New Zealand Peering Exchanges APE and WIX Andy Linton February 2004.
A Business Case for Peering in 2004 (v0.7)
description
Transcript of A Business Case for Peering in 2004 (v0.7)
A Business Case for Peering in 2004 (v0.7)
William B. NortonCo-Founder & Chief Technical Liaison
Equinix, Inc.
(V0.7), XChangePoint Europe October 28, 2005 Updated: Oct 26, 2004
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..
Community Operations Research
1. “Ground Truth” w/dozens of experts2. Write White Paper v0.13. Walk community through WP for
comments4. Revise White Paper into new version5. Present White Paper at conferences6. Solicit comments over lunches and
dinners
White papers so far…
Internet Operations White Papers1) “Interconnection Strategies for ISPs”2) “Internet Service Providers and Peering”3) “A Business Case for Peering”4) “The Art of Peering: The Peering Playbook”5) “The Peering Simulation Game”6) “Do ATM-based Internet Exchanges Make Sense Anymore?” 7) “Evolution of the U.S. Peering Ecosystem”8) “Asia Pacific Peering Guidebook”9) “A Business Case for Peering in 2004”
Freely available. See Web site or send e-mail to [email protected] Google for “William B. Norton”
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
Does Peering Make Sense Anymore?
Transit Prices per Mbps
$1,500
$1,200$1,013
$600$450 $425
$275$125
$0$200$400$600$800
$1,000$1,200$1,400$1,600
1997 1998 1999 2000 2001 2002 2003 2004
The 2004 U.S. Transit Model Transit Prices (August 2004) (TC )
1 Mbps Commit $125 per Mbps10 Mbps Commit $80 per Mbps
100 Mbps Commit $45 per Mbps1000 Mbps Commit $30 per Mbps
Transit Commit Prices
$0
$20
$40
$60
$80
$100
$120
$140
1 121 241 361 481 601 721 841 961 1081 1201 1321
Mbps
$/M
bps
The Optimal Transit Model
• What is the optimal commit level given an expected transit volume?
Transit Prices (August 2004) (TC )1 Mbps Commit $125 per Mbps
10 Mbps Commit $80 per Mbps100 Mbps Commit $45 per Mbps
1000 Mbps Commit $30 per Mbps
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!”
Compare 1M vs. 10M Commit
Mbps
$/Mbps at 1M Commit
$/Mbps at 10M Commit
1 $125 $8002 $125 $4003 $125 $2674 $125 $2005 $125 $1606 $125 $1337 $125 $1148 $125 $1009 $125 $89
10 $125 $8011 $125 $8012 $125 $8013 $125 $8014 $125 $8015 $125 $80 Graphically…
Comparing 1M and 10M Commit
$0
$100
$200
$300
$400
$500
$600
$700
$800
$900
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
$/Mbps at 1M Commit$/Mbps at 10M Commit
Transit Breakeven Point 6.4 Mbps
Compare the heuristics…
Comparing Transit Commit Heuristics
Mbps
$/Mbps at 1M Commit
$/Mbps at 10M Commit
100% Heuristic
Leo Heuristic
Norton Transit Heuristic
1 $125 $800 $125 $125 $1252 $125 $400 $125 $125 $1253 $125 $267 $125 $125 $1254 $125 $200 $125 $125 $1255 $125 $160 $125 $125 $1256 $125 $133 $125 $125 $1257 $125 $114 $125 $125 $1148 $125 $100 $125 $125 $1009 $125 $89 $125 $125 $89
10 $125 $80 $80 $125 $8011 $125 $80 $80 $125 $8012 $125 $80 $80 $125 $8013 $125 $80 $80 $125 $8014 $125 $80 $80 $125 $8015 $125 $80 $80 $80 $80
Cost Savings of Optimal Transit
Optimal Transit : 1M commit v. 10M commit
$0.00$20.00$40.00$60.00$80.00
$100.00$120.00$140.00
1 6 11 16 21 26 31 36 41 46
Mbps
$/M
bps
Leo Heuristic(70%)
100% Heuristic
Norton Heuristic
Same effect between 10 and 100M commits…
1M vs 10M vs 100M CommitsTransit (1M v 10M v 100M)
$0
$20
$40
$60
$80
$100
$120
$140
$160
30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
105
110
115
120
125
130
Mbps
$/M
bps 56.25Mbps
1M Commit
100M Commit
Prefer 100M Commit
Algebraically find the crossover point…
Norton Heuristic
MbpsT
MbpsT
MbpsT
MbpsTiceTransit
Mbps
Mbps
Mbps
Mbps
Commit
/30$
/45$
/80$
/125$Pr
1000
100
10
1
)(
)1()1(
)1(
)1(
)(
)(
)(
)1()1()(
)1()1()()(
*)1(
Pr
Pr
*
**
c
ccc
c
c
c
c
BEc
ccc
cccc
TVT
iffVClCommitLeve
MinTransitVolNextCommitV
iceTransitNextCommitT
nsitVolumeCurrentTraV
icensitCurrentTraT
TTVT
V
VTVTtCostNextTransinsitCostCurrentTra
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)
Transit Breakeven PointsTransit vs. Optimal Transit
$0
$20
$40
$60
$80
$100
$120
$140
1 141 281 421 561 701 841 981 1121
Mbps
$/M
bps
CommitPrice
$/Mbps at OptimalTransit
6.4M
56Mbps
666Mbps
100M Commit
1000M Commit
The Peering ModelPeering 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?...
