Re: [Bloat] Seen in passing: mention of Valve's networking scheme and RFC 5348

[David Collier-Brown <davec-b@rogers.com>]

[-- Attachment #1: Type: text/plain, Size: 1847 bytes --]

The phenomenon is called "lava flow", and is a classic anti-pattern 
illustrated at http://antipatterns.com/lavaflow.htm
Their approach to fixing is ancient, though: there are 
correctness-preserving refactorings for some of the problem space.

Alas, I don't know if middleboxes are correctable... maybe if they are 
ones which only care about the IP layer?

--dave

On 04/04/18 08:45 AM, Jesper Louis Andersen wrote:
> On Tue, Apr 3, 2018 at 5:04 PM Jim Gettys <jg@freedesktop.org 
> <mailto:jg@freedesktop.org>> wrote:
>
>     ​To get to really good RTT's (with low jitter), you still need
>     ​fq_codel (or similar).  You just can't get there by hacking TCP
>     no matter how hard you try...
>
>
> I agree with you on all points here. However, any change which patches 
> an existing bad system is far more likely to win in the long run, also 
> if it is bad in some way. Momentum is a killer of good solutions. I 
> wish I had a ramification for this observation, but I currently don't.
>
> My hunch is that every new generation of young programmers wants to 
> put their mark on the system. As a result, they take what worked well 
> on level N-1 and proceed to build N on top of it. But the beanstalk 
> never withers, so each level is present in said stack, still, after 
> all these years.
>
> (Aside: The codel approach also has worked really well for me 
> internally in Erlang systems as a way to maintain queue load. Far 
> better than many other flow control schemes).
>
>
> _______________________________________________
> Bloat mailing list
> Bloat@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat


-- 
David Collier-Brown,         | Always do right. This will gratify
System Programmer and Author | some people and astonish the rest
davecb@spamcop.net           |                      -- Mark Twain


[-- Attachment #2: Type: text/html, Size: 3548 bytes --]