Synthetic Performance Testing
We used a Supermicro dual Xeon storage server for testing today. The server is very similar if not identical to what a company would already have operating with 10,000 RPM drives today. For the server, we updated the slower Xeons that shipped with the system to E5 2690 8-core processors. We also swapped the slower RAID controller with an Areca ARC-1883 series.
We tested five sets of drives including two Micron 5210 Ion models, 1.92TB and 7.68TB. The comparison products all come from Seagate. We started with the Enterprise Performance 1.2TB, a 10,000-RPM model. Then wanted to add a modern 7,200-RPM series. The Seagate IronWolf Pro 12TB fit the bill. It is our standard drive for testing network-attached storage systems. Finally, we added a Seagate Enterprise Turbo SSHD array to the results to test a higher performance tier HDD.
Sequential Read Performance
Our test system uses Windows Server 2016 so some of our issues at 2 and 4 OIO carry over just like with the Windows 10 PCs. This is the first time we've experienced the sequential read 2 and 4 OIO performance drop with a RAID controller. The issue stands out more with the three HDD arrays but it affects all arrays in the sequential read test. We will look closer at the sequential performance in the mixed workload charts later in this review.
Sequential Write Performance
The 128KB sequential write preconditioning and steady state tests show strong performance consistency in an array behind a controller. The performance tests in the third chart show both 5210 Ion arrays outperforming the 10K Seagate array and on pace with the 15K SSHD.
Sequential Mixed Workload Performance
One of the largest improvements in the sequential workloads over the HDD arrays comes in the mixed test with the arrays performing both reads and writes. The 5210 Ion arrays can easily double the performance of mechanical drives in some of the mixes and it doesn't even take a large amount of data writes to see a massive performance improvement. For example, the Ion arrays with 90% reads easily double the average of the three disk drive arrays.
Random Read Performance
It's been several years since we've put a SSD in the same performance charts with hard disk drives. The results are the same even in powerful multi drive arrays. Flash has the benefit of no moving parts to retrieve data. Random performance is significantly up as while the latency is vastly down. This is how the 5210 Ion runs circles around mechanical drives in complex tasks like machine learning and advanced AI.
Random Write Performance
The 5210 Ion specifications show the 1.92TB model has the fastest random write performance of the three sizes released. A single 1.92TB delivers 13,000 IOPS and a single 7.68TB delivers 4,500 IOPS. Both deliver superior performance to hard disk drives including the SSHD models.
Random Mixed Workload Performance
The 1.92TB's random write performance doesn't give that model a clear lead in an array through all mixed random workloads. As we introduce more reads into the mix, the 7.68TB model catches up the smaller array.
It shouldn't be a surprise to see the Micron 5210 arrays outperforming the two HDD arrays through this entire test. The SSHD array managed to keep pace in the 100% read workload but quickly fell to HDD performance levels with the just 10% writes added to the mix.