Benchmarking the performance of solid state drives, appliances and arrays is the latest play for vendors as they seek to sell their devices into an already crowded market.
Hard disk vs. solid-state drive
Recently, Texas Memory Systems released new performance results for its RamSan-630 SSD appliance. The RamSan-630 system, tested under both the SPC-1 and SPC-2 (Storage Performance Council) benchmarks, is a 3U appliance. The system supports Fibre Channel or InfiniBand connectivity. Fibre Channel was used in the SPC-1 test and InfiniBand was used in the SPC-2 test. The systems have 1.28TB of raw capacity made up of 20 640GB solid state drive modules. The system uses single-level cell flash chips (lithography not specified) and priced out, including three years of support, at around $420,000.
The SPC-1 benchmark results put the RamSan-630 at the top of the leader board. The following is a summary of results and a link to the Storage Performance Council report:
SPC-1 Submission Identifier: A00105
SPC-1 IOPS(tm): 400,503.26
SPC-1 Price-Performance(tm): $1.05/SPC-1 IOPS(tm)
Total ASU Capacity: 8,116.579 GB
Data Protection Level: Protected (RAID-5)
Total Price: $419,292.00
The SPC-2 benchmark results were represented two ‘firsts’ for the SPC-2. The first solid state storage system to complete the benchmark, and the first SPC-2 system to use InfiniBand connectivity. The following is a summary of results and a link to the Storage Performance Council report:
SPC-2 Submission Identifier: B00054
SPC-2 MBPS(tm): 8,323.13
SPC-2 Price-Performance(tm): $49.37/SPC-2 MBPS(tm)
Total ASU Capacity: 8,116.563 GB
Data Protection Level: Protected (RAID-5)
Total Price: $410,926.90
SNIA has now entered the fray with the development of a performance specification for solid state drives. The specification addresses issues related to the performance of a drive fresh out of the box, as opposed to one that has been completely written and needs to begin erasing and reprogramming areas previously written. It also addresses issues related to compressible and in-compressible data, which we covered in our story regarding write amplification:
The specification defines a test hardware configuration and appropriate workloads for establishing benchmark information regarding reads, writes and bandwidth with data points associated with varying block sizes and data sets. Both an enterprise version and client version of the specification are under development and scheduled for release later this year. More information can be found here.
