Re: [homegate] HOMENET working group proposal

[ken carlberg <[e-mail]>]

Fred,

> I think that the need is felt in home, SOHO, and hotel (eg, broadband last mile) networks, and suggestions I have made along such lines have focused on them. However, the topic is closer to the charter of tsvwg IMHO, except in the ways that it manifests in broadband interfaces (the fact of the router, modem, CMTS/DSLAM, and upstream router being separate entities with separate buffers). The problem is in essence that ECN/AQM is generally not implemented in such networks, in part because there is a sense on the carrier side that loss is a bad thing, and in part because tuning the algorithms can be a pain. What the Georgia Tech folks have worked out, as much as anything, is an acceptable configuration driven by user perceptions rather than research optimizations (eg, for them it's not about finding the absolute knee of the curve and making the combined queue be as thin as possible, but finding a pragmatic solution that is predictable and acceptable to the typical user).

given that a lot of these boxes are linux-based, is it a case of not being implemented, or simply not being turned on in the shipped image?  And if the latter, then perhaps Homegate/Homenet offers us a means of configuring ECN/AQM.

And I agree, the algorithms would be something that is advanced in TSVWG.  And we're also in agreement that the home/SOHO/hotel segment is just one part of where bufferbloat can manifest itself.

-ken