Mechanical vs. Solid State Drives, Or Walter's Multithreaded I/O Redux
Walter Bright's recent article about moving from a familiar mechanical hard-disk drive with its "motors and levers and gears and spinning things" to a newer solid-state drive (SSD) was fascinating for a lot of reasons, ranging from the D programming language and multithreading to the issue of adopting new technologies in general.
According to Walter, the SSD's performance improvements were immediate and dramatic. His laptop "boots up promptly, and applications load crisply. It's like a new machine." All of which led Walter to start thinking about how disk I/O slows things down.
It got me thinking too, and while Walter went on to examine how multithreading and parallel programing come into the mix, I started looking for information about how mechanical disks compare to their solid-state cousins. And that took me to a research paper entitled Comparing Performance of Solid State Devices and Mechanical Disks, by Milo Polte, Jiri Simsa, and Garth Gibson.
As the title suggests, the goal of the study was to examine the performance of several SSDs and relate their performance to that of mechanical disks. For the evaluation, they used the IOZone benchmark running in single-threaded mode with varying request size and access pattern on an ext3 filesystem.
If your common-sense assumption is that SSDs are faster than mechanical disk drives, then you're right -- at least according to the authors of the study. What they found was that for sequential access pattern the SSDs are up to 10 times faster for reads and up to 5 times faster than the disks. For random reads, the SSDs provide up to 200x performance advantage. For random writes the SSDs provide up to 135x performance advantage.
The authors also considered the performance numbers against the prices of the SSDs tested (ranging from $110 to $2400) and concluded that "SSDs are approaching price per performance of magnetic disks for sequential access patterns workloads and are superior technology to magnetic disks for random access patterns." That said, the also pointed out that there are open questions, such as integrating SSDs into storage stacks containing mechanical drives and the like.
All in all, a nice job with the paper and also nice to see that there is sometimes measurable basis for your common-sense assumptions.