Re: [aqm] Question re draft-baker-aqm-recommendations recomendation #2

[John Leslie <[e-mail]>]

Wesley Eddy <[e-mail]> wrote:
> On 4/30/2013 6:17 PM, Linda Dunbar wrote:
>> 
>> Do you mean that downstream nodes treat packets with ECN bits marked
>> differently than that packets whose ECN bits are not marked? 
>> In another words, is "congestion experienced (CE)" encoded in the packets? 
> 
>... all that it's saying is that the congested queue set the CE bit
> (when ECT is set), per normal ECN protocol. It isn't saying anything
> new about the downstream nodes that see CE has already been set.

   A downstream node _could_ treat the packet differently: it wouldn't
matter much for what we're interested in.

   Note that a downstream node could reasonably drop a packet already
CE-marked if that node _also_ is congested. It wouldn't actually help
IMHO if the flow containing the marked packet is well-behaved; but
there's really no way to tell whether the flow is well-behaved.

   With room for doubt, it's arguably right to drop it at an even
lighter level of congestion: in all reasonable cases of two or more
nodes congested on the path, we should expect another packet to come
along _not_ already CE-marked.

   (I'm not actually proposing this -- least of all proposing to
_standardize_ this; but perhaps it will help folks to think through
the issue.)

   In some future "perfect world", where CE can mark packets _before_
there is any reason to drop any, I might talk about standardizing
such a practice. ;^)

--
John Leslie <[e-mail]>