Tests on a memory backed pool show performance improvements of ~15-22% for AES-CCM writes, ~17-20% AES-CCM reads, 34-36% AES-GCM writes and ~79-80% AES-GCM reads on a Sandybridge x86-64 CPU, so this looks like a promising optimization that will benefit a lot of users.
== The fix ==
Backport of upstream 31b160f0a6c673c8f926233af2ed6d5354808393 - this is already backported in Groovy ZFS 0.8.3-1ubuntu13
== Test case ==
Run ZFS performance tests from ubuntu_performance_zfs_encryption ubuntu kernel team autotests. With the fix the encryption runs significantly faster, as noted earlier in the SRU justification.
== Regression Potential ==
This fix alters the crypto engine and adds in new optimizations for CPUs that have capable instruction sets. There is a risk that this new crypto code is erroneous.
== SRU Justification ==
Upstream commit 31b160f0a6c673c 8f926233af2ed6d 5354808393 contains AES-GCM acceleration changes that significantly improve encrypted performance.
Tests on a memory backed pool show performance improvements of ~15-22% for AES-CCM writes, ~17-20% AES-CCM reads, 34-36% AES-GCM writes and ~79-80% AES-GCM reads on a Sandybridge x86-64 CPU, so this looks like a promising optimization that will benefit a lot of users.
== The fix ==
Backport of upstream 31b160f0a6c673c 8f926233af2ed6d 5354808393 - this is already backported in Groovy ZFS 0.8.3-1ubuntu13
== Test case ==
Run ZFS performance tests from ubuntu_ performance_ zfs_encryption ubuntu kernel team autotests. With the fix the encryption runs significantly faster, as noted earlier in the SRU justification.
== Regression Potential ==
This fix alters the crypto engine and adds in new optimizations for CPUs that have capable instruction sets. There is a risk that this new crypto code is erroneous.