Interesting. One other thing they don't mention that I found interesting when doing my own digging on dmcrypt speeds a while back is that the 'cryptosetup benc

Interesting. One other thing they don't mention that I found interesting when doing my own digging on dmcrypt speeds a while back is that the 'cryptosetup benchmark' command is only showing the single core performance of each of those encryption algorithms. You can verify this by watching the processor load as it performs the benchmark. That lead me to find that if you have a Linux software RAID you can get much better performance by having 1 dmcrpyt volume per disk and then software RAID the dm devices instead of putting a single dmcrypt on top of the software RAID. Curious if that would stack performance wise with what they found here or if that just happened to help with the queuing issue they identified.