Geoff Huston, APNIC: The Death of Transit and the Future Internet

De IRHM.MP
Aller à : navigation, rechercher

The Death of Transit and the Future Internet
Geoff Huston Chief Scientist, APNIC

This presentation is not a bout any specific network details:

Or specific network plans Or particular services
Or any particular technology Or anything like that

It’s about architecture

And, in particular, about the evolution of network architecture in the Internet

Our Heritage

The Telephone Network

The major technology achievement of the twentieth century

  • Connected handsets to handsets
  • The network was intentionally transparent
  • Real time virtual circuit support between connected edge devices
  • Network-centric architecture with minimal functionality in the edge devices

Computer Networks

The original concept for computer networks was based on the telephone network

  • The network was there to enable connected computers to exchange data
  • All connected computers were able to initiate or receive “calls”
  • A connected computer could not call ”the network” – the network was an invisible common substrate
  • It made no difference if the network had active or passive internal elements

Internet Architecture (c1980’s) “End-to-End” packet design:

  • Connected computer to computer
  • All data is segmented into independent packets
  • The network switching function was stateless

No virtual circuits, no dynamic state for packets to follow

  • Single network-wide addressing model
  • Single network-wide routing model
  • Simple datagram unreliable datagram delivery in each packet switching element
  • hop-by-hop destination-address-based packet forwarding paradigm

Internet Architecture (c 1980’s)

  • TCP/IP Engine
  • TCP hosts

The Result was Revolutionary!

By stripping out network-centric virtual circuit states and removing time synchronicity the resultant packet carriage network was minimal in design and cost and maximized flexibility and efficiency

More complex functions, such as flow control, jiter stability, loss mitigation and reliability, were pushed out to the atached devices on the edge

Role Specialization

In the regulated world of national telephone operators every telephone network was “equal”

Markets do not normally support such outcomes, and we see role specialization as a way of sustaining efficient distribution chains to support public services

We rapidly started differentiating between Internet networks differentiating on roles and services and differentiating on the flow of revenues between networks

The 1990’s Internet

Transit Networks were “special”

  • These were the so-called “Tier One” networks
  • These networks collectively managed the “default-free zone” and arbitrated reachability on the network
  • These networks were at the apex of the money flow within the Internet ecosystem
  • They effectively formed a cartel that defined the Internet as we knew it at the time

Enter Content

Breaking the edge into clients and servers

  • Access networks service the needs of “clients”
  • Clients are not directly reachable by other clients
  • Clients only connect to services

The role of the network here is to carry clients to the service access point

  • The assumption here is that there are many more clients than service points

Who pays whom?

  • The only reason why access networks have clients is because there are content services that clients want to access
  • Carriage access providers should directly pay for content for their users
  • There is no “end-to-end” financial setlement model in the Internet – both “ends” pay for access and network providers setle between themselves. To a carriage network, content is just another client

The Evolution of Content Service

The Tyranny of Distance

But not all clients enjoy the same experience from a single service

Facebook presentation at NANOG 68

Enter Content Distribution

Let them eat data!

The rise of the Content Distribution Network

  • Replicate content caches close to large user populations
  • The challenge of delivering many replicant service requests over high delay network paths is replaced by the task of updating a set of local caches by the content distribution system and then serving user service requests over the access network
  • Reduced service latency, increased service resilience, happy customers!

Today’s Internet Architecture

We’ve split the network into clients and servers

  • Web servers
  • Streaming servers
  • Mail servers
  • DNS servers

Servers and services now sit in CDN bunkers with global replication and DDOS hardening

Users don’t reach out to content any more - the CDNs bring content to users

Service portals are increasingly located adjacent to users And that means changes to the network:

  • Public Networks no longer carry users’ traffic to/from service portals via ISP carriage services
  • Instead, Private Networks carry content to service portals via CDN services

Who’s building now?

Almost all new submarine international cable projects are heavily underwriten by content providers, not carriers

Large content providers have huge and often unpredictable traffic requirements, especially among their own data centers. Their capacity needs are at such a scale that it makes sense for them, on their biggest routes, to build rather than to buy. Owning subsea fibre pairs also gives them the flexibility to upgrade when they see fit, rather than being beholden to a third-party submarine cable operator.”

Tim Stronge of Telegeography, January 2017

  • Fewer cables being built
  • And those that are being built are now single owner cables
  • And the majority are now self-funded

Tim Stronge, Telegeography, Jan 2017

  • And those that are being built are now
  • And the majority are now self-funded

Tim Stronge, Telegeography, Sept 2017

  • Submarine Cables

being built are now

Submarine Telecoms Magazine, November 2018 33

Today’s Internet Architecture

Who needs Transit?

  • If users don’t send packets to users any more…
  • If content is now delivered via CDNs to users via discrete service cones…
  • If there is no universal service obligation…

Then why do we still need Transit Service providers?

Closed Transit?

We see the CDN systems reserve a carriage resource through dedicated bandwidth / wavelength / cable purchase and effectively bypass the open IP carriage infrastructure

Transit?

