WORTH READING more on worth reading "A Piece of the Action"

[David Farber <dave () farber net>]

Begin forwarded message:

From: Louis Mamakos <louie () transsys com>
Date: January 22, 2006 8:58:21 PM EST
To: dave () farber net
Subject: Re: [IP] more on worth reading "A Piece of the Action"

Dave,

As one of the perpetrator of PPPoE, I can say that "I was there" when  
the protocol was conceived, and I can relate the reasons why.  What  
has happened subsequently, I take no credit/blame for.

At the time, I was at UUNET (er, MFS, not thats not right, WorldCom).  
Well anyway, we though of ourselves as "UUNET" at the time before the  
dark times came, but that's another story.  I was working on a design  
for a DSL product offering we wanted to roll out.

At the time, DSL for Internet was very much a new thing.  Many  
vendors where out there and many of them were trying to "add value"  
by making their DSLAM and associated CPE device little baby brain- 
damaged routers, and "IP aware".  They mostly all did a pretty poor  
job of it and it created more problems than it solved.  In  
particular, having DSLAM be an IP router means that your customers  
all have related IP addresses, or at least routing policy and this  
makes competitive equal access to the underlying DSL infrastructure  
impossible.

Every wonder why there's little to no competition in choosing what  
ISP you want to use on your cable plant?  This is why -- it's because  
the CATV IP service is tightly bound with the transport over the  
medium.  We didn't want that to happen again, and we really wanted  
the DSL plant to know NOTHING about IP at all.  Incidentally, this is  
what will allow some DSL ISPs to offer IPv6 to their customers  
sooner; the DSL plant doesn't even know.

Anyway, so we were casting about trying to figure out how to make  
this work.  Of the brain-damaged hardware we knew was deployed, and  
there was a lot of it, pretty much all of it could be lobotomized and  
just operate as in a dumb, simple, Ethernet bridge mode.  That met  
our need for L3 protocol insensitivity and enabled competitive access.

We wanted to enable competitive and wholesale opportunities for ISP  
access here as well; UUNET had been tremendously successful building  
a wholesale V.90 dial-up access network, and we wanted to enable that  
same capability here.  The key to making this work was enable the  
ability to delegate the authentication for the customer wanting  
access; with PPP we did this at UUNET by inventing the RADIUS proxy  
and forwarding the authentication requests to our third-party  
wholesale customers by a simple syntactic examination of the  
principle name being authenticated (e.g., UU/louie or louie () UU NET or  
louie () MSN COM.  We never did UU!louie which would have been a great  
inside joke!)

So, you could have competitive access at:

- the wire loop (being your own DSLAM)
- the logical connection to the CPE device (use their DSLAM, buy some  
ATM or other backhaul)
- the customer (use someone else's router/session termination, bring  
your own email, NNTP, news, etc.)

We then just looked for the obvious way of running PPP sessions over
Ethernet; figuring that PPP ran over every other damn thing.  But  
remarkably, there was no such thing, so we did an 80% solution in 20%  
of the time and it appears to have addressed most of the  
requirements. Ideally, we would have used something simple like L2TP  
over Ethernet, but that didn't exist either, and it would have been  
18 months getting through the working groups.  I expected PPPoE to be  
a proof-of-concept, with the L2TP crowd pickup up from there, but it  
never happened.

Further, because of the way that PPPoE worked, you could share the  
DSL span with multiple end-systems in the home, each connecting to a  
different PPPoE session aggregator because of the discovery process.   
We wanted to enable an AOL-like residential ISP on one PC and a very  
different connection (work-at-home or something) on another PC, each  
with their own PPPoE session.  You could even connect multiple DSL  
CPE devices, each with their OWN DSL span back to different PPPoE  
session aggregators and have multiple sessions operable at the same  
time using a single LAN at the customer premise.  Of course, none of  
this cool flexibility ever got used.  Oh well.

Skip forward 5 or 6 years, and what did we learn:

- our believe that Path MTU discovery would be workable in the face  
of the 1492 byte MTU failed due to zealous filtering of ICMP traffic

- not all of the competitive opportunities to hook into the DSL  
infrastructure were taken advantage of.  There are lots of reasons  
for this, almost none of them technical

Control or choice?  I guess it is a matter of perspective.  When we  
set out to do this, it was to enable options that were not (and are  
still not) available on Cable TV-based Internet services.  And those  
limitations probably were not by accident, if I had to guess.

Today, I believe that cable operators have better control over  
bandwidth, especially upstream bandwidth.  If you think about the MAC  
layer used on a cable TV plant for broadband data, the upstream  
channel cannot be contention based; the CPE device is assigned  
timeslots and thus a fixed upper bound on the bandwidth available.   
Certainly there are session termination devices that can apply  
traffic policers and shapers and other policies to traffic that would  
apply equally to both cable and DSL.  Of course, it's been a few  
years since I've crawled around in that world of building networks so  
the details may have changed.

Louis Mamakos
UUNET alumni



David Farber wrote:
> Begin forwarded message:
> From: Jeff Schult <jss () tftb com>
> Date: January 17, 2006 3:22:07 PM EST
> To: dave () farber net
> Subject: Re: [IP] more on worth reading "A Piece of the  
> Action" (was:  Charging "content providers" ...)
> Reply-To: Jeff Schult <jss () tftb com>
> Dear Dave,
> My own recollection from having worked at a telco or two on the   
> Internet side of things is that they had their eye on multiple  
> tiers  of service almost from the moment that DSL became a  
> potential product  -- and that this was a factor in the telcos all  
> (curiously?) opting  to use and specify PPPoE rather than the  
> simpler cableco DHCP  distribution.
> Someone closer to the fray than I am now could perhaps shed more   
> light on this, but I suspect that the telcos have more control  
> over  the deliver of metered bandwidth to individual customers than  
> do  cablecos?
> -jeff schult
> -------------------------------------
> You are subscribed as louie () transsys com
> To manage your subscription, go to
>  http://v2.listbox.com/member/?listname=ip
> Archives at: http://www.interesting-people.org/archives/interesting- 
> people/


-------------------------------------
You are subscribed as lists-ip () insecure org
To manage your subscription, go to
 http://v2.listbox.com/member/?listname=ip

Archives at: http://www.interesting-people.org/archives/interesting-people/