Your first point is correct - papers ideally lead to innovation and tangible software improvements.
I think a kernel implementation of QUIC is the next logical step. A context switch to decrypt a packet header and send control traffic is just dumb. That's the kernel's job.
Userspace network stacks have never been a good idea. QUIC is no different.
(edit: Xin Long already has started a kernel implementation, see elsewhere on this page)
I think a kernel implementation of QUIC is the next logical step. A context switch to decrypt a packet header and send control traffic is just dumb. That's the kernel's job.
Userspace network stacks have never been a good idea. QUIC is no different.
(edit: Xin Long already has started a kernel implementation, see elsewhere on this page)