2 ms·
Often these “revolutionary” changed get deconstructed and co-opted. Not always a bad thing. Looking through a video he did about this paper, I see a few things
by hinkley 4y ago
Often these “revolutionary” changed get deconstructed and co-opted. Not always a bad thing. Looking through a video he did about this paper, I see a few things that could be popped out.
SRPT (shortest remaining processing time) is there any reason this couldn’t be implemented as a LAN protocol?
Receiver driven congestion control: isn’t this what the transmit window does? Are we not just talking about setting a more aggressive starting value?
“sends packets in any order” and “can only send the first few packets without a grant (ack)” are fighting each other. Especially if you’re building a message oriented protocol, which tends to have shorter conversations. This reads as confused or schizophrenic.
I think of what us being said here is that the Berkeley socket protocol sucks, and that a different one can get from user space to response faster. Great. But do you have to change the wire protocol for that or just introduce a better system call library? This part in particular reads a lot like , “what if Erlang was right and we implemented it at the kernel level?” Which is not a bad question to ask.