Mixed Workload And Steady State
80 Percent Sequential Mixed Workload
We were surprised to see the smaller CS1311 performing slightly better than the higher-capacity model in our sequential mixed workload test. The 240GB CS1311 seems to flush its buffer more frequently than the larger model. And in some workloads, greater performance after the flush makes the smaller drive faster than the large one. In reality, this would probably only happen in rare corner cases though, seemingly at random.
80 Percent Random Mixed Workload
The random mixed workload test shows us exactly what we expected to see based on our previous random transfer results. Both CS1311s find themselves at the bottom of this chart, even when there aren't many outstanding commands. There is only a slight speed-up from a queue depth of two to eight. We see the drives accelerate after that, but by then it's too late. Most desktop workloads aren't intense enough to see commands stack up.
Sequential Steady State
The sequential steady state tests expose a larger gap between the CS1311 SSDs than we've seen in previous benchmarks. The extra CE channels help the 480GB model. Then again, when we put smaller SSDs against larger ones, the lower-capacity drives almost always succumb. To keep our analysis balanced, we're only focusing on similar capacities. The 240GB CS1311 is only there because we received it at the same time and actually want to see how it fares.
In consumer desktop (80 percent reads) and workstation (70 percent reads) steady state workloads, the 480GB CS1311 does well against other low-cost SSDs. Performance drops off dramatically after 70 percent reads until the direct-to-die write algorithm helps increase performance over the two drives with Silicon Motion's SM2256 controller.
Random Write Steady State
The preconditioning test shows us behavior we've never seen from an S10-based SSD armed with TLC flash. Something is different with the emulated SLC buffer that we didn't observe from the MyDigitalSSD or Patriot drives. The random steady state chart shows the final 100 seconds of our metric, where we see the buffered writes that peak to 42,000 4KB IOPS. Then, random write performance drops to a very low level before inching back up to 4500 IOPS.