1. A Business Case for Peering in 2004 (v0.7)
William B. Norton
Co-Founder & Chief Technical Liaison
Equinix, Inc.
(V0.7), XChangePoint Europe October 28, 2005
Updated: Oct 26, 2004

2. Internet Researcher 90% externally focused Many documents on Protocols
Lack of Operations documents
Research: Peering
How does Peering work?
What are the definitions?
What are the “Tricks of the Trade?”
White paper process..

3. Community Operations Research
“Ground Truth” w/dozens of experts
Write White Paper v0.1
Walk community through WP for comments
Revise White Paper into new version
Present White Paper at conferences
Solicit comments over lunches and dinners
White papers so far…

4. Internet Operations White Papers
“Interconnection Strategies for ISPs”
“Internet Service Providers and Peering”
“A Business Case for Peering”
“The Art of Peering: The Peering Playbook”
“The Peering Simulation Game”
“Do ATM-based Internet Exchanges Make Sense Anymore?”
“Evolution of the U.S. Peering Ecosystem”
“Asia Pacific Peering Guidebook”
“A Business Case for Peering in 2004”
Freely available. See Web site or send to
Or Google for “William B. Norton”

5. Disclaimer: Building Models
There is nothing about Equinix in this presentation
We are simply building models here
All prices are estimates, averages, surveyed values, or made up
Your mileage may vary
Adjust the models to meet your situation
Also, this is an early draft

6. Does Peering Make Sense Anymore?

7. The 2004 U.S. Transit Model

8. The Optimal Transit Model
What is the optimal commit level given an expected transit volume?

9. What is the optimal commit level given an expected transit volume?
Optimal Transit Model
What is the optimal commit level given an expected transit volume?
The 100% Heuristic
the Leo Transit Commit Heuristic
“I Don’t want to pay for traffic I don’t use!”

10. Compare 1M vs. 10M Commit
Graphically…

11. Comparing 1M and 10M Commit
Compare the heuristics…

12. Comparing Transit Commit Heuristics

13. Cost Savings of Optimal Transit
Same effect between 10 and 100M commits…

14. 1M vs 10M vs 100M Commits
Algebraically find the crossover point…

15. Norton Heuristic
Q: When do we commit to the next higher commitment level?
A: At the Transit Breakeven Point (TBE)
Plug in for T(c) and T(c+1)
Plug in for V(c+1)

16. Transit Breakeven Points

17. The Peering Model Peering isn’t free…Monthly Costs include:
Local Loop fees (the cost to get in IX),
IX collocation fees, and
IX peering port fees.
Also consider Equipment Costs
which can be amortized into Monthly Cost
Monthly Peering Costs=LL+IX+IXP+E
But how do we compare Monthly Costs against $/Mbps per month?...

18. Peering versus (Optimal) Transit
1) Normalize the Peering and Transit units
$/Mbps (per month)
Vs.
Mbps Exchanged
In Peering or
Transit
Relationship
(Per month)
All as 95th Percentile measures

19. Peering vs. Transit at TV
2) Compare Peering at PV and Transit at TV

20. Example: Peering vs. Optimal Transit
Peering Cost: $9000/Month

21. 2004 Peering vs. Transit Model I – Have to build into IX
Monthly Peering costs
Local Loop
Peering Port
Rack Space
Peering Equipment
Model II – Already at IX
Monthly Peering Costs
Incremental Peering HW
Peering Equipment costs…

22. Peering Equipment Costs
Sources:
Richard Steenbergen (nLayer)
And Patrick Gilmore (Akamai)

23. Model I – Build into IX for 100M Peering
Local Loop Assumptions:
100M Transport into IX: $1000 per month
1000M Transport into IX: $4000 per month
Peering Rack & Port Assumptions:
100M w/ ½ rack at IX: $2500 per month
1000M w/ ½ rack at IX: $5000 per month
Graphically: P vs.T…

24. Model I – 100M Peering

25. 100M Peering at the EPB
Why peer when P>T?

26. Why Peer when P100>T100 ? 1) Performance
2) Greater control over routing
3) Meeting Peering Prerequisites
4) Already at the IX
5) Marketing value
6) 100M Peering is a stepping stone
7) Currently paying high transit fees.
8) Peering Costs are immaterial.

27. 100M Peering Revisited
Assuming only 20%, at 33% peering is closer call.
4*100M=1000M, Peering at 1G…

28. Model I – Build into IX for 1000M Peering
Graphically P vs T…

29. Peering Breakeven Analysis
New vs. Used equipment?

30. Model I – 1000M Peering – new vs. used Equipment doesn’t matter
Project forward – when gigE fills…

31. Adding 2nd gigE for Peering
When does 10GE peering make sense?

32. 10GigE Peering Model
Graphically P v T…

33. Model I – gigEs vs 10gigE
Summary stats for Model I 10G Peering

34. 10gigE Peering Metrics
Traffic exchange for less than $5/Mbps and covering all costs of peering!
Model II – What if I already have colo at an IX?

35. Model II – Already at IX Colo

36. Model II - 100M Peering Peering Port Fee Assumptions:
100M Port only: $1500 per month
1000M Port only: $4000 per month
Graphically: P vs T…

37. Model II – Peering 100M Ports
Graphically 1000M and 10G Peering…

38. Model II – 1000 and 10G Peering
$1.92/Mbps

39. Model II Summary Stats

40. Summary Transit Prices Plummeted Does Peering Make Sense?
Decision Makers: Peering best compared against Optimal Transit
New vs. Used Peering Equipment doesn’t matter
Peering Scales Well in the gigabit ranges

41. Acknowledgements
I am grateful to Richard Steenbergen (nLayer) and Patrick Gilmore (Akamai) for their data points and insights during the initial development of the white paper models. I also thank John Hardie and Lane Patterson (Equinix), Scott Bradner (Harvard), Joe McGuckin (via.net), Leo Bicknell (AboveNet), Dave Woodelet (Shaw Fiber), Rich Collela (AOL), Dan Golding (Burton Group) for their suggestions.