| |
|
Warning: This post is a bit geeky. If you’re not technically minded, you may want to check out the next one…
Mediating the Network
I’m often asked by what do I mean by “mediating the network”. I’ve covered the topic a few times before. However, this time I want to be a bit more specific.
Let’s first review what needs to happen in order for ISP to
maximize the “billable capacity” of their networks in a
tiered Internet. Today, almost all network operators have a flat-rate
best-effort model for service. Australian providers, and very recently, Time Warner’s Cable network
represent some rare exceptions. Under the current broadband billing
paradigm, consumers pay a single monthly fee for as much bandwidth as
they can consume. In the case of the Time Warner trial, and Australian
ISPs, they pay for the amount of best-effort bandwidth they consume by
the byte. The additional charges amount to the absolute quantity of
bandwidth at varying qualities consumed as opposed to charging for the
quantity and quality of that bandwidth.
The Current "Tiered" Internet
Some net neutrality proponents
believe that there is no tiered service on the Internet as it exists
today. I would, of course, respectfully disagree. Large and small
corporations today purchase tiered services from a number of providers.
One of the most notable examples are Content Delivery Networks (CDNs).
These networks promise to reduce transit bandwidth costs by caching
frequently used content closer to the edge of the network nearest the
consumer. They also effectively tier the network by taking advantage of
a “bug” or, in their case, a “feature” of TCP
and the queue management systems that are actively deployed today. The
current dominantly used version of TCP, “Reno”,
has some interesting behavior on the network. It works in such a way
that with shorter round trip times, or delay, those TCP flows from the
CDN are able to obtain a greater share of the bandwidth at bottleneck
points on the network. This has been well understood for a while.
Peer-to-Peer (P2P) networks actively use this strategy very effectively
– often collectively obtaining more than 80% of the available
bandwidth on at broadband congestion points.
Therefore, the Internet is already “tiered” as it is
today. Websites often use CDN services to distribute “high
bandwidth” content like video because it streams or downloads
faster off those distributed servers. This may represent “left
hand lane” best-effort and not “true diamond lane”
service. It is, however, better service than you get from your typical
hosting provider. The cheapskates, on the other hand, often use P2P
networks. They rely on broadband consumers downloading software, which
shares their uplink connection's bandwidth to obtain the lowest cost
distribution. Consumers don’t want to download and share their
computer or bandwidth unless they get something in return.
Unfortunately, for Hollywood, they most often get a freebie in the form
of pirated content in exchange for sharing. The P2P applications can
get even a greater share of bandwidth than the CDNs because they
frequently have even shorter round trip times.
An interesting development has occurred over the past couple of years. New versions of TCP such as those from Netli (now Akamai), FastSoft, and nuMetra
can also grab far more bandwidth from those congested bottlenecks.
There are also UDP-based solutions on the market from companies like Rivulet and EdgeStream. The TCP systems can also operate very stealthily with higher level protocols masquerading on port 80
so they look like fairly ordinary HTTP traffic. This gives the ISPs an
even greater problem. They are basically left with having to ever
increase their capacity without a commensurate increase in revenues
under the current pricing regime. Thus. we have experiments with the Australian pricing model by the likes of Time Warner.
The questions becomes whether, from a competitve standpoint, these new
TCP and UDP implementations are playing by the rules. Being overly
aggressive and not polite on the network can be bad for both the ISPs
and consumers who are not using that aggressive protocol. Competively,
the CDN providers must move to this model for the simple reason that
they all must offer best "left-hand lane" service they possibly can.
Therefore, from a business standpoint, the polite versus aggressive
debate become pretty academic. Every driver knows, particularly those
in Southern California, the difference between diamond lanes and the
left hand lane to the right of it.
Mediation enables diamond lane access for the media grid.
CDNs or Tier 1 networks with only "left hand lane" access will be left in the dust. As
I'll discuss below, the Open Source Peering Agreement (OSPA) creates a uniform diamond lane system for
all ISPs on the Internet and allows the current CDN providers to make
use of it in exchange for conforming to the standards of the agreement.
Bandwidth Quality, not Quantity
Bandwidth has quality characteristics that can be monetized. If we look at it two dimensionally,
packet loss and latency, rather than in a single dimension, bytes,
other much better pricing models could emerge for the ISPs. Obviously,
the big one, today, is high latency video on the Internet. Back in
1996, a mathematician by the name of Tom Kurtz,
provided the basis for an economic and theoretical framework to describe the behavior of packets in
bottlenecks using simple priority, resulting in my interpretation of the ISP opportunity,
shown in the diagram below (email me if you want to discuss it):
In essence, as one the ISPs we’ve been working with described it
to me, they need a general purpose “bandwidth spigot” to fairly (whatever that means) distribute
the packet loss and latency for the bottleneck appropriatly
for each application. This “spigot” is analogous to the trickle
(low volume & pressure), high volume, and high pressure narrow
stream feeds most of us have in our automatic irrigation (water)
systems outside our homes. The combination of the ISTP’s TCP
plus some secure AQM technology in the bottleneck with some various and
sundry other features allows us to create the right kind of
“spigot” so that ISPs may maximize the billable capacity of
their network. By the way, this approach does not violate the principles of net neutrality, as long as ISPs agree to set their mediation policies to adhere to common carriage principles.
