Toshiba selected a typical enterprise SSD storage configuration as a reference, comprising a total of eight 1.6TB SATA enterprise SSDs in a RAID6 configuration. These SSDs were installed in the reference server and driven by an internally connected RAID controller. To make the comparison, twenty-four HDDs with capacities up to 2.4TB were used as these cost approximately the same as the eight SSD. They were connected in a JBOD configuration using a second RAID controller and configured as RAID10, which is the fastest configuration with protection, albeit with a 50% redundancy overhead due to mirroring. As HDDs are relatively inexpensive when compared to SSDs, this was considered to be an equitable and valid test, even though the HDDs offered 28.8TB versus the 9.6TB from the SSDs.
The performance of both solutions was evaluated and compared using a synthetic random workload that included sixteen concurrent reading and writing tasks. This test generated the expected results; for small block sizes, which required a significant amount of data seeking, the resulting performance is dominated by the IOPS specification of the drive. In short, this means that the SSD solution offered significantly better performance.
However, as the block size rises, so the sequential performance specification increasingly dominates the result. Although SSDs are twice as fast as HDDs, when twenty-four HDDs are used in parallel, the HDD solution exhibits higher performance for block sizes above 64kB.
In real world applications, the actual workload usually comprises multiple different block sizes. This was approximated using a mix of block sizes from 4k to 2M. In this test, the HDD solution gave the best performance, showing that an array of HDDs with multiple spindles working in parallel can outperform SSD solutions in real world applications.
Toshiba has produced a detailed white paper covering this testing with comprehensive results tables. To download your copy, please click here: