I'm simulating CUBIC using TCP Linux and tracking cwnd and sending rate but the results are somewhat puzzling.
I would expect the sending rate to equal cwnd * MSS / delay but it doesn't. It appears to be correct in itself (~90% of the link bandwidth), but the formula above suggest a rate about 3.5 times higher for all my experiments, with varying bandwidth, number of TCP flows and delay. What am I missing? -- Jasin