Peering versus (Optimal) Transit1) Normalize
the Peering and Transit units
Peering vs. Transit
Traffic Exchanged (Mbps)
$/M
bps
Fixed Monthly Cost of Peering
Metered Transit
Peering Breakeven Point (Peering=Transit)
$/Mbps (per month)
Vs.
Mbps ExchangedIn Peering or Transit Relationship(Per month)
All as 95th Percentile measures
Peering vs. Transit at TV
2) Compare Peering at PV and Transit at TV
Sample Blended Peering w/Optimal Transit vs. Optimal Transit(Assume 20% peering in Blend)
Peering (20%) Mbps
Peering at PV
Transit (80%) Mbps
Transit at TV
Peering and Transit Blend
Transit Only Mbps
Optimal Transit (100%)
Monthly Cost Savings
25 $360.00 100 $60.00 120.00$ 125 45.00$ -$9,37550 $180.00 200 $45.00 72.00$ 250 45.00$ -$6,75075 $120.00 300 $45.00 60.00$ 375 45.00$ -$5,625
100 $90.00 400 $45.00 54.00$ 500 45.00$ -$4,500125 $72.00 500 $45.00 50.40$ 625 45.00$ -$3,375150 $60.00 600 $45.00 48.00$ 750 40.00$ -$6,000175 $51.43 700 $45.00 46.29$ 875 34.29$ -$10,500200 $45.00 800 $45.00 45.00$ 1000 30.00$ -$15,000225 $40.00 900 $44.44 43.56$ 1125 30.00$ -$15,250250 $36.00 1000 $40.00 39.20$ 1250 30.00$ -$11,500275 $32.73 1100 $36.36 35.64$ 1375 30.00$ -$7,750300 $30.00 1200 $33.33 32.67$ 1500 30.00$ -$4,000325 $27.69 1300 $30.77 30.15$ 1625 30.00$ -$250350 $25.71 1400 $30.00 29.14$ 1750 30.00$ $1,500375 $24.00 1500 $30.00 28.80$ 1875 30.00$ $2,250400 $22.50 1600 $30.00 28.50$ 2000 30.00$ $3,000425 $21.18 1700 $30.00 28.24$ 2125 30.00$ $3,750450 $20.00 1800 $30.00 28.00$ 2250 30.00$ $4,500475 $18.95 1900 $30.00 27.79$ 2375 30.00$ $5,250500 $18.00 2000 $30.00 27.60$ 2500 30.00$ $6,000525 $17.14 2100 $30.00 27.43$ 2625 30.00$ $6,750550 $16.36 2200 $30.00 27.27$ 2750 30.00$ $7,500575 $15.65 2300 $30.00 27.13$ 2875 30.00$ $8,250 100M Peering vs. Optimal Transit at EPB
$0.00
$10.00
$20.00
$30.00
$40.00
$50.00
$60.00
$70.00
$80.00
$90.00
$100.00
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70
Mbps Exchanged
Effective Peering [email protected]
Transit
Example: Peering vs. Optimal TransitPeering Cost: $9000/Month
MbpsT
MbpsT
MbpsT
MbpsTiceTransit
Mbps
Mbps
Mbps
Mbps
Commit
/30$
/45$
/80$
/125$Pr
1000
100
10
1
Blend Peering and Optimal Transit vs. Optimal Transit (assuming only 20% Peering is attainable)
$-
$20.00
$40.00
$60.00
$80.00
$100.00
$120.00
$140.00
Mbps
$/M
bps
Peering and Transit Blend
Optimal Transit (100%)
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
• Peering Port• Incremental Peering HW
Peering Equipment costs…
