[PLUG-TALK] Netflix neutrality

Keith Lofstrom keithl at gate.kl-ic.com
Mon Mar 2 22:39:13 UTC 2015


On Mon, 2 Mar 2015, Keith Lofstrom wrote:
KL> The airlines are a common carrier.  But they won't let me bring 
KL> along my car as luggage.  Streaming movies is like moving a car 
KL> compared to moving a purse - gigabits of data compared to the 
KL> kilobits this email uses.
KL>
KL> What Hollywood and Netflix are asking for, and what the politicians 
KL> want to give them, is displacing millions of purses so they can move 
KL> thousands of cars.

On Mon, Mar 02, 2015 at 09:51:12AM -0800, Paul Heinlein wrote:
PH> OK, I'll bite.
PH> 
PH> Where is it written that sending e-mail is a more appropriate use of 
PH> network bandwidth than sending video content?
PH>
PH> Without disagreeing with the obvious assertion that video consumes 
PH> more bandwidth than e-mail, I'd ask, so what?

You can move as much video as you want - if you choose to distract
yourself rather than learn something, it is your life.  However, if
you choose to slow down a thousand other uses, and use the law to 
enforce that, you are stealing from others.  Demanding high video
Quality of Service on channels designed to move autonomous packets. 
Video QOS means everyone else must wait.  If the net was truly
neutral, instead of "netflix neutral", then video would look far
more juddery on saturated channels than it does, because it would
get the same best-effort treatment as other less demanding uses.

I gave the example of email, not because it is the only suitable
use, but because it is what the internet was designed for.  Your
car is designed to move passengers.  If I filled the passenger
compartment with 10 tons of concrete, it probably wouldn't work
so well, and you wouldn't be very comfortable in there.  If you
want to move tons of concrete, buy a concrete truck - but don't
expect to get one for the same price as your Honda.  And if you
corrupt the laws so that Honda owners must subsidize concrete
trucks, some Honda owners will be priced out of the market,
and there will be more concrete trucks than there should be.

The internet map is endpoints, fiber, and the packet switches in
between.  For N endpoints, there is some architectured fraction
of N^2 switches.  I mentioned I designed them, right?  Well,
O(N^2) was the startup's business model, and we were growing the
business at twice the (rapid) internet growth rate for a while
[ ln(N^2) = 2*ln(N) ], until the VCs told us to ignore our
incredibly profitable customers and design endpoints instead. :-(

N^2 hasn't changed.  The internet is very big now, it is paid
for by the N endpoints, but has grown as O(N^2).  A traceroute
will show that your path to netflix goes through perhaps a dozen
switches - the network must handle the load, to all N endpoints,
and the graph combines into switches to reduce the total number of
fibers.  What traceroute doesn't show is the O(N^2) connectivity
of those switches;  as you add more switches, you add more
interconnect between them, too, within and outside of peering
centers.  Read "Tubes" by Andrew Blum;  peering isn't some dinky
operation like Pittock or Easystreet, these are literally acres
of hangar-sized buildings full of fiber optic cable connecting
thousands of racks of Juniper and Barracuda and Cisco backbone
routers, bigger than most data centers, in hundreds of locations
worldwide.

Netflix could, if they chose, solve this problem in many different
ways.  They could arrange for video servers in the same building
as all of the thousands of distribution routers (such as
car01.bvtn.or.frontiernet.net which my connection goes through)
so they don't have to burden a whole network of peering switches.
They could provide encrypted buffering hardware to the end
customers, described below.  Or, they could pay their share of
the capital costs of a high QOS global video routing network.
That is a lot more expensive than a best-effort-with-retry packet
network with limited buffering, and the financial risks are huge.

But no - Netflix wants to impose the cost of moving their flood
of time-sensitive data on Frontier (or Comcast, or CenturyLink)
and the peering hubs, O(N^2) more/faster/bigger switches without
compensation.  They must borrow to pay for their connectivity
based on the old assumptions of symmetrical traffic, while
servicing their debt with the same N customer endpoints. 
And when the endpoint providers try charging users extra for 
exceeding bandwidth caps, they get hammered for that.  

