The Community for Technology Leaders
RSS Icon
pp: 1
Xiang Song , Shanghai Jiao Tong University, Shanghai
Jian Yang , Shanghai Jiao Tong University, Shanghai
Haibo Chen , Shanghai Jiao Tong University, Shanghai
Flash-based solid-state drive (SSD) is now being widely deployed in cloud computing platforms due to the potential advantages of better performance and less energy consumption. However, current virtualization architecture lacks support for high- performance I/O virtualization over persistent storage, which results in sub-optimal I/O performance for guest virtual machines (VMs) on SSD. Further, current software-based I/O virtualization violates the "don't hide power" principle due to inefficient support for some advanced SSD commands (e.g., TRIM) and constrained parallelism, leading to sub-optimal performance and life cycle. This paper observes that the massive internal parallelism and the block emulation in the flash translation layer (FTL) make flash-based SSD an ideal candidate to support high-performance I/O virtualization for persistent storage. Based on this observation, we propose VFlash, the first storage I/O virtualization architecture that extends existing SSDs with trivial hardware changes to directly expose multiple virtual SSDs to guest VMs. Performance evaluation using a modified FlashSim with two FTL schemes (i.e., DFTL and FAST ) shows that VFlash incurs only small performance overhead over native SSDs and can efficiently exploit parallelisim.
simulation of multiple-processor systems, B.4.4 Performance Analysis and Design Aids, C.4.g Measurement, evaluation, modeling
Xiang Song, Jian Yang, Haibo Chen, "Architecting Flash-based Solid-State Drive for High-performance I/O Virtualization", IEEE Computer Architecture Letters, , no. 2, pp. 1, RapidPosts RapidPosts, doi:10.1109/L-CA.2013.22
264 ms
(Ver 2.0)

Marketing Automation Platform Marketing Automation Tool