Results: Write Saturation And Over-Provisioning Tests
Random Performance Over Time
A saturation test consists of writing to a drive for a specific duration with a defined workload. Technically, this is an enterprise-class write saturation test, where the entire LBA space of the SSD is utilized by a random write at high queue depths.
This is 500 minutes worth of one-minute averages, with a 20 minute breakout on a per-second basis. Again, this sort of arbitrary workload on a fresh drive results in a peak of just 22,000 IOPS (though few random workloads will continuously pepper the entire LBA space with random writes at a queue depth of 32). In this test specifically, SanDisk's clever nCache technology imparts lower performance. In most of the workloads the X210 will be exposed to, nCache should do the opposite, speeding up benchmarks instead of the other way around.
As far as the raw data goes, our 20-minute slice of steady-state performance breaks down fairly well considering the X210 has just 7% spare area to rely on. It averages 4445 IOPS, showing a minimum of 2472 IOPS and a maximum of 7384. That's not bad. But we want to know more. Perhaps a quick look at various levels of over-provisioning will tell us something.
Over-Provisioning
We can't truly over-provision a drive in the lab. There are numerous ways to trade capacity for better performance; we could convince the OS that our drive has fewer logical block addresses for example, but the SSD knows otherwise. Even so, certain workloads are much better off if we sacrifice a bit of capacity. That leaves more room for the drive to perform background processes (like garbage collection), and the results can be profound, even if the controller still has to map the existing space. If we start with the number of bytes available to the user (256,052,966,400 in this case), we can run the write saturation test again. Each successive time (after a secure erase), the amount of accessible space is reduced.
The following chart shows five separate points: the full span (as above), 93%, 84%, 50%, and 25%.
There is a near-doubling of performance when we sacrifice an additional 7% of the addressable LBA space. The next step up, leaving 84% of the drive's capacity, allows the X210 to sustain almost 17,000 IOPS. To multiply that number, you need to give up 50% of the 256 GB SSD's space, a hefty price to pay, for sure.
Lastly, this is a 20 minute breakdown of the same over-provisioning experiment taken from the end of the multi-hour plot and displayed in one-second increments. As you can see, the spread between minimum, maximum, and average IOPS can get pretty massive.
