[e2e] pre-BOF discussion: "ABR to the Internet"

[Michael Welzl]

Hi,

I'm CC'ing the list and hope you don't mind - I would like
to get some discussion going.


>Seems like it's a topic that's better handled by the 
>IRTF, especially since there are so many open questions. 
>IMHO, forming an IETF working group (assuming that 
>there's interest in this) would require a charter that's 
>focused on standardizing a specific solution once it 
>has been found.  Are there any very specific approaches 
>that have shown a lot of promise in test environments?  
>If so, it may be worth having a BoF to discuss these. 
>It's an interesting topic, though. 
>-Anoop 

I got some good results with PTP, but the way it was used
is very simplistic - and therefore the gain isn't that
impressive. The poster at
http://www.tk.uni-linz.ac.at/~michael/ptp/docs.html
has an example simulation trace - here, RAP only increases the
bandwidth if the available bandwidth trend from PTP is positve.
I have a very specific idea on how to make better use of the
information but just didn't make it in time    :(     I hope
I can get a tech-rep out in a week.

The SCI'01 multilevel ECN paper
http://www.tk.uni-linz.ac.at/~michael/abr2internet/durresi.pdf
also contains good results - okay, it's only one more bit there,
but I don't see why this can't be extended to Explicit Rate.

Then there are reasons at:
http://www.packet.cc/files/Ex-Rate-Fl-Con.htm

PTP is actually part of my Ph.D. thesis which IS a very specific
approach. It's end2end (not combined with DiffServ) and will use
a specific endpoint mechanism based on Content Type 4. The protocol
is described in draft-welzl-ptp-05.txt and available for ns and
Linux, both the router and end node part.

I could have restricted the scope by not putting so many open
questions on the page ... but I would like others to be able and
incorporate their work, too (see the papers on Explicit Rate signaling
in a DiffServ context at the web site).

Cheers,
Michael