Split from bug 1584. We need to investigate if we can boost performance of OpenSSH on multi-core systems by multi-threading the AES CTR handling. One implementation of that is done here: http://www.psc.edu/networking/projects/hpn-ssh/ There are some hurdles though: - It bypasses OpenSSL's EVP, which means that we won't get AES-NI (see bug 3002) and similar things. We might lose more than we gain. - Modern OpenSSH prefers ChaCha20 over AES. - The server might want other ciphers or chaining methods, even though it is not OpenSSH's default.
If we want to use OpenMP, then bug 5615 needs to be fixed as well.
This is vague and HPN-SSH themselves admin that AES SNI is faster than their version. ssh isn't even close to being a performance bottleneck at the moment, so investigation bugs like these aren't productive.
