Internet Economy and Content PeeringAlcatel -LucentBell Labs Research PaperBill KrogfossNovember 2, 2011
Agenda1. Status of Peering2. Internet Economic Model3. Content Peering4. Peering Ratios5. CDN Economics
Internet Economic Mysteries and Contradictions “Why should Google or other large CP be allowed to dump all this traffic on my network for free?” T2/T3 ISPs are complaining about rising consumer traffic, but T1 ISPs are not complaining Netflix 10K “We believe our financial responsibility is to bring the traffic to ISP doorstep and the ISPs responsibility is to deliver it” Level3 (CDN) vs. Comcast: “ Why should peering ratios be limited?” Content Providers and CDNs : “Content Peering is a win-win proposition for us and ISP” Alcatel-Lucent – Internal3 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
Internet Economics for Transit and Peering Transit pricing has fallen dramatically, and now competes with peering Peering break even point is several Gbps and increasing as transit continues to decline (Can transit price drop be a consequence of peering?) Macro Trend (33K ASN study from 2004-2010) Shows Increased Peering and Transit– vast majority of peering relationships in top 2% (90%) Peering vs. Transit Costs CDF of AS Peers (05-09) 100% *source:Drpeering.net 12.00 No co-location 90% 10.00 80% 8.00 Increased peering Break-even$/Mbps 70% 6.00 Transit rate 4.00 What is impact of peering and transit prices converging?...... 60% ’09-’10 Diminishing Returns? 2.00 50% with co-location - 40%No Colocation 1 2 4 6 8 10 1 2 3 4 5 6 7 8 9 10Peering Costs10G Port $ 3,500 Gbps (inter ISP BW) Num ber of Peers10G Transport $ 5,000Colo $ 2,000Equipment $ 500Total $ 11,000 With Colocation No Colocation 2005 2006 2007 2008 2009 2010 Alcatel-Lucent – Internal 4 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
Peering for Performance MSN Routes Facebook Routes 2007 HP Lab study latency: MSN &Google had 24 ‘degree’ networks, DT NTTFacebook 3 degree gblxMSN FB ST Internet Impact: 50 global 3914 BT 2914 traceroute servers showed 60% of BT NIC TPG routes to large ASPs avoided T1 ISPs (Internet Backbone)* Facebook 300ms vs 157ms Yahoo/Youtube Routes with zero T1 ISPs Jupiter Research - doubling delay 35 can result in one-quarter fewer 30 customers 25 Routes 20 Content Peering provides a 15 significant performance 10 advantage 5 0 d u r e g JP ok a Content Peering provides a significant performance advantage o a e t ce te of oa id lo gl ho ub di sin bo ds os pa ba to oo pe o pl Ya ut ho fo ce en icr au ys G ki yo Ya fa wi fri m m eg m HP Laboratories, May 21, 2008* Alcatel-Lucent – Internal 5 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
Internet Economics Basics Internet Ecosystem: Shared Infrastructure, Shared Revenue & Expenses Content Provider Revenue is shared as Transit Revenue to other ISPs Each ISP fares differently due to its type and routing topology Majority of Revenue for Internet is CP and Subscriber However, Subscriber revenue is fixed (doesn’t vary with traffic), CP varies with traffic No new revenue generated by transit fees Internet Ecosystem (Shared Infrastructure) (for the Internet Ecosystem) Shared Revenues ISP B$10 Generators (K) ISP G ISP F ISP B ISP A $5 $5 ISP F B ISP ISP F ISP A $2 ISP A ($5) $S ISP C ISP C ISP E ISP CG $10 $7 ISP ISP D $3 ISP G ISP D$ K + $S ISP E ISP D Traffic (Mbps) ISP E Consumers (S) Content Revenue literally funds the inter domain links (i.e. Internet) Alcatel-Lucent – Internal7 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
Internet Economic Model (before Content Peering) Ecosystem is ‘conservative’ in nature. No new revenue generated through traffic distribution. Traffic generated is traffic consumed proportional to subscribers Inter-domain Economic Model (focus on transit/peering economics) Scenario 1 (no content peering): Nominal (T1) Transit $5/Mbps, 5K subs/T2 & T3 ISP, sub revenue $30/mo/sub Internet Income View ISP Profit, increasing BW per Content ProviderTier 1 450,000 ISP E 400,000 T1 Transit Rate Profit T1 350,000 300,000 Profit T2 $k ($USD) 250,000 CPC $kTier 2 ISP C ISP D CPD 200,000 150,000 Profit T3 $s $s 100,000 No Content 50,000 Subscriber revenue CPA $k Peering $k -Tier 3 ISP A ISP B CPB 0 10 20 30 40 $s (Gbps) $ sub = s Internet Profit (P) = Content Revenue ( ∑ ki ) +Subscriber Revenue ( ∑ si ) = K + S Alcatel-Lucent – Internal 8 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