My usage?  I don't just run email - I sometimes move VOIP phone
packets over a VPN between our home and my wife's office.  But
that doesn't work anymore, because the network saturates, not
enough QOS even with significant buffering.  Now, when my wife
wants to use her office phone system to call patients or
hospitals, she drives back to work (15 minutes each way), or
just stays at work until it is late.  We used to use that link
3 or 4 times a week, 15 minutes at a time, consuming about
40 kbps on a supposedly 15 Mbps link.  Now our low bandwidth
VPN is too juddery - just like downloads, web surfing, etc. 
We get pushed aside for 4 megabit video.

Bandwidth is a limited resource.  It costs more to have more.
Trenched fiber bundles are only $100K per mile, and a customer
fiber drop and optical network terminal is only $2K, but 
switching many-to-all in a dozen hops across a continent can
cost billions of dollars.  Above about 10 Gbps, the only way
to get more bandwidth through a switch is by adding parallel
channels through smaller/faster switch fabrics. 

If hyper bandwidth was easy, then a Google data center would fit
in a shoebox of exahertz-clocked logic.  Instead, they buy acres
of warehouses full of relatively slow computers doing pretty
much the same thing for millions of customers in parallel.  
Switching data between fibers is a similar process.


PH> I know what I'd say: we've improved and magnified the infrastructure 
PH> that provides electrical power, and our uses of it have changed 
PH> accordingly.

Ah - but the electric network has limits, too.  It takes 15 years
from idea to electricity to design, permit, and build a long
distance power transmission line.  There is no place to store
power - if you generate slightly too little, brownouts, too
much, and the power becomes water spilled over the dam or extra
heat sent up the smokestack.  If everyone in a residential
neighborhood drew power to the limit of their 200A load center
(circuit breaker box), it would blow the "oversubscribed" breakers
feeding the neighborhood.  This is not theoretical - when the
power lines are pushed to 120% capacity (after all the industrial
users have been forced to shut down) then the lines can heat to
hundreds of degrees, stretch and sag, touch trees and short out. 
That is why Asplundh makes big bucks chopping big holes in the
trees around PGE's power lines.  If they do not chop big holes
often enough, local or regional blackouts can result.  Everything
has a designed use, everything has a design capacity.  Exceed it,
and bad stuff happens.

More capacity costs big bucks, network and electrical.  Electricity
is "multicast" - everyone gets the same 60 Hz power signal.  Internet
bits are unique, though multicast can reduce transmission costs.
Cable TV is somewhere in between, but is inherently multichannel
multicast and thus a much cheaper way to transport shared content.

KL> Netflix wants to feed impulsive customers who demand instant 
KL> gratification.  Hollywood doesn't want you to have a stored copy 
KL> (like this email you are reading) on your computer hard drive, just 
KL> a few buffered frames in your RAM. This is an expensive abuse of a 
KL> limited-bandwidth packet network designed for different purposes.

PH> Or, somewhat differently, this is a pragmatic response by people who 
PH> make their living producing entertainment to the threat of their 
PH> handiwork being stolen. (Dehumanizing all the people who make their 
PH> living in the entertainment field by labeling them as "Hollywood" is a 
PH> cheap rhetorical trick.)

Speaking of "cheap rhetorical tricks" - Do every one these
entertainers demand nationwide petabit data distribution to do
their jobs?  Many write books.  Many perform live.  And some
lost their jobs stamping DVDs.  Those disk makers probably should
have demanded "plastic neutrality", and subsidies for their
distribution systems from all the other users of distribution
systems.  But wait, they did ... media mail!  That, and franking,
and free rural delivery, etc. drove the post office bankrupt.

You know what I mean by "Hollywood" - the large corporations that 
have bought legislators and warped the laws to protect their
obsolete business models.  The MPAA is not an employee's union.
Stolen?  Are any of the movies missing?  Or is it only that the
revenue is insufficient to buy enough cocaine for the bosses?


> Frankly, I like the subscription model much better than the purchasing 
> model. Most shows I like to watch once, perhaps twice. I have no 
> interest whatsoever in owning it on DVD. Subscribing to a service 
> meets my needs much better -- so the free market is doing its job.

