Performance degradation after OpenSSL 3.0.14 uptake

[pruthig]

Hi Team,

We have upgraded our tool to use OpenSSL 3.0.14 instead of OpenSSL 3.0.13. As a part of our regular OpenSSL uptake process, we performed the performance testing on OpenSSL 3.0.14. We didn't expect any deviation, but, surprisingly, we observed degradation in 8 symmetric encryption algorithms we validated for performance. Following are the performance numbers for 1000000 iterations.

: <Time taken for 1000000 iterations>

OpenSSL 3.0.13

AES-256-CFB_OpenSSL_fips : 49213 ms
AES-256-GCM_OpenSSL : 1045 ms
AES-256-CBC_OpenSSL : 2122 ms
AES-256-CBC_OpenSSL_fips : 19083 ms
AES-128-GCM_OpenSSL : 952 ms
AES-128-GCM_OpenSSL_fips : 8602 ms
AES-128-CBC_OpenSSL : 1564 ms
AES-128-CBC_OpenSSL_fips : 14089 ms

OpenSSL 3.0.14

AES-256-CFB_OpenSSL_fips : 49696 ms
AES-256-GCM_OpenSSL : 1054 ms
AES-256-CBC_OpenSSL : 2125 ms
AES-256-CBC_OpenSSL_fips : 19153 ms
AES-128-GCM_OpenSSL : 959 ms
AES-128-GCM_OpenSSL_fips : 8661 ms
AES-128-CBC_OpenSSL : 1573 ms
AES-128-CBC_OpenSSL_fips : 14140 ms

Performance degradation

AES-256-CFB_OpenSSL_fips : -0.98%
AES-256-GCM_OpenSSL : -0.86%
AES-256-CBC_OpenSSL :-0.14%
AES-256-CBC_OpenSSL_fips : -0.37%
AES-128-GCM_OpenSSL : -0.74%
AES-128-GCM_OpenSSL_fips : -0.69%
AES-128-CBC_OpenSSL : -0.58%
AES-128-CBC_OpenSSL_fips : -0.36%

[t8m]

There do not seem to be any relevant changes between 3.0.13 and 3.0.14 that should affect the performance. The only one remotely related change is: 4654e94

How does your performance measurement work? Does it by any chance spawn a new process for each iteration? If so, that could explain the difference. It would also somewhat explain the otherwise unexplainable difference between the fips and non-fips measurements above.