[ih] congestion control... solved?

[Jack Haverty]

On 2/13/23 17:48, Dave Taht wrote:
> Imagine ... someone managed to get the flow queuing packet scheduling mechanism ... into billions of machines, and many of the congested routers, on the internet, and for all we know, a ton of switches.
This is "the" key issue, IMHO.  Personally, I don't consider something 
"solved" until it is deployed into the field and the original problem 
goes away.   It's good to see that some science (math) has been 
developing in the networking problem space.   But new algorithms and 
protocols aren't a solution unless they are deployable.   I wrote a bit 
about this in RFC 722 after running into the issue in the ARPANET.   
Algorithms and mechanisms in a network must be designed to be not only 
functional but also deployable, ideally without the need to throw the 
existing systems away and start from scratch.

The people who were working on ARPANET internals back in the 80s were 
confident that they were making progress on mechanisms for flow control, 
congestion control, and other networking issues that surfaced as the 
network was used and grew.   Much of that work involved thinking about 
how any new mechanism would be introduced into the ARPANET in a 
non-disruptive way.   I suspect that the people working on other network 
technologies - SPX/IPX, Appletalk, SNA, XNS, DECNET, ISO, etc., had 
similar goals.

That was a hard problem.

IMHO, the Internet architecture makes the problem even more difficult.   
In the ARPANET, there were only hundreds of switches and lines, and the 
topology, operation, and software development was all managed by one 
organization (BBN).  In the Internet, the same mechanisms are spread so 
that parts are in the switches (IP) and other parts are in the endusers' 
computers (TCP), now billions of them, designed, built, deployed, and 
operated by many different organizations and people.

That's a really hard problem....

Jack