Mixed Workload And Steady State Testing
In the past, we could look at four-corner performance to get a good idea of how a product would act in the real world. That's no longer the case, though. Mixed-workload tests yield more useful info on what your SSD does while multitasking and running background software.
Our mixed workload testing is described in detail here, and our steady state tests are described here.
Our new mixed workload tests show performance at several queue depths with 80% reads and 20% writes. The SM2256 comes up short of several existing SSDs, but proves quicker than SanDisk's Ultra II (a low-cost drive with SanDisk three-bit-per-cell flash).
Silicon Motion fares a bit better when we benchmark with random data.
This set of tests typifies a sequentially-accessed mixed workload at steady state. It's an extreme use case - for example, what you might see after several hours of editing large video files and then rendering to the same drive. Most users won't experience anything quite like this.
When I first created this chart, I didn't expect to see any product spike and dip so wildly. But the SM2256 is all over the place. Its steady state 4KB performance varies more than any other SSD I've ever tested.
This chart illustrates the final 100 seconds of our test. Again, steady state benchmarking is well outside what we consider normal usage, but we still get a good idea of how well a product might perform in a RAID environment. We look for consistent behavior across the graph, without wild peaks and valleys. That's the opposite of what Silicon Motion's sample drive shows us when it's taxed by 4KB writes for hours.