TEST SETUP AND METHODOLOGY
In testing all enterprise drives we focus on long-term stability. In doing so, we stress products not only to their maximum rates, but also with workloads suited to enterprise environments. We use many off-the-shelf tests to determine performance, but we also have specialized tests to explore specific behaviors we encounter. With enterprise drives, you will see that we do not focus on many consumer-level use-cases.
When testing SSDs, the drive is purged and then preconditioned into a steady state before capturing its performance results. We also disable all write caching on the DUT when possible, this ensures consistent results that are compliant with SNIA standards. Our hope is that we present tangible results that provide relevant information to the buying public.
|PC CHASSIS:||Thermaltake Urban T81|
|MOTHERBOARD:||ASRock X99 WS-E|
|CPU:||Intel Xeon E5-2690 v3|
|CPU COOLER:||Thermaltake Water 3.0 Ultimate|
|POWER SUPPLY:||Thermaltake Toughpower 1500W Gold|
|GRAPHICS:||MSI GT 720 |
|SYSTEM COOLING:||be quiet! Silent Wings 2|
|MEMORY:||Kingston ValueRAM DDR4 2400MHz ECC 64GB|
|STORAGE:||Crucial MX200 500GB|
|OS:||Windows Server 2012 R2|
This Test Bench build was the result of some great relationships and purchase; our appreciation goes to those who jumped in specifically to help the cause. Key contributors to this build are our friends at ASRock for the motherboard and CPU, be quiet! for the cooling fans, and Thermaltake for the case. We have detailed all components in the table below and they are all linked should you wish to make a duplicate of our system as so many seem to do, or check out the price of any single component. As always, we appreciate your support in any purchase through our links!
The Storage Networking Industry Association has an entire industry accepted performance test specification for solid state storage devices. Some of the tests are complicated to perform, but they allow us to look at some important performance metrics in a standard, objective way.
SNIA’s Performance Test Specification (PTS) includes IOPS testing, but it is much more comprehensive than just running 4KB writes with Iometer. SNIA testing is more like a marathon than a sprint. In total, there are 25 rounds of tests, each lasting 56 minutes. Each round consists of 8 different block sizes (512 bytes through 1MB) and 7 different access patterns (100% reads to 100% writes). After 25 rounds are finished (just a bit longer than 23 hours), we record the average performance of 4 rounds after we enter steady state.
- Purge: Secure Erase, Format Unit, or vender specific
- Preconditioning: 2x capacity fill with 128K sequential writes
- Each round is composed of .5K, 4K, 8K, 16K, 32K, 64K, 128K, and 1MB accesses
- Each access size is run at 100%, 95%, 65%, 50%, 35%, 5%, and 0% Read/Write Mixes, each for one minute.
- The test is composed of 25 rounds (one round takes 56 minutes, 25 rounds = 1,400 minutes)
Unlike our other performance tests, the SNIA tests only last for a relatively short period of time each (1 minute), but they cover many more access patterns and transfer sizes. All tests were done at a QD of 32. The Micron 5100 ECO and MAX achieved 94K AND 95K IOPS, respectively, in 100% read performance at 4KB, which is more than the 93K rating. Taking a look at 100% writes, we can see that both drives, again, surpassed their ratings. The 5100 ECO achieved 28K IOPS while the 5100 MAX achieved 83K. Furthermore, during the 4KB 65%/35% read/write portion of the test the drives averaged 48K and 75K IOPS, which is just about where it needs to be.
To specifically measure latency, we use a series of 512b, 4K, and 8K measurements. At each block size, latency is measured for 100% read, 65% read/35% write, and 100% write/0% read mixes.
The Micron 5100 MAX’s average latency is obviously better than the 5100 ECO’s, however, in pure read applications we can see that they are pretty much tied. For 100% read they averaged 0.33ms at the 0.5KB and 4KB file sizes and 0.66ms at 8KB. During the 65%/35% read/write mix the 5100 ECO’s average latency reached up to 2.2ms while the MAX’s results all maintained under 2ms. In 4K & 8K writes the MAX model’s latency is nearly half that of the ECO model.
Moving onto the maximum latency results we can clearly see that the MAX model’s extra over-provisioning helps to keep the latency down compared to the ECO model, but at the 0.5KB file size it still can have latency spikes over 30ms at a QD of 32.