Home page logo

interesting-people logo Interesting People mailing list archives

Re: Good submission -- How to fix the Internet New Short Video: "Network Neutrality in 30 Seconds" (Part 1)
From: David Farber <dave () farber net>
Date: Thu, 28 Aug 2008 05:48:44 -0400

Begin forwarded message:

From: "Michael O'Dell" <mo () ccr org>
Date: August 28, 2008 12:48:14 AM EDT
To: Bob Frankston <Bob19-0501 () bobf frankston com>
Cc: dave () farber net
Subject: Re: [IP] How to fix the Internet New Short Video: "Network Neutrality in 30 Seconds" (Part 1)

You are making the classic mistake of confusing "local connectivity"
with "Internet Service".  the former has to do with "wires" that move
bits over local or metropolitan distances and don't necessarily have
anything to do with Internet Service, although it is routinely used
for that.

"Internet Service" is the ability to exchange packets with any other
subscriber to "Internet Service", no matter where they are in the
world. that's what Internet Service is all about.

If you price a T1 connection to a major ISP, you get a quotation
with *two* components - one for the T1, which is the local connectivity
to the ISP, and a second component, sometimes called the "port charge",
is for the actual Internet Service.  the T1 is usually a pass-through
cost from the facilities provider and marked up some
modest amount to cover overhead. in the case of most T1 lines,
the cost for the T1 itself is determined by the LEC or a CLEC
if you're luck enough to be served by one of the few remaining.

The charge for Internet Service is derived from the cost of building
and operating the network, both long-haul and metro, at sufficient
quality to satisfy customers as well as continue to make investments
for growth.  For instance, a router suitable for major backbone
service these days can trivially cost $250K and much more.
all the equipment needed in major service node can easily
run to several million dollars per node. there are 35 NFL
cities in the US - that's a good proxy for the number of major
metro clusters, and there's probably several large nodes in
each of those major clusters. if you do the sums, just that
equipment alone can scare $500mm to death. (we're talking
initial buildout.)

then building a long-haul fiber plant to tie it all together
is an interesting exercise. even if an national-scale ISP
buys fiber or lambdas, the equipment is there and somebody
paid for it and they want a return on their investment.

Most terrestrial fiber systems need optical
amplifiers every 120 kilometers or so. Ultra-long-reach systems
that would raise that to 300-400 kilometers haven't really
worked out like people had hoped.  Yes, these systems can
run DWDM, but the transmission hardware required is non-trivially
expensive. for "3R" regeneration, most systems require demuxing
the lambdas and regenerating each one individually, then muxing
them back together for the outbound optical amplifier.

It is these national-scale and internationa-scale networks
which *are* the commercial Internet these days.
the MSOs and pure DSL providers and the like
only provide the local tails for the most part.
they do not constitute the "big I" part of the Global Internet.

it is the span, reach and scale of these large networks that allow
people pretend that it doesn't cost any more to send packets
to Beijing from NYC than it does to send them to Boston.

what you pay your ISP for Internet Service (as opposed to local
connectivity to the ISP) pays for the right to send packets
willy-nilly without regard for the geography of the destination.
if you ISP is smaller than national scale, it almost certainly
buys service from a larger ISP, and that service takes the
traffic everywhere - that's what any customer is paying for -
the ability to send packets anywhere in the Global Internet
with impunity and not worrying about getting charged by
the packet or by the mile.

to make that last part work, it's about scale and getting
the law of large numbers on your side. (also known as
statistical multiplexing)

at the end of the day, an ISP *must* understand their real costs
because if they don't charge enough for Internet Service,
well, that's when it gets ugly.  and this business of understanding
network costs is an entire other topic.


