One slightly dubious thought. Andrews and Arnold, my ISP, has a line monitoring display in its clueless.aa.net.uk server and this shows upstream and downstream traffic rates amongst other things. Since switching over to the ZyXEL modems from the DLink DSL-320B-Z1 devices I was using before I see a surprising difference in the downstream rate graphs. The downstream rate line is absolutely as fast as a pancake, solidly locked on to 100% of the maximum with no tiny droops or little glitches in throughput. I wonder, is it possible that with the old modems I was seeing a sign of the effects of dropped corrupt packets and hence tiny gaps because of delays before L4 retransmissions timers expired? Or is that just my imagination glorifying the new modems ?
If the link is really busy anyway, because there are multiple TCP streams running in parallel, perhaps you would never see any gaps in that case anyway. So if that is the case then maybe I am just fooling myself. I am looking at a session where I am downloading a whole load of TV episodes from Netflix. The display suggests that four episodes are being downloaded in parallel. Of course there is no reason why this should mean that there is a one-to-one correspondence between episodes and files and TCP connections. Indeed, they could simply split up the content of one single episode and send it split over n TCP connections if they want to try and push the link harder because they regard TCP performance as sub-optimal. So would that just cover any gap up anyway? I don’t know how long the timeout before retx is anyway in the case of a packet dropped due to corruption.
But if there really are visible gaps, and the gaps are due to corruption-dropped packets, then this would have arisen because the DLinks were running so very close to the limit before, with SNRM downstream of 0.6-0.8 dB sometimes. Also there was no support for any L2 retransmission protocol available in the DLinks whereas now I enjoy Broadcom PhyR so there should be no corruption.