To what extent is size a factor in SSD performance?

Question

To what extent is the size of an SSD a factor in its performance?

In my mind, correct me if I'm wrong, a bigger SSD should be, everything else being equal, faster than a smaller one. A bigger SSD would have more erase blocks and thus more leeway for the FTL (flash translation layer) to do garbage collection optimization. Also there would be more time before TRIM became necessary. I see on Wikipedia that it remarks that "The performance of the SSD can scale with the number of parallel NAND flash chips used in the device" so it seems throughput also increases significantly. Also many SSDs contain internal caches of some sort and presumably those caches are larger for correspondingly large SSDs.

But supposing this effect exists, I would like a quantitative analysis. Does throughput increase linearly? How much is garbage collection impacted, if at all? Does latency stay the same? And so on. Would the performance of a 8 GB SSD be significantly different from, for example, an 80 GB SSD assuming both used high quality chips, controllers, etc?

Are there any resources (webpages, research papers, presentations, books, etc) that discuss correlations between SSD performance (4 KB random write speed, latency, maximum sequential throughput, etc) and size? I realize this does not really sound like a programming question but it is relevant for what I'm working on (using flash for caching hard drive data) which does involve programming.

If there is a better place to ask this question, eg a more hardware oriented site, what would that be? Something like the equivalent of stack overflow (or perhaps a forum) for in-depth questions on hardware interfaces, internals, etc would be appreciated.

Answer 1

Thanks James. It doesn't fully answer my question but there are some useful tidbits. To summarize some of the relevant parts of those talks:

SSD performance can increase by 10% or more as the "spare area" on the SSD increases.
Increases in spare area on MLC (vs SLC) is particularly good at increasing performance. Most of the SSDs which are reasonably big tend to use MLC, which has somewhat higher latency and lower write bandwidth, but comparable read bandwidth.

So drawing from this, it seems that a bigger SSD will probably have higher performance than a smaller SSD, unless the reason the smaller SSD is smaller is because it uses SLC or it has a higher percentage of spare.

This is mostly what I expected but it's nice to see confirmation.

Another useful but again not entirely comprehensive look comes from just scanning some of the tech specs on the web. For example I look a look at OCZ's page here on the Vertex Turbo. It seems that throughput does not come at all close to linearly increasing with drive size. It takes a jump from 30 GB to 120 GB to see a significant difference, and even then mainly for write speed not read.

Answer 2

Check out MEMS002 and MEMS003 from Intel's Developer Forum found here: https://intel.wingateweb.com/us09/scheduler/catalog/catalog.jsp

Answer 3

Storage Review has an excellent review of 6 Corsair Force Series GT drives in different sizes (from 60GB to 480GB). The summary is that larger drives in the family perform better in most scenarios (but not all). 240 GB seems to be the sweet spot (at least for the SandForce controller).

Answer 4

Anandtech discusses this briefly, based on an investigation by IBM Zurich (PDF). I think the overall conclusion is that there should be 10-30% of spare area on an SSD to maintain a decent write amplification ratio. Most drives include a reserved quota of about 7% (and enterprise drives will offer more); it is up to the user (and the TRIM command) to keep extra space free to maintain full performance.