Re: [tsvwg] FQ & VPNs

[Jonathan Morton <[e-mail]>]

> On 20 Feb, 2021, at 1:26 am, Bob Briscoe <[e-mail]> wrote:
> 
> You are talking about instantaneous rate fairness, with no concept of occupancy over time. There's been so much more water under this bridge since the 1980s/90s. Try this: https://bobbriscoe.net/presents/1210isoc/1210isoc-briscoe.pdf
> 
> Clearly you never read/absorbed what I wrote for you (specifically for you more than anyone else) about this:
> Per-Flow Scheduling and the End-to-End Argument: https://bobbriscoe.net/pubs.html#FQ_v_E2E

I've read your papers already.  I happen to disagree vigorously with their conclusions, and I believe others with far more standing than myself also do so.

> I can tell 'cos:
> a) you somehow think I am arguing for RTT-fairness (when have I ever done that?) and

I'll quote you:  "…it would be preferable to deploy a FIFO AQM there."

If you pass AIMD traffic through a single FIFO with a single AQM instance, what you get is a (slow) convergence towards equal cwnd values and thus RTT-fairness.  That is not a controversial statement, but the truth.  And that is what you explicitly recommended just a few sentences before asking this question.

> b) the assertion that max-min fairness is the gold standard. If you had read anything from me, you would at least make a nod to my explanations about why max-min only makes any sense during a persistent famine (I mean famine for hours/days/weeks), and in all other scenarios max-min is the opposite of gold - it's the most rubbish monkey-metal crazy standard in a non-famine world.

That is precisely where I'm aware of your point of view, but diametrically opposed to it.

At this stage I can only recommend that you install Cake, turn on the host-and-flow fairness options, and just try it out for a while.  You might be pleasantly surprised by how well it works in practice.  In fact, until you have done so, I will simply not respond to any further posts by you on this subject.

 - Jonathan Morton