Improving The Performance Of I/O Intensive Workloads Using SSD-Aware Multi-Level Caching Hybrid Storage System An Empirical Study
Processor speed has increased at a steady pace year after year, while storage system speed is still lagging behind. This speed gap has become a major bottleneck for most computers including high performance computer systems. Many attentions have been given to optimize the storage system performance using various approaches such as caching, intelligent prefetching and scheduling techniques, nevertheless, the storage system remains the performance bottleneck for most computer systems due to its internal design that rely on moving parts to locate and retrieve data. The advent of Solid State Devices (SSD) has opened a new door to storage optimization. These devices are capable of producing not only exceptional bandwidth, but also random I/O read performance that is orders of magnitude better than that of standard rotating mechanical disks due to the absence of moving parts. There are growing motivations in using SSDs (along with the current storage devices) as a new layer in the memory hierarchy to form a hybrid storage system in order to exploit the potential of both type of devices. In this paper, we have conducted an extensive empirical and comparative study of an I/O intensive workloads running on a real machine with SSD-aware multilevel caching implemented at the file system level. We measured and demonstrated the effectiveness of using SSD as a read cache extension to the main memory in order to reduce the disk hits ratio that would otherwise cause substantial delay in workload performance. We show that the workloads IO Per Seconds (IOPS), the file system IOPS, Storage IOPS, latency per I/O and dollar amount per I/O is also improving by using a hybrid storage system with an SSD-aware multi-level caching.
SSD ZFS Multi-level caching Hybrid storage system
Abdullah Aldahlawi Esam El-Araby Tarek El-Ghazawi
ECE Department The George Washington University Washington DC, USA EECS Department The Catholic University of America Washington DC, USA
国际会议
2010 International Conference on Future Information Technology(2010年未来信息技术国际会议 ICFIT 2010)
长沙
英文
914-921
2010-12-14（万方平台首次上网日期，不代表论文的发表时间）