Internet Economy with Content Peering Peering often benefits one ISP at the expense of another Net effect of increased peering is reduced profit for the system and most ISPs InternetProfit (P )= ContentRev ( K ) +Sub Rev ( S ) - Peering Costs (C p ) ⇒ P = K + S − C p T1 Reduced Income Internet Profit (P) = ∑ηi (1 − ri )ti + s − ∑ C p rti ISP B CP ‘A’ CP ‘B’ i $$ $$ $ i i ISP A ISP B Internet Profit w/ Increasing Content Peering Reduced Expense Reduced OPEX CP ‘B’ ISP A CP B 1,800,000 Peer 1,600,000 Total Ecosystem Profit Content Peering View 1,400,000 1,200,000 ISP E 1,000,000 800,000 T2 CP C CP D 600,000 ISP C ISP D T3 400,000Peering Candidates For CP B 200,000 T1 CP A CP B - ISP A ISP B No Peer 1 x CP 2 x CP 3 x CP 4 x CP Increasing Content peering Peering Candidates For CP C Alcatel-Lucent – Internal 10 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
Internet Economy with Content Peering Profit of Internet with Increased Peering and Increased Traffic 3.50 What is the ultimate impact on the 3.00 T 40% CAGREcosystem as Traffic gets very large? 2.50 Millions USD 2.00 Pnp − Pp 1.50percentage loss in ecosystem = Pnp 1.00 K + S − (K + S − Cp ) Cp lim = lim = 0.50 t →∞ K +S t →∞ K +S 0.00 No Peer 1 x CP 2 x CP 3 x CP 4 x CP t pcp rc p lim = Increasing Peering t →∞ r (ηβ )t + S (1 − r )(ηβ ) Ultimate loss in Ecosystem with increasing peering Transit cost (t → ∞) 300% Profit Loss in Ecosystem 250% K + S − C p ≥ 0; condition for profitability 200% Cp-> η 150% t (1 − r )(η ) + S − trc p ≥ 0 100% Profit becomes loss S + tη η r≤ ; if t is large than r ≤ 50% tc p + tη cp +η 0% 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 as c p → η , r ≤ 0.5 amount peered away Peering Cost and Transit Cost Converge Cp=η/4 Cp=η/3 Cp=η/2 Cp=η/1.2 Alcatel-Lucent – Internal 11 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
What about increasing Peering Ratios?Today’s Internet is made up of many Content Heavy Networks and Eyeball Heavy Networks(access networks) Peering between ISPs and Content Networks reduces the ability for the Internet to profit from transit income Traffic grows at rate proportional to transit cost ( kt), but when a gets very large it grows at a rate of ( kt − k p ) . Recall that peering and transit costs are converging! Internet Profit Ecosystem with increasing asymmetric peering ratiosP = ktT + S − k p aTo 600kt ; unit cost of transit 500 a=2k p ; unit cost of peering T1 Increasing asymmetry a=20 400 txt (T - aTo) a=40 ($MUSD) 300 kt a:1 a=100 T2 CDN 200 P (peer) T profit To aTo 100 a=100, kt ->kp - 20 40 60 80 100 What happens as ‘a’ gets large? Gbps With increasing asymmetry there is no margins in selling transit ( kt ⎯⎯ k p ) → Alcatel-Lucent – Internal 12 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
Internet Economics: Content Delivery NetworksScenario: Impact of Content Delivery Networks CDNs look to ISPs as large ContentProviders, economically Do CDNs bring new revenue into Ecosystem? CDN creates new Revenue:if CDN rate > CP Transit Rate ( ε > β ) CDN shares in Existing Revenue:if CDN rate ~ CP Transit Rate ( ε ≅ β ) CDN degenerates Revenue:if CDN rate < CP Transit Rate ( ε < β ) Content Peering View ISP E CP A T/P CDN CP n CP C ISP C ISP D CP D CP A T/P CP B ISP A ISP B P = ∑ Pi +impact kof − hi ) + s) + ( εη )∑ tiwithK+S; without Content peering Look at Pn= ∑ ( i (1 CDNs on ISPs hi= and β = ε i i i Alcatel-Lucent – Internal13 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
CDN Conclusions Scenarios 1: CDN Rate ~ CP Transit Rate (with and w/o CDN Peering) Scenario 2: CDN Rate > CP Transit Rate (no CDN peering) Cache Hit Rate (50%), With and Without CDN Peering Impact of CDNs on ISP Economy (CDN rate = Transit) ISP Impact with Increasing CDN Value CDN rate(ε) > Transit Rate (β)2,000,000 2,000,000 Total ISP Profit1,800,000 1,800,0001,600,000 1,600,0001,400,000 CDN Impact 1,400,000 CDN Impact1,200,000 1,200,0001,000,000 1,000,000 800,000 CDN Peering Impact 800,000 600,000 Profit: CDN Peering 600,000 CDN Increasing Value (CDN Rate >> Transit) (ε = 1.6) 400,000 400,000 200,000 Profit: CDN 200,000 (ε = 1.4) - - 10 20 30 40 50 10 20 30 40 50 Gbps Gbps Alcatel-Lucent – Internal 14 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.
ConclusionsBy its very structure the Internet profit is inequitably divided between theinfrastructure providersOriginally Peering provided a method to combat high Transit Costs Today Peering is primarily a performance tool rather than an economic tool – content providers that peer have a significant performance advantageThe consequences of content peering is a real economic loss – difficult toquantify in a highly complex ecosystem, but as peering and transit pricesconverge it becomes worseIncreasing Asymmetry for peering only exacerbates the economic impact ofcontent peeringCDNs postively contribute to Internet Ecosystem only if they are able to showvalue beyond just transit replacement Alcatel-Lucent – Internal15 | Presentation Title | Month Year Proprietary – Use pursuant to Company instruction.