Degradation and Steady-State Performance
Unfortunately, SSDs are not always as fast as in their “fresh” state. In most cases their performance goes down after some time and in real life we deal with completely different write speeds than what we see on the diagrams in the previous chapter of our review. The reason for this phenomenon is the following: as the SSD runs out of free pages in the flash memory, its controller has to clear memory page blocks before saving data into them, which causes substantial delays. Although, modern SSD controllers can alleviate the performance drop by erasing unused flash memory pages ahead of time, when idle. They use two techniques for that: idle-time garbage collection and TRIM.
Of course, users are more interested in the consistent performance of their SSDs over a long period of time rather than the peak speed they are going to see only during the initial short-term usage period, while the drive is still “fresh”. The SSD makers, however, declare the speed characteristics of “fresh” SSDs for marketing reasons. That’s why we decided to test the performance hit that occurs when a “fresh” SSD becomes a “steady” one.
To get a complete picture of SSD performance degradation we ran special tests based on the SNIA SSSI TWG PTS (Solid State Storage Performance Test Specification) methodology. The main idea of this approach is to measure write speed consecutively in four different cases. First we measure the “fresh” SSD speed. Then we measure the speed after the SSD has been fully filled with data twice. The third test occurs after a 30-minute break during which the controller can partially restore performance by running the idle-time garbage collection. And finally, we measure the speed after issuing a TRIM command.
We ran the tests in synthetic IOMeter 1.1.0 RC1 benchmark, where we measured random write speed when working with 4 KB data blocks aligned to flash memory pages at 32 requests queue depth. The test data were pseudo-random.
The Kingston SSDs all perform as expected in this test. The SandForce platform is like a diagnosis. Its highs and lows are reflected in the same measure in any SSD from any brand. We can see that the Kingston SSDs do not do much background garbage collection and their performance can only be restored by the TRIM command. Yet even TRIM can’t bring them back to their original speed: the steady-state performance of the Kingston SSDs is going to be somewhat lower compared to their out-of-box state.
Interestingly, TRIM works differently on the HyperX and HyperX 3K and for a good reason. The HyperX 3K uses memory with a shorter service life, so Kingston has adjusted its garbage collection and made it less aggressive than in the flagship model. This lowers the write amplification factor and increases the service life of the SSD but the latter suffers a heavier performance hit when filled with data. The same optimization has been done to the SSDNow V+200 which is equipped with flash memory rated for 3000 rewrite cycles as well.
Thus, the steady-state performance of the HyperX 3K is going to be somewhat lower in terms of writing compared to the HyperX and to other SandForce-based SSDs with synchronous memory whose manufacturers do not bother to tweak the default algorithms. We measured the write speed of the SSDs once again to see that difference. You can see the results in the following diagrams (the SSDs were filled with data at random):
As it turns out, Kingston's HyperX 3K and SSDNow V+200 with their adjusted garbage collection algorithms are only slower than other SSDs of the same class when writing data with a long request queue. That’s not very relevant for desktop SSDs because desktop applications can hardly generate such a queue. So, Kingston's solution seems to be viable.
We can also see that every SSD based on the SF-2281 controller suffers a performance hit when transitioning from its out-of-box to steady state. The diagrams show that the SSDs with third-party controllers can outperform SandForce-based products in more scenarios than expected. We must keep it in mind, however, that write operations are less frequent in everyday applications than read ones, so the rather low performance of the Kingston SSDs illustrated by the last three diagrams isn’t a disaster.