Interesting People mailing list archives

Re: Interesting -- comment from author -- F.C.C. to Look at Complaints Comcast Interferes With Net - New York Times

From: David Farber <dave () farber net>
Date: Wed, 16 Jan 2008 02:05:16 -0800

________________________________________
From: Bob Frankston [Bob19-0501 () bobf frankston com]
Sent: Wednesday, January 16, 2008 12:18 AM
To: richard () bennett com; 'Bob Frankston'
Cc: David Farber; 'ip'; 'Lauren Weinstein'
Subject: RE: [IP] Interesting -- comment from author -- F.C.C. to Look at Complaints Comcast Interferes With Net - New 
York Times

Perhaps it's one of those things where you had to be there before there were networks and were just copper wires and 
radios. When Dave Farber was playing with token rings and Bob Metcalfe was putting the Aloha net on a Coax as his class 
project. TCP is a neat hack but is not at all fundamental. You should also look at the history of UUCP

You need to be careful about confusing end-to-end with womb-to-tomb. End-to-end means you don't assume anything but if 
you're lucky you'll get a few packets through and then can discover what works and drive the dynamic.

The network is not any more fundamental to networking than rails are to travel. The wired home network is copper -- 
it’s not a network as such. It's just a metal we use to simply exchanging packets and it's not even necessary as we 
become more adept at using radios.

An important point of end-to-end is that you don't serve the social needs in the network -- other than by facilitating 
people finding their own solution that is, essentially by definition, beyond your imagination. This is why I cite 
Minitel as an example -- it was farsighted but eventually the future veered off in a different direction. This is why 
adaptionism not only doesn't work in evolution but would frustrate the process by locking it into a local optima.

Perhaps reading http://www.frankston.com/?name=VONRailroads and http://www.frankston.com/?Name=OurCFR might be a start.

-----Original Message-----
From: Richard Bennett [mailto:richard () bennett com]
Sent: Tuesday, January 15, 2008 23:07
To: Bob Frankston
Cc: dave () farber net; 'ip'
Subject: Re: [IP] Interesting -- comment from author -- F.C.C. to Look at Complaints Comcast Interferes With Net - New 
York Times

Please see my comments in-line.

Bob Frankston wrote:

After writing this I Richardï¿½s letter that distinguishes between

ï¿½Internet Congestionï¿½ and ï¿½Fairnessï¿½ ï¿½ that certainly sounds like a

moral distinction rather than a technical distinction. Please define

fairness.


Network fairness is something like pornography, we can't always define

it precisely but we know it when we see it. In most of the fora where

this has been an issue, such as IEEE 802, we've tended toward a set of

definitions that would make Utilitatrians of the Bentham/Mill school

happy: "the greatest good for the greatest number." But we're typically

not that philosophical and simply true to prevent particular network

stations from starving others. This has been a concern of LAN designers

throughout the last 30 years of network engineering, Bob, and it's not a

"carrier-centric" concern. Ethernet, Token Ring, Token Bus, WiFi and UWB

all have fairness engineered into their media access control layers. No

network engineer worth his paycheck wants to deploy a system in which

one hungry station can effectively block network access for others,

either through accident or design.

A good engineer works within the limits given ï¿½ a very good engineer

questions the limits (and gets fired).


This is a very cute remark, and I liked it even better in your NN Squad

formulation, where you added "...because the constraints serve a policy

need higher than mere science or reality." I'll confess that science and

reality bias my perspective on network architecture. I have this

eccentric belief that networks that function well - meaning they're

fast, robust, and inexpensive - serve the social needs of liberal

democratic society better than those that don't function well. I haven't

verified this belief, and I am aware that it's not universal. The Amish,

for example, aren't at all into Ethernet or WiFi, although they have

apparently embraced cell phones. The argument that all good engineers

are unemployed is certainly novel, I have to admit.



And I do hope you're aware that some limits are imposed by laws of

physics that serve no particular political agenda.

I can understand that from a carrier-centric point of view there are

hogs that upset their resource allocation models. The question is not

whether the Internet needs a fairness system ï¿½ the question is whether

the Internet as an emergent property of the end-to-end constraint can

even define such a concept and who do we get along so well despite the

lack or, perhaps, because of the lack of an imposed notion of

fairness. This is indeed echo of calling modem users hogs.


This is a very disappointing remark, as you're resorting to name-calling

when we might be having an interesting discussion; it's not just

carriers who care about fairness, of course, but by invoking the devil

you're surely hoping to avoid serious discussion.



