On 03/26/2019 10:01 AM, Alexei Starovoitov wrote:

The next tcp congestion control algorithm should be implementable in bpf.

Congestion control do not need per-packet decision, and can indeed be
implemented in bpf.

Some researchers even have implemented them in user space (CCP if I remember well)

If kernel was extensible to that degree likely there would have been
no need to bypass it and invent quic.

QUIC is a complete re-design, and works around fundamental things
that TCP can not change.

bpf in TCP could not have changed the rise of QUIC or any other
modern transport, really.

Lets stop this discussion, we are digressing :)