On Tue, Nov 28, 2017 at 9:17 PM, Mikael Abrahamsson <[email protected]> wrote:
> On Wed, 8 Nov 2017, Toerless Eckert wrote: > > I am primarily thinking that there could be a higher demand for >> TCP (end-to-end) retransmissions when using WiFi because the L2/WiFi >> local retransmissions are insufficient. And if so, what the >> characteristics >> > > I don't know of any work you're asking for, but from my own experience > radio networks (wifi + cellular) has the following (general) > characteristics: > > They deliver packets in-order per host. > My experience is different. It is more similar to this post https://www.ietf.org/mail-archive/web/rmcat/current/msg00815.html > > The try very very hard to deliver all packets (true for unicast only on > wifi), using L1/L2 retransmits. > > Because of this, they sometimes "stall" shorter or longer times, so you > might get latency spikes of hundreds of milliseconds that are gone in the > next second. My personal record is 180 SECONDS of RTT on a 2G network. > > These latency spikes might cause TCP to believe there was packet loss and > cause retransmits, where there instead "only" was packet delay. > > Radio networks have airtime schedulers, so a single packet and a train of > packets might have very different network experience. Some network types > send multiple packets in a single transmit opportunity, and these transmit > opportunities might be tens of milliseconds apart. > > -- > Mikael Abrahamsson email: [email protected] > >
