Re: [re-ECN] Complete draft Charter for proposed CONEX WG

[John Leslie <[e-mail]>]

[e-mail] <[e-mail]> wrote:
> 
>> allow senders to inform the network of the level of congestion they
>> expect their packets to encounter. This information is currently only
>> visible at the transport layer. 
> 
> Although I agree with Mirja's comment, an argument that could be made to
> say the information is visible in the network if you use DPI equipment...

   That is an argument we need to rebut: Deep Packet Inspection is not
sufficient to show congestion management -- merely to infer reasonable
guesses from artifacts of congestion managment, which is not good
enough to act upon.

> However I think highlighting that only the end-systems are DESIGNED to
> see the info is a good thing...

   Even this is subtly wrong. There are many different congestion-control
schemes in current use. Most, but not all, are designed to infer packet
loss and interpret that as congestion. We are not talking about getting
every router along the path to have the same view of packet loss: we
are talking about getting every router along the way to have the same
view of expected congestion _whether_or_not_ any packets are ever
dropped.

> > With the output of CONEX, it will be
> > possible to provide sufficient information in each IP datagram so that
> > any node in the network can see the expected rest-of-path congestion.
> 
> I think I would prefer to say "The main output of CONEX will be a
> protocol that allows each IP datagram to carry sufficient information to
> allow any node in the network to see the whole path congestion."

   This is a good issue to raise.

   But I don't agree with Toby: any congestion _earlier_ in the path is
of no concern to a router -- there is nothing reasonable to _do_ about
it. Even if we don't actually design algorithms about what to do about
rest-of-path congestion, that remains the concern to consider. (Said
differently, there's no point in punishing congestion _after_ the last
bottleneck in the path.)

> I make the subtle distinction between rest of path and whole path so
> that we leave the door open for Matt Mathis's mark on drop suggestion...

   I'm happy to leave that door open, but I don't think Toby's suggestion
is necessary to do so.

--
John Leslie <[e-mail]>