> > I haven't looked into why on the RPi plan9's tcp performance > > is about 30-40% of that on linux (which works near wire speed). > > For the local case it doesn't matter much in any case. > > (a) allocb() relies on deathly slow malloc; cf. qallocb in 9atom, which upps > performance quite a bit > (b) usb is not as fast, > (c) send and recieve in plan 9's tcp are not as decoupled as they could be, > this leads to latency in sending after the window opens, or latency in opening > the window.
we did get a lot of performance out of a proper NewReno implentation, which happened after the original rpi port. a review of changes in sources from the original work didn't turn up any noteworthy changes, but then again just reviewing the source code isn't all that effective. :-) - erik