Once the CDN caches sit “inside” the Edge NAT of the Access ISP then the entire wide area network becomes a marginal activity compared to the value of the content feeds!

Internet Names and Addresses?

If the Internet is (or maybe soon will be) a collection of discrete CDN service ‘cones’ then why do we expect end users to pay for the maintenance of:

  • A global address plan?
  • A global name system?
  • A single global network?

It’s not just Death of Transit

It’s the re-purposing of the entire network

  • Service provisioning sits within cloud providers and distributed data centres
  • Edge computers are now acting as televisions into the clouded world of data
  • The distinction between personal and public data realms is disappearing into the realm of corporately owned private data empires

Exactly where are we?

  • We started this journey building a telephone network for computers to communicate between each other
  • But now one-way content distribution lies at the core of today’s Internet
  • This content distribution role is an enterprise service framework rather than a public carriage service
  • The internal parts of the carriage network are now being privatized and removed from public regulatory oversight

If CDN feeder networks are private networks, and there is litle residual public carriage other than last mile access networks, then what do we really mean by “public communications policy”?

In the regulatory world ‘content’ is commerce, not carriage!

In today’s Internet what do we mean in a policy sense by concepts such as:

  • “universal service obligation” “network neutrality”
  • “rights of access” or even “market dominance”
  • when we are talking about diverse CDNs as the dominant actors in the Internet?

The Large and the Largest

The world’s 10 largest publicly traded companies, as ranked by their market capitalization, Q3, 2018

Content Really is King!

  • None of these seven technology companies are a telephone company, or even a transit ISP, or even an ISP at all!
  • All of them have pushed aside carriage networks in order to maintain direct relationships with billions of consumers
  • These valuable consumer relationships are based on content services, not carriage
  • There are not thousands of content service platforms any more
  • There are just a few left
  • And the space is dominated by a small number of dominant actors who set the rules of engagement for all others

“The size and scale of the atacks that can now easily be launched online make it such that if you don't have a network like Cloudflare in front of your content, and you upset anyone, you will be knocked offline.

In a not-so-distant future, if we're not there already, it may be that if you're going to put content on the Internet you'll need to use a company with a giant network like Cloudflare, Google, Microsoft, Facebook, Amazon, or Alibaba.

Without a clear framework as a guide for content regulation, a small number of companies will largely determine what can and cannot be online.

htps://blog.cloudflare.com/why-we-terminated-daily-stormer/ August 2017

Consolidation?

Boston Globe , June 14 2018

With a small number of truly massive enterprises at the heart of the area of digital content and service is this still a space that is shaped by competitive pressures?

Or do these dominant incumbents get to set their own terms of engagement with each other, with users, and even with the public sector?

We’ve been here before…

A term applied to America in the 1870 – 1890’s about the building of industrial and commercial corporate giants on platforms that were a mix of industrial innovation and enterprise with elements of greed, corruption and labor exploitation

During this period in the United States the dominant position within industry and commerce was occupied by a very small number of players who were moving far faster than the regulatory measures of the day.

The resulting monopolies took the US decades to dismember, and even today many of these gilded age companies remain dominant in their field

At some point in the past decade or so the dominant position across the entire Internet has been occupied by a very small number of players who are moving far faster than the regulatory measures that were intended to curb the worst excesses of market dominance by a small clique of actors.

These actors have enough market influence to set their own rules of engagement with:

  • Users,
  • Each other,
  • Third party suppliers,
  • Regulators and Governments

By taking a leading position with these emergent technologies, these players are able to amass vast

What is this all about?

This is no longer just a conversation about incremental changes in carriage and communications within the Internet. For me, the essential topic of this conversation is how we can strike a sustainable balance between an energetic private sector that has rapidly amassed overarching control of the digital service and content space, and the needs of the larger society in which we all would like some equity of opportunity to thrive and benefit from the outcomes of this new digital age.

Is it that these enterprises are:

  • so big?
  • exploitative of their workers?
  • distorting markets?
  • extracting monopoly rentals from consumers?
  • not providing consumers what they want?
  • In the quest for ever-faster service delivery we are seeing the return of proprietary solutions in applications and service delivery platforms that expose as litle as possible to the underlying network platform
  • Perhaps the problem is the looming demise of open technologies and open technology standards
  • Akamai uses Fast to improve content delivery
  • Google uses QUIC and BBR
  • Facebook and WhatsApp use strong encryption to hide the application from the network and the platform
  • Applications are no longer constructed on a platform of common libraries provided by the platform
  • Applications are now paranoid and avoid exposing their behaviour wherever and whenever possible
  • Applications are increasingly reluctant to use standard open technologies in standard and open ways
  • In its place we are seeing a resurgence of various closed technologies that create a set of datacentre-to-application bindings that are impervious to all third parties
  • These closed architectures make minimal assumptions about a common network protocol, a common name space or even a common name space
  • What happens to the efforts that support open technologies, open standards and open networking in such a world?

I just don’t know!

But I’m not sure that it’s all good!

Thanks!