On Tue, Oct 28, 2014 at 05:57:10PM -0700, Garrett D'Amore wrote: > It may be cheaper to throw a little more hardware at the problem and > parallelize than to try extraordinary measures like a user space tcp stack. I also agree that a user space TCP stack is an extraordinary measure. :D So right now I'm only using it on the raw input, and not on the nanomsg side where the volume is lower. But it would be nice to use nanomsg with the self-managed memory if I commit to not changing its content until nanomsg calls me back to free it post-TX. Matthew.