SSD Performance With Random Data
Many types of data are compressible, but compressed image files, zip files, video files, etc are generally not compressible. Also, Mac OS X system files and application are compressed already and therefore incompressible. Therefore, performance of an SSD with incompressible random data is a relevant metric for some users.
Several things should be kept in mind regarding performance for random data:
- A drive that does not compress data will write more flash memory on average, and is thus likely to consume more power and wear out the flash memory faster.
- In general, reads are far more important than writes, since most data is read many more times than it is written (think image files, sound files, video, etc that is written once only, then played repeatedly over time).
- A drive that has to read less data (because it is compressed) will likely consume less power.
- File-copy speed, scratch file speed, etc could all be affected.
Using DiskTester, an 8GB file was written (in 8MB chunks) containing an incompressible stream of random data:
disktester create-files --num-files 1 --file-size 8G --fill random Drive
The OWC and Apple drives read at similar speeds, but the OWC Mercury Extreme Pro (Sandforce controller) suffers badly on writes, dropping to about 40% of its usual speed for compressible data, down to hard drive speeds.
The fact that the Apple SSD does not slow down on writes suggests that it does not compress data for writes. But this makes the Apple SSD twice as fast for writes as the OWC SSD, keeping in mind that this is not a common situation to have pure random incompressible data.
Yet for perspective, if the Apple SSD is twice as fast for writes as measured with random data, the Photoshop large file comparison shows that it is only about 11% faster for that particular intensive test, suggesting that for most everything else, the OWC Mercury Extreme Pro would win out by dint of 30% higher speed for more common situations. Expect the Sandforce version 2 controller to make all this a moot issue, with double the speed of the drive tested here.