The nuMetra design for this system will operate in all world
jurisdictions by allowing the ISPs to handpick those mediation policies
for their network.
A central tenet of ours has been to recognize that "fairness" in
whatever flavor you mean can only be enforced inside the network.
Relying on software developers who develop congestion control
algorithms to be "polite" has no commercial value. The creation of
these polite TCP implementations should be left to academia. The
commercially viable and valuable work is in mediating those bottlenecks.
ISPs Need a Flexible Business Model and Solution
In a way, many ISPs really don’t care about Hollywood’s piracy problem.
Some believe, such as Time Warner, that they can cost factor all of
those P2P users, and high volume video consumers into the pricing
equation and make more money from their broadband service by getting
the consumer to pay by the byte. We'll see if the consumer bites on
that value proposition.
Whatever new or existing network equipment they may be providing to do will almost
certainly not have the quality of bandwidth pricing flexibility that a
system like the one I describe above. In my mind, ISPs should go
for systems that maximize flexibility and monetizable capacity while
helping Hollywood with anti-piracy and offering the consumer a superior
experience.
The IP Bandwidth Food Chain
I hope I’ve made it clear that P2P is like the
“hobos” of the great depression on today’s IP
bandwidth train. Normally, they let them on, only if they could hop a
ride on the caboose. ISPs, today, are allowing them in the coach
passenger cars. In the current best-effort networking paradigm, they
sit at the top of a pyramid. They also offer a really lousy customer experience. Joost's problems
are just one example of this. Fundamentally, P2P systems are for
hackers and teenagers to pirate content. The complexity of using
something like BitTorrent makes it all but impossible to easily watch
what you want when you want it. Web-based distribution systems like Hulu and YouTube
offer a much better and more satisfying consumer experience. In terms
of the IP bandwidth economy, the consumer and distributor both get what
they pay for.
If you’ve read our wiki and have been following what we’re doing, you probably recognize that the Tier 1 ISPs are next down on the bandwidth food chain. In essence, they never have to pay for IP Transit,
so they should have the second lowest cost structure for delivering
video across the Internet. CDN-only operators, and Tier 2 ISPs then
Tier 3s are next down pyramid in terms of cost structure. The diagram
looks something like this:
Our Open Source Peering Agreement (OSPA)
may change the pricing dynamics for the current peering and transit
agreements by creating a new world order of Tier 1s. Level 3, a true
Tier 1 provider, for example, just announced pricing for its CDN
services that are comparable to that of IP Transit. They’re
pushing the commoditization effort of CDNs forward rapidly and beginning to focus on higher value-added services.
Implicitly, they and other Tier 1s recognize these distributed
“grids” are the only way video is going to be well
supported on the current and next generation web.
Since under the OSPA, “new world order” InterStream Tier 1+ ISPs must share their “transit keys”
with one another so that common carriage is enforced by allowing ISPs
to unlock each other’s premium bandwidth services at the
bottlenecks. The OSPA, in essence, allows the GPL to enforce common carriage practices across the Internet. If what net neutrality really means is common carriage, then I’m totally in favor of it as long as ISPs are allowed to put those hobos in the caboose.
The Key Questions
We’re now left with a couple of questions: How will the new
“Tier 1” network evolve under the InterStream approach?
Will ISPs be willing to adopt open source?
The answer to the first question is pretty obvious. We’ll have
exactly who we’ve got today or we’ll get some intersection,
or perhaps even a subset of the current group.
It could remain as the same oligopoly or, more likely evolve into a
much more complex ecosystem. We’ll see as we finish the process
of defining the OSPA as more ISPs participate.
We believe ISPs will readily adopt a mediation system based on open source. They have a natural incentive for the patent infringement protection it offers them as well as dramatically reducing their capital and operational costs. I’m not the only one
in Silicon Valley who feels this way. I expect the vast majority of the ISPs and network operators
are looking to help improve their profit margins as well as be willing
to support common carriage over their networks.
We’ll see over the next several months how the OSPA will reshape
IP bandwidth economics. I expect the same breakthroughs will occur with
net neutrality as long as the net neutrality proponents stick to their
knitting with a common carriage message and not push a bill that could have disastrous unintended consequences.
Jeff Turner
|
|
|
