To read more on our test methodology visit How We Test Enterprise SSDs, which explains how to interpret our charts. The most crucial step to assuring accurate and repeatable tests starts with a solid preconditioning methodology, which is covered on page three. We cover 4KB random performance measurements on page four, explain latency metrics on page seven and explain QoS testing and the QoS domino effect on page nine.
Our typical test regimen begins with six-hour preconditioning periods before each and every test, but our focus in this evaluation is to highlight the throughput capabilities of the FSA 200 platform. This removes the requirement for successive SSD preconditioning runs with random workloads as we make our blatant attempt to extract the maximum performance, and instead we merely used a 128K sequential write pass to precondition the SSDs.
The thrust of our testing is not to compare the performance and scalability of the NVMe-powered Intel DC P3700 SSDs against the SanDisk Fusion ioMemory SX300 SSDs. We only present the test results of 8 and 16 SSDs in RAID 0, which is the best configuration to achieve the maximum performance in a single host. Dual-host configurations will achieve double the performance.
There were time limitations on the evaluations process; this is quite a bit of hardware to have out on loan. These test results reflect a singular unique test environment, and with differing configurations, tuning and specialization, other users will extract varying levels of performance. There are a number of factors that affect performance at both the system and software level, and because such large SSD RAID setups are fairly rare and specialized, your mileage may vary (as always).
The 16-SSD Intel RAID 0 array barrels onto the chart with 3,055,735 4K random write IOPS at 1,024 OIO. To put this into perspective, this is nearly 12 GB/s of random throughput. It would take roughly 3,819 of the fastest 15K HDDs to generate this performance (assuming perfect RAID scaling). We are well over the rated Intel SSD specifications with 190,983 random 4K write IOPS per SSD, but this is because the SSDs are not in a steady state during the test. Conventional preconditioning methodology be gone; we are pushing the limits here.
The Intel SSDs scale well. The 8-SSD array, which we spread between two canisters, reaches 1,635,825 IOPS under the heaviest load. The 16-SSD SanDisk ioMemory SX300 follows with 1.4 million IOPS, and the 8-SSD SX300 array hovers at just under a million IOPS.
The 16-SSD Intel array punches a hole in the PCIe sky with 2.9 million IOPS under heavy load, which is over 190,000 IOPS per SSD—well under the Intel 450,000 IOPS rating for the individual SSDs. We are subject to the vagaries of software RAID during these tests, so it is difficult to discern if this is a limitation of the chassis, host or the operating system. The 8-SSD Intel array pulls into second place with 2.6 million IOPS, or 352,070 IOPS per drive, which follows the general trend of better scaling with fewer SSDs.
The SanDisk SX300 offers solid performance metrics of 1.8 million IOPS with 16 SSDs, and just over a million IOPS with the 8 SSD array. The Intel SSDs provide more performance under light loads (4 to 16 OIO) in both the 4K random read and write segments of our testing.
8K random write performance is important in a number of applications, and the 16-SSD SanDisk SX300 array sneaks up and steals the win from the Intel SSDs with an average of 1.276 million IOPS. The 8-SSD SanDisk array weighs in at over a million IOPS, while the 16-SSD Intel array sticks its foot in the door at 651,714 IOPS.
The Intel SSD arrays take the lead at 4 and 8 OIO, but the SanDisk SSDs take the lead for the remainder of the test, albeit with some variability below 128 OIO.
The 16-SSD Intel array returns to the limelight during the 8K random read tests with a score of 1.668 million IOPS, which is followed closely by 1.6 million IOPS from the 8-SSD array. The 8 and 16-SSD Fusion arrays provide 1.1 million and 673,931 IOPS, respectively, during the tests.
The Intel SSDs take the lead in our latency measurements during the test, though the SanDisk SSDs provide a better QoS profile in the 128 OIO breakout.