In the Database pattern the drive is processing a stream of requests to read and write 8KB random-address data blocks. The ratio of read to write requests is changing from 0% to 100% with a step of 10% throughout the test while the request queue depth varies from 1 to 256.
You can click these links to view the tabled results:
We will build diagrams for request queue depths of 1, 16 and 256.
There is a couple of SSDs that deliver outstanding results at a request queue depth of 1. One of them is the highly consistent OCZ RevoDrive which copes neatly and confidently with every load. The other is the Crucial C300 whose behavior betrays very advanced deferred writing algorithms. It also takes first place at pure reading. However, The Crucial C300 only shows such results when connected via SATA300. Its performance is far from outstanding on the Marvell controller.
Among the remaining SSDs, the RAID0 array and the Colossus, being dual-controller solutions, have higher results. The Vertex 2 with new firmware 1.1 is fast, too.
As the request queue gets longer, the leaders make their preference clear: the RevoDrive wins at reading while the Crucial C300 breaks all records at writing. It is a mystery to us how the C300 can accomplish such wonders, but its performance – tens of thousands of write operations per second! – is truly impressive. Among the other products we can single out the RAID0 array built out of two Corsair F120 drives. Alas, the scalability of the array is not 100%.
The Intel X25-V, G.Skill and Colossus are the losers in this test. The 40GB SandForce-based model doesn’t slow down much, though.
Winding up this part of our tests, we will build diagrams showing the performance of each SSD at five different request queue depths.
This pair of diagrams makes it clear what benefits can be achieved by changing a single SSD with a RAID array. The write speed improves considerably, which also has a positive effect on the performance under mixed load. And if you can load your disk subsystem so high that there occurs a requests queue, the reading performance will grow up with the RAID array as well.
If the first diagram were not annotated, one might thought it to refer to some SCSI or SAS drive with all those characteristically shaped graphs indicative of read request reordering and aggressive deferred writing. It seems a small miracle that Crucial with Micron managed to achieve that from flash memory.
The second diagram suggests that we must give up using Marvell controllers. SSDs behave awfully when connected via them.
These SandForce-based devices behave in the same way, even the two Vertex 2 drives with different firmware.
The Colossus produces characteristically shaped graphs, too. This time around, the graphs are typical of Indilinx controllers.
As for the Intel X25-V, its behavior is highly specific, too. It features an excellent performance at reading (especially for this, low-capacity model) but has only mediocre results at writing.
The RevoDrive is similar to the SandForce-based products in this test but copes with mixed loads better. It slows down when there are only write requests to be processed.