Exploring The Performance Of A Full SandForce-Based SSD
One of the reasons we brought back fresh out-of-box testing in Iometer was to free up time to explore other aspects of performance. For this round-up, we specifically want to investigate the read performance of a full drive. After all, a 60 GB drive isn't very large, so it's very likely that you'll be dealing with this exact scenario on and off.
In theory, read speeds shouldn't be significantly impacted by the drive's free space. But there is a noticeable drop-off on the lower-capacity SF-22xx-based SSDs. Compression is one reason for this. Compressed data has to be decompressed, which requires sufficient free space. SandForce's garbage collection mechanism is a second explanation. Because most of its clean-up happens in the foreground, decompression requires a rotation of scratch space so that the NAND wears out evenly. This is known as wear leveling.
Both factors penalize the 60 GB drives we're rounding up more than the larger models as a result of limited NAND bandwidth, a side-effect of leaning on fewer NAND dies. We're using HD Tune to illustrate.
The image above shows how read performance varies. After using Acronis True Image Home to clone our system drive, we set HD Tune to read the entire LBA space sequentially. As you can see, reading back information written to our 60 GB Agility 3 falls between 150-200 MB/s. Once we move on to free space, performance accelerates to 450 MB/s. That's a big difference.
But what happens when we get more specific and narrow our focus to incompressible data? After adding about 6 GB of MP3s and H.264-encoded video, we can clearly discern performance at its worst on a SandForce-based drive. Read performance drops to about 150 MB/s. That's why you see performance bounce up and down when we only have Windows and a few apps installed: some of that data is compressible, and some isn't.
That's not the end of the story, though. We repeated this same tests on all of the drives in our round-up and found that the results only apply to 60 GB SF-22xx-based SSDs accessing asynchronous NAND. Let's take Intel's SSD 520 for a spin to compare the performance of a drive with higher-performing synchronous flash.
We already established (using early Iometer testing) that reading back free space is similar regardless of a drive's NAND interface. Yet, comparing the SSD 520 to OCZ's Agility 3, we see read performance in the space occupied by Windows and our application data falls between 250-300 MB/s. In a worst-case scenario, incompressible data is read back at 250 MB/s. Sure, that's low compared to the 450 MB/s we see otherwise. But it's also 100 MB/s faster than the 60 GB Agility 3 and its asynchronous NAND-based competition.
Now, to put it all in context. Check out the results from our 64 GB m4 in the chart below. Crucial mostly employs background garbage collection, and it doesn't lean on compression for its commendable performance. As a result, read performance is largely unaffected by fill state.
