fe31nz:
So no-one seems to have ever done much thinking about a better system than TCP already uses, because there is no point for them.
https://en.wikipedia.org/wiki/TCP_congestion_control#Algorithms
![]() ![]() ![]() |
|
fe31nz:
So no-one seems to have ever done much thinking about a better system than TCP already uses, because there is no point for them.
https://en.wikipedia.org/wiki/TCP_congestion_control#Algorithms
--
Please note all comments are the product of my own brain and don't necessarily represent the position or opinions of my employer, previous employers, colleagues, friends or pets.
dzh:
fe31nz:
So no-one seems to have ever done much thinking about a better system than TCP already uses, because there is no point for them.
https://en.wikipedia.org/wiki/TCP_congestion_control#Algorithms
Congestion is not the problem being discussed, although it obviously can have a big effect on download and upload speeds. The problem is the TCP window size and getting it ramped up quickly enough so that your TCP connection is not spending most of its time waiting for the next ACK packet before any more data gets sent.
Talkiet:
https://en.wikipedia.org/wiki/QUIC
This looks interesting as it avoids using TCP and does its own protocol, which is pretty much what is required in order to avoid the TCP window problem. But on first glance, it appears to be using multiple data streams, which should not be necessary if you do not have the TCP window problem to contend with. QUIC appears to be using multiple data streams to avoid slowdowns from packet loss. Since packet loss is fundamental to how IP routers handle congestion, any good protocol has to pay attention to how to handle packet loss well. But I am not sure that just having one data stream slow down due to packet loss and allowing other data streams to speed up to compensate is the right way to go about it - your steams might end up contending with each other at the overloaded router(s). In the specific case of the OP, when dealing with monster size files, best speed should be achievable with one stream and complex packet loss and retransmission control, as the requirement is for the entire monster file to arrive intact as fast as possible, rather than to have a stream of data arriving in a form where the data that has arrived can be used as it arrives (eg streaming TV).
fe31nz:
Congestion is not the problem being discussed, although it obviously can have a big effect on download and upload speeds. The problem is the TCP window size and getting it ramped up quickly enough so that your TCP connection is not spending most of its time waiting for the next ACK packet before any more data gets sent.
I'm far from expert here, but isn't Slow Start part of congestion protocol?
That almost seem like a key point citing marketing slide of Google's BBR
linux and windows 11 2h22 can apparently enable BBR
Jase2985:
linux and windows 11 2h22 can apparently enable BBR
I have been running BBR on my Ubuntu systems for several years. I did not realise Windows 11 now could do it also - I will have to enable it there too.
and
|
![]() ![]() ![]() |