You can still get the service - just pay for the bits you use,
and pay peak hour rates that feed through to the network owners,
so they can purchase the equipment needed to support your peak
hour use.  

The easiest way to patch up the switched packet network that we
have is for video subscribers like you to have encrypted set top
boxes with perhaps a terabyte of storage, which pre-download
(during slack hours) simultaneous multicast content (say the
first five minutes of every major Hollywood title a few days in
advance of release) then pull the remainder of the movies you
select in advance during slack hours.  If you are compelled to
chose your evening's entertainment at the last minute, rather
than select it in advance, you pay last minute per-megabyte
rates, whatever clears the market during high demand times. 

So, you might end up paying $0.50 to download movie at times of low
demand, and $20 for a movie chosen when the network is saturated. 
My VOIP connection, using 1% of the bandwidth for 20% of the time,
might pay $0.001 for a daytime call, and $0.04 for an evening call.

But to demand that my content gets shoved aside, and that the
network providers subsidize yours, is not "the free market doing
its job".  Comcast doing whatever the hell they want with their
equipment to maximize their revenue is the free market doing
its job.  If you don't like the way they do it, invest in a
competitor.  If you are wrong, kiss your own retirement savings 
goodbye - not somebody elses.

And that is not hyperbole.  I am designing a satellite network,
http://server-sky.com, which will someday deliver terabit-per-
square-kilometer data service to the tropics, and the billions
of underserved people there, without fiber networks or peering
or monopolies.  A demand-priced internet service that does not
cede control and redistribute network revenue to Netflix and
"half a dozen giant media conglomerates headquartered north
of Wilshire Blvd. in Southern California".  In a free market,
alternative internet providers have a chance to compete by
offering customer-tailored solutions.  You get what you want,
I get what I want, and we both pay enough to attract providers,
or get paid enough to grow our provider businesses.

KL> Want to displace thousands of political discussions and muzzle 
KL> individual free speech so vidiots can watch the latest action movie?
 
PH> I suspect this is at the moral heart of your argument, but the 
PH> discussion of the merits of trivial vs. political speech is
PH> something different than that about network neutrality.

Besides drawing your Venn diagram wrong (political speech can be
trivial or not, a matter of preference), network neutrality is
the government's business ONLY because of the first amendment. 
Local governments (violating the first amendment) restrict the
number of telecommunication carriers a political jurisdiction
can have.  Much of the "net neutrality" argument stems from
giving one or two providers a monopoly on information traffic,
then whinging when the providers use the monopoly monopolistically. 
Or more precisely, when they make different monopolistic decisions
than the monopolistic decisions that we personally prefer,
benefiting their own monopoly rather than other monopolies.

Personally, I say "screw monopolies".  Yes, the chaos and
ferment of a free market is confusing, sometimes expensive. 
Sometimes people lose their sinecure and must learn new jobs. 
But a free market wrested from Bell Telephone's control is
the reason you have a smart phone rather than a Bell model
500 dial telephone.  The re-consolidation of the Bells as
Verizon is why the sound quality sucks and texts cost a
million times more than internet packets.

If a jurisdiction offers underground tunnels to run hundreds
of different fibers (or whatever) through, then a region can
offer dozens of colocated providers, offering many different
services for many different needs.  I agree with Russell
Senior that the city should own the tunnels, though I
disagree that the city should own what runs through them.

With many services available, I can get my slow jittery
gigabyte distro downloads from one service, and push a
40 kbps VOIP connection through a different channel,
both optimized for the traffic, and both a lot easier
to provide than high-QOS high-speed video-on-demand,
which you may choose to pay for.

Cheap, fast, good.  Pick two.

Network neutrality is about freedom of expression.  Libre,
not gratis.  In a world where powerful interests maintain
their monopolies by silencing the powerless, a network of
networks that protects the free speech rights of everyone
is a treasure.  If net neutrality is redefined as Hollywood
and Netflix Subsidy, throttling small participants to make
room, then we have lost our freedom and lost our minds.

Keith

-- 
Keith Lofstrom          keithl at keithl.com



More information about the PLUG-talk mailing list