Increasingly, we pay close attention to the performance consistency of enterprise-class SSDs. This is what separates a good drive from a great one when all of the corner case testing seems equal. Over the past year, we measured this in terms of large block transfers in our Enterprise Video Streaming section. Armed with this data and our exclusive analysis, the peaks, valleys, and frequency of each become clear. If you look at the information for long enough, you start to see fingerprints for each drive.
We start with large-block transfers because, in enterprise video applications, if you don't buffer or write data fast enough, you can lose it completely. Random 4 KB transfers are slightly more academic, but they also match database transfers more closely. With this sort of workload, you might not lose data, but the system will slow down.
For the following tests, we subject SanDisk's Optimus Eco and Intel's SSD DC S3700 to 25 hours of continuous random 4 KB writes across each drive. We record IOPS every second, giving us 90,000 data points. We then zoom in to the last 60 minutes to more coherently visualize the results.

Again, the Optimus Eco gives Intel's SSD DC S3700 a run for its money. The Intel drive's consistency is slightly better, but SanDisk still does well. Every single one-second average falls between 28,500 and 38,000 IOPS (0.84 and 1.12 ms), while 73% are above 33,000 IOPS.

Looking at a histogram of the data, we see a nice, even distribution. Often when we generate this data, we see a high concentration centered around the vendor's claimed spec, but a non-trivial amount of data landing at higher latency levels. Typically, we attribute that to wear leveling or garbage collection. Even Intel's SSD DC S3700 exhibits some of that behavior.
The Eco isn't like that. Although the S3700 demonstrates better performance consistency, it is only slightly better, and that advantage probably won't be noticeable in real-world apps.
- Meet SanDisk's Optimus Eco SSD, With Up To 2 TB
- Under The Hood Of SanDisk's Optimus Eco
- SanDisk's Guardian Technology, Broken Down
- Testing SanDisk's Optimus Eco
- Results: 4 KB Random Performance And Latency
- Results: Performance Consistency
- Results: Enterprise Workload Performance
- Results: Sequential Performance
- Results: Enterprise Video Streaming Performance
- SanDisk Takes On Intel's Enterprise SSD Crown
Every single one-second average falls between 28,500 and 38,000 IOPS (0.84 and 1.12 ms)
The Samsung Pro is not an enterprise drive. They were comparing Intel's enterprise drive vs Sandisk's.
Many consumer SATA drives are a lot less, but enterprise drives aren't always quite that low. The S3700, at 800GB, is 6W typical and 8W burst. The Eco isn't quite as 'eco' at 400GB, but for 2TB, is actually pretty good. Many of the PCIe add-in SSDs that provide better performance at the same capacity are at least 10-15W and sometimes 25W. There aren't a lot of 6Gbps SAS SSD comparisons, now that 12Gbps drives are out, but even the Toshiba MK line is rated at 6.5W. We plan on doing more power consumption testing in the future.
As was said earlier, the Samsung 840 products are not enterprise class. They do not provide the endurance or power loss protection. They could possibly be used in workstations, but not beyond that. Samsung does offer the 843T, but that product is more in line with the Intel S3500 and does not have the random write performance to come close to the Eco. The 843T is also much more expensive than the 840 series.
When comparing to HDD, there isn't a single SSD that will come close on price, enterprise or not. On the flip side, there isn't a single HDD that can come close on performance either. In order to get that much flash storage, you were previously limited to multiple SSDs or PCIe add-in cards, the Eco allows you to have that capacity in a smaller form-factor while drawing less power. Considering the $/GB, which is in line for it's class, it makes sense since there are plenty of enterprise customers buying 800GB drives, at least enough that companies keep producing them.
Every single one-second average falls between 28,500 and 38,000 IOPS (0.84 and 1.12 ms)
No, we didn't, but can you be a little more clear with your question? I am not an expert in the area, but the law, as it applies to performance testing, is valid if the number of jobs in the system is equal to those being completed. Meaning that no new jobs are created in the system and no jobs are lost forever. So, if jobs were being lost, you might see consistent performance centered around the bottleneck, which is not the device under test. Since the system is artificially creating and tracking IOs, I don't believe any are being lost or accidentally created. Also, with our Enterprise Video testing, which I wrote to test a specific use-case, the data is generated and validated as it is written, so there is no chance of data loss. If that doesn't answer your question, please let me know, I am always interested in new subjects....