If they are accepted, the parallel cryptography should be configurable. It's only a win on systems with kernel threads and, according to the documentation, it's only win when there is no hardware acceleration.
I can attest that the dynamic buffer sizing is a performance win. Some operating systems will need a setsockopt on SO_RCVBUF in the server code to take full advantage.
On 11/8/20, 2:20 PM, "openssh-unix-dev on behalf of Demi M. Obenour" <openssh-unix-dev-bounces+herbie.robinson=stratus.com@xxxxxxxxxxx on behalf of demiobenour@xxxxxxxxx> wrote:
Dear OpenSSH Developers:
What are your thoughts on the high-performance networking patches?
I understand that some of them (such as NULL encryption) will not
be accepted, but am more interested in the others, such as parallel
cryptography and dynamic buffer sizes.
These patches are used in production already, and I am wondering if
the OpenSSH developers would be interested in making them part of
upstream OpenSSH.
Sincerely,
Demi
_______________________________________________
openssh-unix-dev mailing list
openssh-unix-dev@xxxxxxxxxxx
https://lists.mindrot.org/mailman/listinfo/openssh-unix-dev