As the Internet being "an emergent property of the end-to-end

constraint," my goodness, there must be something more basic to it than

the TCP error-recovery and flow-control mechanism. Consraints, you see,

don't do anything unless there'a a prior reality to be constrained, and

that would be the mesh of routers, packets, and links that hold it all

together. You're going to have a hard time applying end-to-end

constraints if you don't have any packet delivery mechanism.



And whether we have modems or not, the essence of any packet-switched

network is the shared data link. This was the deal, remember, that made

it interesting to investigate the utility of packet-switching to begin

with: if we relax the hard allocation limits of circuit switching and

instead create a massive pool of bandwidth that we can all draw against,

might not there be some interesting implications? And certainly hogging

is one of them, just as bursting is another. Rationalizing the bandwidth

needs of diverse applications against a common pool of bandwidth is what

this whole Internet experiment is all about.

Iï¿½m responding with a caveat since a full response would be much

longer and I only want to touch at some top level issues which are

more policy than a particular algorithm. This is the

Malthusian/zero-sum issue. Malthus did argue that we have a finite

buffet and weï¿½re going starve if weï¿½re not very careful ï¿½ we must

allocate the scarceness rather than creating abundance. This doesnï¿½t

mean there arenï¿½t limits but we shouldnï¿½t hobble ourselves by focusing

only on the limits.


We network engineers certainly do care about limits, but I don't think

it's fair to accuse of being obsessed with them. Certainly, we make our

living by changing them. The Ethernet folks didn't just sit back and

collect royalty checks after the 2.94 Mb/s Experimental Ethernet was

installed at PARC, they went on a did the Blue Book, then the twisted

pair versions that were practical to install and manage, and the 100

Mb/s version, and so on. But network administrators certainly do have to

live within the limits of their budgets and purchasing options, this

being a material world and all.

The idea that a ï¿½hogï¿½ blocks others makes this sound like a 200 car

train that blocks a track ï¿½ if anything we have 200 independent cars

with others getting between them. The question is what opportunity or

percentage of slots each participant gets. You can have a local

algorithm to allocate a percentage of this capacity but that doesnï¿½t

scale to large networks.


Most of the recent complaining about network neutrality violations has

actually been layer two, private network concerns. The Comcast affair,

for example, arose (in my opinion) because of the comany's concern about

congestion in the upstream first-hop caused by a relatively small number

of BitTorrent users. Their application of a "local algorithm" got a lot

of people excited because they used some layer four protocols to

implement it. So we can't just pretend that nobody cares about "local

algorithms" when the evidence before us says otherwise.

What does it mean to ï¿½connect to an ISPï¿½? If a carrier allocates a

ratio of 1:50 or 1:10000 for capacity allocated to each subscriber vs

the capacity to the rest of the network (through a thin peering

orifice) then complains when some take more than their percentage ï¿½ is

that false advertising or naï¿½ve modeling? What makes this problem more

difficult is that it doesnï¿½t allow for local traffic to take advantage

of local abundance while backbone capacity is artificially limited to

create value through scarcity. As one drills down on this any policy

depends on an arbitrary definition of ï¿½fairnessï¿½ and a guess as to how

this interacts with the network as a whole?


It's certainly the case that ISPs - and I'm using the term to denote

"first mile network infrastructure providers" actually - have to model

expected patterns of usage to get the most of their networks. We have

asymmetric DSL and even more asymmetric DOCSIS because the usage

patterns that undergird the models are old and new symmetric

applications have shot them full of holes. The carrier response is

partly counter-measures such as RST spoofing, but it's also transition

to newer standards and better models, such as Comcast's announced plan

to roll out 100 megabit symmetric service.

Again, the real question is why we are focusing on scarcity? If a

fiber to the home is about $1k (and going down or far less if I share)

then we could finance that at $50/month and pay it back fairly

quickly. Instead we pay that amount for 1/1000^th capacity or less?

And then weï¿½re told weï¿½re hogs? What percent of the local physical

facilities are allocated to the carriersï¿½ own use. If Verizon can

delivery multiple VoD streams over a 20Mbps connection then why are we

fighting over a few megabits shared among users? Why arenï¿½t we asking

why we donï¿½t have more capacity per connection?


Some of us have asked those questions, and the general answer has been

(if you're paying attention) that upgrades are coming. In fact, all the

ISPs are more or less constantly adding capacity, so the question is

whether they're adding it fast enough to satisfy their customers and

slow enough to satisfy their investors. This being a capitalist world,

that sort of a balance is to be expected. Verizon has been a real

trailblazer in bringing high bandwidth to the home, and I wish more

telcos and cablecos were like them.

