Transport protocols and bulk file transfer

[Larry Dunn <>]

Scott,
Having quickly read the paper,
it's "another queue management algorithm" (obvious).
To be embedded at router/path queues.
As such, a little outside of what we can "build".
And can't "test" much until it shows up in gear we use.

Though could certainly put medium-speed versions (Linux, etc.)
together for testing, etc.

I'll have to ponder it a bit more- seems potentially aggressive
with dropping, as it measures for min-standing-queue,
and goes into drop-mode above that value.

Also, RED looked great on paper and linearly simulations.
Wasn't until a couple/few years after people started working with
it that its multi-parameter fragility was documented.

Though this is described as a "parameterless" algorithm,
there were definitely choices made (target queue size, re-baselining frequecy,
transition from drop- to not-drop, etc.)
that may or may not stand up over time...
For example the plots showed behavior/improvement for low-speed
connections. Will the dynamics hold up at high bitrates? Don't know...
Most folks characterize the "buffer bloat" issue as it relates to
aggregated home-speed access, often with buffers in DSLAMs
getting our of control (for small number of users).
So not sure if that translates to 10Gpbs (though the atricle suggests it 
does...)

As you know, I'd never discount work from Van/Kathie...

Larry
--

On May 8, 2012, at 7:47 AM, Scott Brim wrote:

> Kathie Nichols and Van Jacobson have a new AQM algorithm out:
> <http://queue.acm.org/detail.cfm?id=2209336>
> 
> Thoughts please.  Could this be adapted to networks like ours?
> 
> p.s. Van's related slides from 2006 are fun:
> <http://www.pollere.net/Pdfdocs/QrantJul06.pdf>
>