As we noted in our 60/64 GB SSD round-up, we're changing the way we test using Iometer. In the past, we tested random transfers by benchmarking with four workers active in Iometer. This mimics threaded environment, as each worker represents a different application accessing the drive. However, that approach is also overoptimistic because four workers with one outstanding I/O operation each actually translates into a queue depth of four. In order to make things more realistic, we cut the number of workers to one, actually yielding a queue depth of one.
We made two additional tweaks to testing, better tuning the suite for our purposes here today. You see, the average desktop workload experiences a lot of bursty data transfers. However, our prior methodology emphasized sustained performance. In trying to reflect increasingly real-world results, our Iometer scripts now run for one and a half minutes. That's a substantial reduction from our previous scripts that ran for five minutes. Additionally, we're testing queue depths from 0 to 31 in increments of three, since SATA's Native Command Queuing extension only supports that many outstanding commands.
Random Read Performance
At a queue depth of one, each SSD performs similarly on a 3 Gb/s or 6 Gb/s connection, except for the SandForce-based Vertex 3.
As you hit seven and sail past 10 outstanding commands, the 3 Gb/s standard constrains the entire field to the 200 MB/s.
Interestingly, Crucial's m4 is substantially slower than the competing 6 Gb/s-capable drives, though it does enjoy a performance boost attached to the faster interface.
For the other two drives (Samsung's 830 and OCZ's Vertex 3), SATA 6 Gb/s demonstrates its benefits at queue depths of four and higher. Once you hit seven outstanding commands, the interface allows these drives to deliver more throughput than a second-gen port would otherwise enable.
Random Write Performance
As you can see in the chart below, Samsung's 830, Crucial's m4, and OCZ's Vertex 3 all lose a little bit of performance attached to a 3 Gb/s interface, even at a queue depth of one, accounting for a roughly 9% drop. That's despite the fact that all of these benchmark results fall well under what even a first-gen SATA interface can accommodate.
Although Intel's SSD 320 doesn't impress in this test, the gap between it and other SSDs becomes much wider once you move to a higher queue depth. For example, at more than 10 outstanding I/Os, the 240 GB Vertex 3 and 256 GB m4 top out around 180 MB/s plugged in to a 3 Gb/s port. That's roughly 70 MB/s faster than Intel's SSD 320, which hits its limit at 110 MB/s.
Once you switch to SATA 6Gb/s, the 240 GB Vertex 3 and 256 GB m4 accelerate significantly. At queue depths higher than four, both drives push past the 200 MB/s barrier easily.
- Should You Feel Bad That Your Board Only Supports SATA 3Gb/s?
- Test Setup And Benchmarks
- Benchmark Results: Storage Bench v1.0 And PCMark 7
- Benchmark Results: 4 KB Random Performance
- Benchmark Results: 128 KB Sequential Performance
- SandForce: Performance With Incompressible Data
- Real-World Tests
- Buy The SSD You Can Afford, Not The Fastest One
- Storage Bench v1.0, In More Detail
- More Background On Our Benchmarks