Solid State Devices (SSDs): PerformanceModeling and Analysis of Flash-based Storage Devices

H. Howie Huang, Shan Li, Alex Szalay, Andreas Terzi. Solid State Devices (SSDs): PerformanceModeling and Analysis of Flash-based Storage Devices. MSST 2011.

This paper models and analyzes the performance of flash-based Storage Devices (SSDs) with a Linear Regression Tree . Compares Intel and Samsung SSDs. They compare and analyze 3 SSDs: Intel X-25M, OCZ Apex and Samsung using a a couple of transactional and web search workloads.

Comments/Critiques:

1. Future of SSDs: I think flash storage would definitely have a place in future WSCs. However, in the next 10 years, I am unsure whether they will be used as a caching layer between the DRAM and the Disks or will replace Disks themselves (like Disks replaced magnetic tapes). An interesting trend to look out for will be how price/GB of DRAMs, SSDs and Disks vary over time.
2. MTBF of SSDs: An interesting analysis to look out for will be to find out the mean time between failures of SSDs. This in turn would affect the choice of their wide adoption in datacenters.
3. Fast Reads; Slower writes: Luiz Barrosso mentioned in his talk that poor write performance is one of the key challenges in using SSDs as a storage device. It would be interesting to see if they will find a place as the infrastructure-of-choice for read heavy workloads.
4. Rethinking Infrastructural Software Stack: If the SSDs indeed replace the disks, this would call for  a redesign of the underlying file system itself. Currently GFS/HDFS for instance heavily depend on sequential writes as opposed to random writes. While SSDs are better suited for the latter.