Random Read and Write Patterns
Now we will see how the performance of the SSDs in random read and write modes depends on the size of the requested data block. The requests queue depth in this test equals 1.
Wow! What a great difference between the “brothers-in-chip” – Crucial m4 and Intel 510! Looks like Intel bet the streaming tests, while Crucial – the IOps per second. The performance difference on small data chunks in phenomenally high. I wonder what we are going to see in file tests, which approach will prove the best?
It is interesting that when we switched Crucial m4 to SATA-300 interface the drive slowed down by about 1.5 times on small data blocks.
In the vertex 3 clan the leadership is in the hands of the 240 GB model, and the largest model suddenly yielded to the smallest one. Why did it happen? Does the 480 GB model use different memory?
Let’s see what happens with writes:
And here we see yet another surprise from Crucial. These SSDs are slow on small data blocks and then they suddenly speed up on 4 KB blocks.
I have to point out that it is not a problem of our patterns, because all requests in them are adjusted for 4 KB. So, it is most likely the result of some software “optimizations”.
Intel SSDs act much more reasonable: no sudden jumps or drops in speed…
Well, things are exactly the same here: Vertex 3 SSDs do not like small data blocks during random writes. And judging by the vertex 2 results, we can conclude that this dislike is inherited.
Now let’s see how our testing participants, who demonstrated such diverse results in synthetic tests, will perform in file benchmarks.