体读写性能有了明显的提升。全闪存SSD是一种性能优秀的SSD,基于全闪存SSD的Ceph系统表现也更好。与传统的分布式系统的读写延迟制约于后端存储介质不同,在基于全闪存SSD的情况下,Ceph系统的软件栈也需要优化以匹配硬件读写速度的提升。因此,提出基于全闪存SSD的Ceph存储系统的混合优化策略,针对Ceph的加锁机制、Crush算法等方面进行了优化,实验结果表明性能优化效果大约在2.8%~4.5%之间。

     

    Abstract: Ceph is a classic distributed storage system, but with the continuous improvement of data and experimental requirements, the lack of read-write delay of Ceph restricts its further development. With the combination of Ceph and new storage media, such as solid state disk (SSD), the overall read-write performance of Ceph has been significantly improved. Non-volatile memory express (NVMe) SSD is an excellent SSD, and Ceph system based on NVMe SSD performs better. Different from the traditional distributed system, the read-write delay of NVMe SSD Ceph isn't limited by hardware. This paper considers that in the case of NVMe SSD the software stack of Ceph also needs to be optimized to match the improvement of hardware read-write speed. Therefore, this paper proposes a hybrid optimization strategy for NVMe SSD Ceph: in terms of the locking mechanism of Ceph, Crush algorithm etc. The experimental results show that the improvement effect is about 2.8%~4.5%.

     

/

返回文章
返回