Bob Frankston wrote:
This does create a big problem -- why should my local connectivity be burdened with the cost of getting bits across the ocean? The benefit of each bit across the ocean bears little relationship to the value of my watching a high school football game. Any funding model that attempts to link the two is fanciful at best and generally perverse. The benefit accrues to society as a whole and thus should or even must be funded that way. We see this in failures which can affect economies. All the more reason to treat it as a national or international expense and not as what amounts to a tax on my local bits. The Internet has only one problem – that we treat it as a scarce resource rather than trying to maximize the opportunity.
-----Original Message-----
From: David Farber [mailto:dave () farber net]
Sent: Wednesday, August 27, 2008 18:46
To: ip
Subject: [IP] How to fix the Internet New Short Video: "Network Neutrality in 30 Seconds" (Part 1)
  Begin forwarded message:
From: "Michael O'Dell" <mo () ccr org>
Date: August 27, 2008 5:34:10 PM EDT
To: dave () farber net
Subject: Re: [IP] Re: New Short Video: "Network Neutrality in 30 Seconds" (Part 1)
Excuse me, but the last time I checked,
the Internet was a global phenomenon and
the customers actually care about global connectivity.
a US MSO just trading peer traffic with US big networks
will NOT get the customers everwhere in the world.
in spite of urban legend to the contrary,
getting across very large bodies of water is
*very* expensive. the cost of underwater
bandwidth is *much* higher than terrestrial,
and btw - that ain't free either.
"The Internet only has 3 problems:
           (i) it isn't infinitely fast
           (ii) it isn't perfectly reliable
and      (iii) it isn't free."
just fix those and we're done!
 David Farber wrote:
Begin forwarded message:
From: Barry Gold <bgold () matrix-consultants com>
Date: August 27, 2008 11:25:54 AM EDT
To: "Patrick W. Gilmore" <patrick () ianai net>
Cc: Dave Farber <dave () farber net>
Subject: Re: [IP] Re:   New Short Video: "Network Neutrality in
30    Seconds" (Part 1)
Patrick W. Gilmore wrote:
> I have a couple questions regarding your assumptions.
>> First, the ISP has to pay somebody else (a backbone provider) for a
>> "hose" (water main) to transmit their packets to the rest of the
>> internet.  If their customers demand more bandwidth, they have to
>> buy a
>> bigger water main. And that needs to be paid for somehow: either by
>> raising rates for all their customers, by charging the heaviest
>> users
>> extra, by imposing bandwidth "caps" to encourage the heaviest
>> users to
>> buy a higher service tier, or some combination.
> First, why does the broadband provider have to pay a "backbone"?
> Most of the filtering, capping, limiting, etc. is centered around
> the P2P debate.  Cable-cos and DSL providers almost all have free
> peering with one another, and P2P is almost exclusively traffic
> between end users on these networks.  Where did the "backbone" come
> in?
OK, having read the Wikipedia article on "Tier 1 networks," I'm
inclined to agree with you.  It appears that most large cable ISPs
are probably "Tier 2" but peer with each other and hence don't pay
transit/settlement charges.
What that means for this discussion is that most large ISPs
(cablecos and telcos) really don't have external charges as an
excuse for limiting user traffic.  There's still the other issue I
raised, however.
> Also, Verizon and SBC both own Tier One providers, who pay no one.
> They are what you would cll the "backbone".  (Leaving aside how
> very wrong that word and its connotations are in this discussion.)
> And what do you think the chances are the large MSOs aren't
> thinking about buying a Tier One or becoming one themselves?
Yes.  I was talking about cablecos, not telcos.  I agree that
Verizon & SBC/AT&T own their own "backbones" and, as such, any
settlements are purely "funny money" (internal funds).
>> Second, the current system is asymmetric(4).  The installed fiber/
>> wire
>> base can carry multi-terabits downstream (CO to customer) but only
>> a few
>> gigabits upstream (customer to CO).  This usually applies to the
>> "hoses"
>> they buy from backbone providers too: a gigabyte of outgoing packets
>> costs significantly more than a gigabyte of incoming packets.
> First, you admit DOCSIS 3.0 is symmetric, and you ignore the DSL
> providers.
Yes.  I was specifically talking about cablecos.  There is no
asymmetry for telcos and other DSL providers, and _over the next few
years_ the assymetry will disappear for cablecos as well.  So my
analysis applies only right now, not 2-3 years from now.
But even granting the idea that upstream bandwidth is
> limited, there is no cost associated with it.   The "backbone"
> providers charge on the greater of in or out, not the sum of the
> two.  Since you admit downstream capacity is greater than upstream,
> the cable-cos will not be charge for outgoing bits and
> realistically never will be.
Again, this is probably true of cablecos/telcos.  Brett reports
buying asymmetric connections from larger providers, and while I
think he is biased, I have no reason to disbelieve his first-hand
reports.  Assuming Brett is telling the truth, it seems likely that
most other small providers will do something similar: pay less for
an asymmetric connection because it _costs_ significantly less than
one that can carry the same bit rate in both directions.
Or maybe not.  Perhaps Brett is in an unusual position, having to
buy his connection "retail" from a cableco rather than directly from
a Tier 1 or even Tier 2 provider.  In which case, the problem
affects only him and other similarly-placed (semi-)rural providers.
In which case, regulators could (and IMHO should) make appropriate
exceptions for smaller providers like Lariat and only apply NN rules
to larger ones.
Remaining, however, is the congestion problem.  If a cableco is
currently running DOCSIS 2, they have a lot more bandwidth available
for downloads than for uploads, and customers who make a more
symmetric use, uploading nearly as much as they download, can cause
a congestion problem for other users who share the same cable node.
I think that is sufficient reason for them to "cap" or at least
delay upward traffic for the heaviest users.
But, repeating what I said earlier: I think any such limits should
be based on _quantity_, not on _protocol_.  Just count the packets
and the bits, and don't worry about what's inside them.  (These
aren't the bits you're looking for.  They can go about their
business.  Move along...)
Archives: https://www.listbox.com/member/archive/247/=now
RSS Feed: https://www.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com
Archives: https://www.listbox.com/member/archive/247/=now
RSS Feed: https://www.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com

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

  By Date           By Thread  

Current thread:
  • Re: Good submission -- How to fix the Internet New Short Video: "Network Neutrality in 30 Seconds" (Part 1) David Farber (Aug 28)
[ Nmap | Sec Tools | Mailing Lists | Site News | About/Contact | Advertising | Privacy ]