The question is not whether we get our fixed percentage but what we

can do with what we can get independent othersï¿½ use and what we can

come to expect.

For now the high order bit is scarcity created by policy ï¿½ it is not

scarcity created by so-called hogs. Unless youï¿½re willing to talk

about the biggest hogs ï¿½ carriers who think they own our ability to

communicate.

And I havenï¿½t even addressed the whole idea of containing packets to

partitioned paths nor the lack of wireless connectivity nor ï¿½


I appreciate your idealism, Bob, as I'm a child of the sixties and big

fan of Cheech and Chong. But we do have to be realistic on occasion,

even thought it's boring.



I hope this helps.



RB

-----Original Message-----

From: David Farber [mailto:dave () farber net]

Sent: Monday, January 14, 2008 18:21

To: ip

Subject: [IP] Interesting -- comment from author -- F.C.C. to Look at

Complaints Comcast Interferes With Net - New York Times

________________________________________

From: Richard Bennett [richard () bennett com]

Sent: Monday, January 14, 2008 4:23 PM

To: David Farber

Subject: Re: [IP] Re: F.C.C. to Look at Complaints Comcast Interferes

With Net - New York Times

As the author of the article in question, I'll gladly defend it. The

fundamental point I was trying to make is simply that there's a huge

hole in the architecture of the IETF protocol suite with respect to

fairness. I'm a layer two protocol designer (Ethernet over UTP, WiFi 11n

MSDU aggregation, and UWB DRP are in my portfolio), and in the course of

my career have devoted an embarrassing amount of time to engineering

fairness in network access. Most the younger generation takes it as

given that if you understand TCP/IP you understand networking, but in

fact most of the progress in network architectures over the last 30

years has been at layers 1 and 2. And with the TCP-centric mindset, they

tend to believe that all problems of networking can be solved by the

application of the right RFCs. But in fact we all connect to our ISP

over a layer 2 network, and each of these has its own challenges and

problems.

The carriers are often criticized for not using packet drop to resolve

fairness problems, but that's not really the scope of packet drop, which

is actually a solution to Internet congestion, not to the lack of

fairness that may (or may not) be the underlying cause of the

congestion. We need a different solution to fairness at layer 3,

especially on layer 2 networks like DOCSIS where packet drop closes the

door after the horse has run off.

The buffet analogy needs a little refinement. What the bandwidth hog

does is block the line to the all-you-can-eat buffet so that nobody else

can get any food. That's not a behavior that would be tolerated in a

restaurant, and it shouldn't be tolerated in a residential network.

Unfortunately, it wasn't the huge problem when DOCSIS was designed, so

the 1.0 and 1.1 versions of the technology don't address it, certainly

not as well as Full-Duplex Ethernet, 802.11e WiFi, and DSL do.

Some may argue that the Internet doesn't need a fairness system as it's

mostly a local problem, and I have some sympathy for that point of view.

But in the final analysis, we all know that some of our bits are more

important than others, and the network will work better if the layer 3

and layer 2 parts can communicate that sort of information between each

other.

I don't view this as a moral problem as much as an engineering problem.

Moral philosophy is certainly a fascinating subject (as is video

coding), but it's outside the scope of the current discussion.

RB

David Farber wrote:

________________________________________

From: Bob Frankston [bob37-2 () bobf frankston com]

Sent: Saturday, January 12, 2008 1:01 AM

To: David Farber; 'ip'

Subject: RE: [IP] Re: F.C.C. to Look at Complaints Comcast

Interferes With Net - New York Times

Moral court again ...

Does this mean I can't share files with my neighbor because of the

cost of peering with a remote provider? Will someone judge that

backing up over the net is not an appropriate use of the network? Am I

not allowed to backup to peers?

-------------------------------------------

Archives: http://v2.listbox.com/member/archive/247/=now

RSS Feed: http://v2.listbox.com/member/archive/rss/247/

Powered by Listbox: http://www.listbox.com


-------------------------------------------
Archives: http://v2.listbox.com/member/archive/247/=now
RSS Feed: http://v2.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com

Current thread:

Interesting -- comment from author -- F.C.C. to Look at Complaints Comcast Interferes With Net - New York Times David Farber (Jan 14)
- <Possible follow-ups>
- Re: Interesting -- comment from author -- F.C.C. to Look at Complaints Comcast Interferes With Net - New York Times David Farber (Jan 16)