[tor-talk] (no subject)
coderman at gmail.com
Tue Aug 11 23:32:49 UTC 2015
On 8/11/15, Roman Mamedov <rm at romanrm.net> wrote:
> *Repeatedly headbangs on the desk*
> Uhm so what was I talking about. Ah yes, I believe that's not the case. It
> would add a great deal of benefit actually.
it would be useful, particularly on systems with native acceleration
of supported crypto primitives.
> As to why it's not implemented, I think simply because no one has coded it
correct. in particular, making the Tor internals efficiently
multi-threaded is difficult due to the particulars of crypto and Tor
network I/O in the implementation.
i thought i had a list of Trac tickets to the gist of this matter,
alas i cannot find them. perhaps someone else has a convenient collection?
this also came up in context of using CUDA or OpenCL to accelerate
network crypto via CPU offload to GPU.
the good news is that it is maybe less hard now, than it was some
years ago, to make this transition to well threaded internals in Tor.
maybe soon, even closer yet. and as you mention, patches welcome since
the best fix is code under test :)
More information about the tor-talk