3 ms·
Yes, but that's how TCP/IP (the Internet) has worked successfully for 40 years. Basic congestion control uses packet drops as a signal of congestion and slows
by swinglock 7y ago
Yes, but that's how TCP/IP (the Internet) has worked successfully for 40 years.
Basic congestion control uses packet drops as a signal of congestion and slows down. You get packet drops if a queue was filled somewhere, which would be the bottleneck between two stations. That's business as usual.
The best queue to fill is as close as possible to a station, i.e. in the user modem or first hop, if not the application itself. If we begin to get congestion between ISP networks it won't be as pretty but it won't "break the Internet" and it's literally their job to upgrade the connectivity between networks before that happens. Plenty of bandwidth means you don't need QoS.
That being said, there are modern congestion control algorithms that try to be more clever. Your web browser may already be using one of them.
- zozbot234 7y ago> Basic congestion control uses packet drops as a signal of congestion and slows down. You get packet drops if a queue was filled somewhere, which would be the bottleneck between two stations. Bufferbloat gets in the way of that. As does packet loss due to non-congestion reasons, e.g. noisy or unstable connections at the physical layer.
- the8472 7y agonetflix&co are probably using BBR to shovel data to end users which takes latency into account too, not just packet drops.
- swinglock 7y agoLarge queues without AQM is indeed problematic but the industry is far more aware of this nowadays than it used to be. If the EU really wanted to save the Internet they should have done something about that, only it's 20 years too late.