On 28/05/2020 16:50, Matthew Wright wrote:
> I was literally just typing the same response, in order to stay under
> the Long Fat Network (LFN) definition you need to be under
> 100microSeconds RTT, which unless the line is sub 15Km even without
> hardware switching and processing times. Then you’re into TCP tuning etc
> to try and improve that.
>
> Your microwave link will likely (I assume) be within that kind of
> distance which means with the same TCP scaling you’re seeing higher
> throughput
>
> UDP should prove this as you don’t require the ACK and the data can just
> be fired down the line fire-and-forget style
That's what I was alluding to in the links I sent in the first reply to
the OP.
A 10ms RTT (which is feasible if the circuit is of reasonable distance)
needs a TCP Window in excess of 1MB to get close to the limit (excluding
overheads) of 950 Mbps.
Where's PeterG when we need him?
Ray