Peering Equipment CostsUsed Peering Equipment100M Peering EquipmentFoundry Big Iron 4000 w/24 FEs $2,4003yr Straight Line Depreciation $66.67 per month
1000M Peering EquipmentCisco 12008, loaded, with 2*GE $7,0003yr Straight Line Depreciation $194.44 per month
New peering Equipment1000M Peering EquipmentCisco 7301 $13,0003yr Straight Line Depreciation $361.11 per month
High End 1000M Peering EquipmentJuniper M7i, loaded, dual gigEs $28,9253yr Straight Line Depreciation $803.47 per month
Sources: Richard Steenbergen (nLayer)And Patrick Gilmore (Akamai)
Model I – Build into IX for 100M Peering
Model I : Monthly 100M Peering CostsTransport 100 Mbps 1,000.00$
IX Fees (Port + 1/2 rack) 100 Mbps 2,500.00$ Base IX Participation Peering Costs 3,500.00$
Amortized used Foundry Equipment Cost 66.67$
Total Monthly Cost of 100M Peering 3,566.67$ Local Loop Assumptions:100M Transport into IX: $1000 per month1000M Transport into IX: $4000 per month
Graphically: P vs.T…
Peering Rack & Port Assumptions:100M w/ ½ rack at IX: $2500 per month1000M w/ ½ rack at IX: $5000 per month
Model I – 100M Peering100M Blended Peering vs. Optimal Transit
$0.00$500.00
$1,000.00$1,500.00$2,000.00$2,500.00$3,000.00$3,500.00$4,000.00
1 28 55 82 109 136 163 190 217
Mbps
$/M
bps Peering
Optimal Transit
100M Peering at the EPB100M Peering vs. Optimal Transit at EPB
$0.00
$10.00
$20.00
$30.00
$40.00
$50.00
$60.00
$70.00
$80.00
$90.00
$100.00
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70
Mbps Exchanged
$/M
bps
Effective Peering [email protected]
Peering
Transit
$50.59
$45/Mbps
Why peer when P>T?
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.
100M Peering RevisitedModel I 100M Peering vs. Optimal Transit Only
$0.00
$10.00
$20.00
$30.00
$40.00
$50.00
$60.00
$70.00
$80.00
$90.0060 78 96 114
132
150
168
186
204
222
240
258
276
294
Mbps
$/M
bps Blended Peering
optimal transit only
Add FastE Ports
Commit to 1000M Transit
Assuming only 20%, at 33% peering is closer call. 4*100M=1000M, Peering at 1G…
Model I – Build into IX for 1000M Peering
Model I : Monthly 1000M Peering CostsTransport 1000 Mbps 4,000.00$
IX Fees (Port + 1/2 rack) 1000 Mbps 5,000.00$ Base IX Participation Peering Costs 9,000.00$
Amortized used Cisco 12008 Equipment Cost 194.44$
Total Monthly Cost of 1000M Peering w/used Cisco12008 9,194.44$
Graphically P vs T…
Peering Breakeven AnalysisModel I - 1000M Peering vs. Optimal Transit
$0.00
$10.00
$20.00
$30.00
$40.00
$50.00
$60.00
$70.0015
0
207
264
321
378
435
492
549
606
663
Mbps
$/M
bps
Optimal Transit(@4xP)
Model I 1000MPeering
Peering=Transit
306 Mbps
New vs. Used equipment?
