Here’s the deal: A number of print and Web publications run benchmarks on solid-state drives just like they do on standard, magnetic storage. This typically entails firing up a few synthetic benchmark applications–programs that offer speed measurements for a drive, but do so in a fashion that isn’t very correlated to real-world use–alongside a few closer-to-real-world benchmarks like PCMark, Iometer, or some kind of measurement from actual applications and games.
These tests are frequently done on blanked (if not unpartitioned) drives, but measured across the whole of the drive as much as possible. By that, I mean that a number of the synthetic benchmarks concoct average speed scores for the hard drive’s various operations by measuring its performance from a number of different locations across the drive’s spinning platters. Reviewers can’t just slap a blank drive in a system, load up a copy of an operating system, and run tests in the OS to simulate the drive’s real-world use. They would have no way of controlling the exact methods by which the operating system and hard drive store data after the copy, resulting in incomplete points of comparison against other products–or a lack of apples-to-apples testing.
This isn’t a critique of testing methods. But it’s important to understand this back-story a little bit, because these same methods are frequently used to test solid-state drives as well. The problem with that is that a solid-state drive can show different performance depending on how many of its flash cells are being used. According to AnandTech, a drive that stores some element of data across all of its blocks–the smallest part of a solid-state drive’s flash memory that can be written at any given time–will suffer drops in both its read and write performance.
We’ll say that again: Once you’ve filled your SSD with enough data, performance chokes.
How bad does it get? Depending on the drive, you could see drops of tens of megabytes per second in reads to over a third of your drive’s available write performance. This translates into real-world performance losses of anywhere from three to fifteen percent–at least, based on the particular batch of drives AnandTech tested. The site still found that its tested SSDs outperformed the best of the magnetic storage offerings, mostly due to their speedier random-access capabilities. That, and the fact that the affected read speeds of solid-state drives are still beyond those of conventional magnetic storage.
As for the takeaway, AnandTech’s results bring forth two points to consider. First, the Declaration of SSD Independence: not all solid-state drives are created equal. This should be obvious, but it’s worth emphasizing for those new to this storage spectrum. Drives can be optimized for low random access times, high bandwidth, or a combination of both. AnandTech found that those in the first category, like Intel’s X-25M and X-25E line of SSDs, tended to fare much better in post-slowdown performance than SSDs optimized for high transfer rates.
Second, AnandTech’s findings cement the need for increased accuracy in solid-state reviews. What good is a benchmark result that will vary in wild and dramatic fashions after you’ve used the drive for a normal period of time? It would be uncouth to point fingers. But you shouldn’t trust the results of any reviewer who only runs a cursory series of benchmark runs on an empty drive. Until this SSD slowdown is corrected on a hardware or operating system level, there’s just too much of a chance that reported results won’t reflect your real-world performance. And given how much you’ll be spending for that brand-new SSD, you really should pick up the best device you can afford to buy.
Follow us on Twitter @geektech for your daily fill of tips, hacks, hardware, and even more nerdery.