nwen 402 – peering & exchange t2 2015 with material from geoff huston, andy linton &...
TRANSCRIPT
NWEN 402 – Peering & Exchange
T2 2015
With material from Geoff Huston, Andy Linton & Valerie Schaeffer
Outline• Connecting the big players of the Internet• What we want it to look like and what it actually
looks like• The problem• The solution
NWEN402 2
The Sum of Many Parts• The Internet is the sum of more than 30,000
component service providers (ISPs)
• Each ISP has its own network with services, tariffs, customers, policies.
The Well-Ordered Internet• This view is based on a conventional
distribution infrastructure
• Every relationship is bilateral– a provider sells services to a consumer
• Tiering of the ISP sector– Tier 1 - global backbone transit networks– Tier 2 - national wholesale transit networks– Tier 3 - local retail access ISPs
The Well-Ordered Internet
Client
Provider
Local ISP
Regional ISP
National ISP
Transit ISP
Local ISP Local ISP Local ISP Local ISP Local ISP
Transit ISP Transit ISP
National ISP National ISP National ISP
Regional ISP Regional ISP Regional ISP Regional ISP
• The resultant structure is a hierarchy of relationships
The Internet - as we know it• The competitive ISP industry tends to equilibrate on the lowest local
cost structures• There are no objective criteria to identify who is the provider and who is
the customer• Debt is better than profit as a means of leverage of ISP value
– there are fewer ways of establishing true value
• underlying carriage tariffs shape Internet-based ‘locality’• Within each local tier cell ISPs tend to SKA peer - or not
– bluff is a critical component of the peering game
• Strict tiering blurs because of the confusion over value identification
– is content of equal value to transit?
The Internet - as we know it
Exchange
Exchange
ExchangeExchange
ISP
ISP
ISP
ISP
ISP
ISP
ISPClient Net Client Net
Client Net
Client Net
Client Net
Client Net
Client NetClient Net
Client Net
Client Net
Client Net
Client Net
Client Net
Client Net
Client Net Client Net
The Problem - as we see it• how to interconnect many thousands of
component networks while:• minimizing local cost everywhere by:
– localizing transit traffic– matching diverse import, export and transit policies– avoiding super dense traffic black holes– maintaining stability and quality– both technical and financial
• and also adding thousands more component networks
The Role of the Exchange• An examination of the rationale for public
Internet exchanges.
• https://www.youtube.com/watch?v=cpAUedhcjAY&feature=plcp
The Exchange Router• Too simple• Router-based exchanges impose transit policy
A
Exchange Router selects preferredpath to destination A
A
Route PeerMesh
Bilateral peering allowseach ISP to select preferredpath to destination A
The Exchange L2 Switch– An L2 switch does not implement routing policy– Routing policy is then the outcome of bilateral
agreements
The Distributed Exchange
Switching Mesh
PeeringVirtual Circuits
• Use of L2 virtual circuits to support bilateral peering eliminates the need for co-location
Adding Value to the Exchange
• exchanges represent a very efficient centralized service launch point
UsenetServer
DNSServer
Web Hosting Services
Multicast Router
Route Server
Service Environment
Web CacheServer
Economics of an exchange• Nicholas Economides “The economics of the Internet Backbone”, Law
& Economics Research Paper Series, June 2005.
• Cremer Jacques, Patrick Rey & Jean Tirole “Connectivity in the Commercial Internet” Journal of Industrial Economics, 2000.
• https://www.youtube.com/watch?v=tR1sLLOYxnY
17
Bring the rain• Software Defined Exchange
NWEN402 18
Gupta, Arpit, Laurent Vanbever et al.. "SDX: A software defined internet exchange." In Proceedings of the 2014 ACM conference on SIGCOMM, pp. 551-562. ACM, 2014.
Economides, Nicholas. "The economics of the Internet backbone." NYU, Law and Economics Research Paper 04-033 (2005): 04-23.
D’Ignazio, Alessio, and Emanuele Giovannetti. "Predicting internet commercial connectivity wars: The impact of trust and operators’ asymmetry." International Journal of Forecasting (2015).