Model I – 1000M Peering – new vs. used Equipment doesn’t matter
Peering 1000M: Comparison Between Used and New EquipmentTotal Monthly Cost of 1000M Peering w/used Cisco12008 9,194.44$ per MonthMinimum Cost of Peering (at 705Mbps) 13.04$ per MbpsPeering Breakeven Point (w/$30/Mbps transit) 306.5 Mbps
Total Monthly Cost of 1000M Peering with new Cisco 7301 9,361.11$ per MonthMinimum Cost of Peering (at 705Mbps) 13.28$ per MbpsPeering Breakeven Point (w/$30/Mbps transit) 312.0 Mbps
Total Monthly Cost of 1000M Peering w/new Juniper 7i 9,803.47$ per MonthMinimum Cost of Peering (at 705Mbps) 13.91$ per MbpsPeering Breakeven Point (w/$30/Mbps transit) 326.8 Mbps
Project forward – when gigE fills…
Adding 2nd gigE for PeeringPeering 2nd gigE
$0.00
$5.00
$10.00
$15.00
$20.00
$25.00
$30.00
$35.00
$40.00
$45.00
$50.0020
5
300
395
490
585
680
775
870
965
1060
1155
1250
1345
Mbps
$/M
bps Optimal Transit (@4xP)
Model I 1000M Peering
Peering Breakeven @306Mbps
2nd gigE
When does 10GE peering make sense?
10GigE Peering ModelModel I : Monthly 10-gigE Peering CostsTransport 10000 Mbps 16,000.00$
IX Fees (Port + 1/2 rack) 10000 Mbps 12,000.00$ Base IX Participation Peering Costs 28,000.00$
Amortized Cisco 6509, 4*10G, 4xXENPAK $1,527.78
Total Monthly Cost of 10000M Peering 29,527.78$
Graphically P v T…
Model I – gigEs vs 10gigE1G & 1G-10G vs Optimal Transit
$0.00
$20.00
$40.00
$60.00
$80.00
$100.00
$120.00
$140.00
$160.00
$180.00
$200.0050 485
920
1355
1790
2225
2660
3095
3530
3965
4400
4835
5270
5705
6140
6575
7010
Mbps
$/M
bps Optimal Transit (@4xP)
Model I 1000M Peering
Model I 1-10G Peering
Summary stats for Model I 10G Peering
10gigE Peering MetricsSummary Stats for Model I Peering 10000M w/new 6509
Effective Peering Bandwidth 7,050.0 MbpsMinimum Cost of Peering (at 7050Mbps) 4.19$ per MbpsPeering Breakeven Point (20% Peering) 306.48 MbpsEffective Peering Range 306.5 until 7050.00 MbpsAt the 1410 before using 2nd gigE 20.94$ per Mbps
Traffic exchange for less than $5/Mbps and covering all costs of peering!
Model II – What if I already have colo at an IX?
Model II – Already at IX Colo
Model II - 100M PeeringModel II : Monthly 100M Peering CostsTransport -$
IX Port Fees (rack present) 100 Mbps 1,500.00$ Base IX Participation Peering Costs 1,500.00$
Equipment presentAssume 24 Fast Ports already installedTotal Monthly Cost of 100M Peering 1,500.00$
Peering Port Fee Assumptions: 100M Port only: $1500 per month1000M Port only: $4000 per month
Graphically: P vs T…
Model II – Peering 100M PortsModel II+100M Peering
$-
$10.00
$20.00
$30.00
$40.00
$50.00
$60.0030 85 14
0
195
250
305
360
415
470
525
580
635
690
745
800
855
910
965
1020
1075
1130
1185
1240
1295
1350
1405
1460
1515
1570
1625
1680
Mbps
$/M
bps
Optimal Transit
Peering with FastE cards
Graphically 1000M and 10G Peering…
Model II – 1000 and 10G PeeringModel II : Peering vs. Transit
$0.00
$10.00
$20.00
$30.00
$40.00
$50.00
$60.00
73 383
693
1003
1313
1623
1933
2243
2553
2863
3173
3483
3793
4103
4413
4723
5033
5343
5653
5963
6273
6583
6893
Mbps
$/M
bps Optimal Transit
Model II : Peering gigEsModel II : Peering gigEs to 10GE
$1.92/Mbps
Model II Summary StatsSummary Stats for Model II Peering 10000M w/new Cisco 6509
Effective Peering Bandwidth 7,050.0 MbpsMinimum Cost of Peering (at 7050Mbps) 1.92$ per MbpsPeering Breakeven Point (20% Peering) 89.81 MbpsEffective Peering Range 89.8 until 7050.00 MbpsAt the 1410 before using 2nd gigE 9.59$ per Mbps
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
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.