8KB Random Read And Write
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 8KB random performance measurements on page four, explain latency metrics on page seven, and explain QoS testing and the QoS domino effect on page nine.
The 8k random write tests follow a similar pattern and the two value entrants fall to the bottom of the tests. The Intel DC P3520 QoS measurements top 450ms under the heaviest load, while the PE3110 reaches a top latency measurement of 50ms. The increased QoS measurements only come during very heavy workloads, which isn’t the DC P3250’s intended environment. We also note a higher DC P3520 QoS envelope during the 32 OIO 8k random write breakout. The histogram reveals a broad range of latency distributions for the DC P3520 and the PE3110, though we do note that the PE3110 begins with higher latency.
Power consumption continues to be a very encouraging story for the value models. The DC P3520 requires nearly half the power of its P3700 flagship predecessor, which illustrates how much of an advantage that IMFT 3D NAND provides. This is likely due to the reduced power consumption of each die, but density also helps eliminates some die, as well - density is just a win-win for power consumption.
The DC P3520 and PE3110 still lag the test filed in IOPS-per-Watt efficiency metrics, but it is important to note that we derived these metrics while the device is under full load, which is a rare encounter for even a heavy workload, let alone a mainstream read-centric application. This reality diminishes some of the perceived gains of the higher IOPS-per-Watt metrics, and the value parts will provide a tangible gain on both fronts during mainstream workloads.
Once again, the low end of the scaling tree reveals great performance from the DC P3520 under 32 OIO. The SSD scales well with increased intensity, and both the latency and QoS-over-IOPS charts reveal a great performance profile, especially when we cast the DC P3520 in the light of reduced cost. The QoS breakout at 32 OIO reveals the same somewhat odd envelope during the read workload. On a brighter note, we were able to come within a mere 1,000 IOPS of the random read specification.
The 8k mixed workload finds the DC P3520 continuing to leverage its superior channel count to lead the PE3110 for the majority of the test, but the two SSDs achieve near parity after the 30/70 read/write workload. Both of the value entrants continue to provide exceptional performance consistency. Flagship high-endurance SSDs couldn’t muster such consistent performance even a few short years ago. The DC P3520 might have a bit of a firmware tuning issue under heavy load at the 90/10 mixed read/write workload, as the same erratic performance that we encountered during the 4K mixed tests crops up again at this mixture. We also noticed the increased QoS latency during the mixed tests at 256 OIO, but the heavy workload is beyond Intel’s QoS specifications, which stop (appropriately) at QD128.