EricssonResearch / scream

SCReAM - Mobile optimised congestion control algorithm
BSD 2-Clause "Simplified" License
174 stars 53 forks source link

Rather large difference between numbers shown #48

Open mirabilos opened 1 year ago

mirabilos commented 1 year ago

With the bandwidthtester receiver on an 100 Mbit/s ethernet link:

 summary  22.1  Transmit rate = 93584kbps, PLR =  0.00%( 0.00%), RTT = 0.009s, Queue delay = 0.025s
 summary  24.1  Transmit rate = 94336kbps, PLR =  0.00%( 0.00%), RTT = 0.008s, Queue delay = 0.024s
 summary  26.1  Transmit rate = 93545kbps, PLR =  0.00%( 0.00%), RTT = 0.012s, Queue delay = 0.033s

The -log CSV differs rather a lot though: the highest I get (shortly before ^Cing) is:

That’s about 10 Mbit/s less! Where does this difference come from?

Looking at the code, the “Transmit rate” command line output seems to be tracked completely independently of the values that actually end up in the CSV. This doesn’t make any sense to me… shouldn’t the bandwidth be dependent on how many bytes per second you can actually get through the link?

Isn’t ACKed bitrate (plus packet overhead) the better estimate?