4 ms·

Unfortunately, full-disk encryption absolutely kills SSD performance because it makes the data look random (i.e. incompressible). It will wear out the SSD much

Unfortunately, full-disk encryption absolutely kills SSD performance because it makes the data look random (i.e. incompressible). It will wear out the SSD much faster than using it without would, because the hardware compression unit in the controller can sometimes achieve 8:1, and therefore have to rewrite only 1/8th of the NAND cells that it otherwise would.

gojomo 15y ago
I think you're right on the effect but wrong on the implied grave magnitudes "absolutely kills" and "much faster", because: • not all SSDs even have hardware compression • modern workloads have less highly compressible data than in the past: large-media formats include their own compression, and bulk data processing often does its own application-level compress/decompress on store/load I'd be interested to see any benchmarks that quantify the speed/lifetime hit that whole-disk encryption might cause for SSDs, but my hunch is that the effect would be slight in normal scenarios.
- culturestate 15y ago
  Anand benchmarked FileVault(1) and his conclusion was "Overall the hit on pure I/O performance is in the 20 - 30% range. It's noticeable but not big enough to outweigh the benefits of full disk encryption." 1. http://www.anandtech.com/show/4485/back-to-the-mac-os-x-107-lion-review/18 http://www.anandtech.com/show/4485/back-to-the-mac-os-x-107-...
  - kgtm 15y ago
    Yes, but that result is most likely on a Sandforce-powered SSD, which sports different transfer speeds based on the payload (compressible/incompressible). Like the GP notes, this is not a problem with SSDs in general.
    - culturestate 15y ago
      That benchmark is with a stock Apple SSD, which IIRC are not Sandforce-based. To be clear, I'm actually in agreement with Anand and OP - the benefits of FDE far outweigh the consequential I/O hit.
conformal 15y ago
been running with FDE on an SSD for >3 years now, still works fine. sure it's slower but how often do most people really push their disk past 20 MB/s?

gojomo 15y ago

I think you're right on the effect but wrong on the implied grave magnitudes "absolutely kills" and "much faster", because: • not all SSDs even have hardware compression • modern workloads have less highly compressible data than in the past: large-media formats include their own compression, and bulk data processing often does its own application-level compress/decompress on store/load I'd be interested to see any benchmarks that quantify the speed/lifetime hit that whole-disk encryption might cause for SSDs, but my hunch is that the effect would be slight in normal scenarios.

culturestate 15y ago
Anand benchmarked FileVault(1) and his conclusion was "Overall the hit on pure I/O performance is in the 20 - 30% range. It's noticeable but not big enough to outweigh the benefits of full disk encryption." 1. http://www.anandtech.com/show/4485/back-to-the-mac-os-x-107-lion-review/18 http://www.anandtech.com/show/4485/back-to-the-mac-os-x-107-...
- kgtm 15y ago
  Yes, but that result is most likely on a Sandforce-powered SSD, which sports different transfer speeds based on the payload (compressible/incompressible). Like the GP notes, this is not a problem with SSDs in general.
  - culturestate 15y ago
    That benchmark is with a stock Apple SSD, which IIRC are not Sandforce-based. To be clear, I'm actually in agreement with Anand and OP - the benefits of FDE far outweigh the consequential I/O hit.

conformal 15y ago

been running with FDE on an SSD for >3 years now, still works fine. sure it's slower but how often do most people really push their disk past 20 MB/s?