On Cooperative Settlement Between Content, Transit and Eyeball ISPs Richard T.B. Ma Columbia University Dah-ming Chiu, John C.S. Lui The Chinese University of Hong Kong Vishal Misra, Dan Rubenstein Columbia University Outline • Current ISP Settlement Problems • ISP Models • Profit Sharing Among ISPs • Implications A view of Internet Service Providers (ISPs) • The Internet is composed of Autonomous Systems (ASes). • An ISP is a business entity. – Might comprise multiple ASes. – Provide Internet access. – Objective: maximize profits. ISP Different classes of ISPs • Eyeball ISPs – Provide Internet access to customers: – Place Large investment in infrastructure. – E.g. AT&T, Verizon … • Content ISPs – Provide contents via the Internet. – Serve customers like: • Transit ISPs – Tier 1 ISPs: global connectivity of the Internet. – Provide transit services for other ISPs. – Cover a large geographic area. Current ISP settlements Customer-Provider Settlement Zero-Dollar Peering Settlement Transit ISP Transit ISP Transit ISP Content ISP Eyeball ISP ISP positions on current settlement Transit Not enough revenue to recover investments. Other ISPs are free-riding on our facilities. Eyeball Content Providers Home-users’ monthly fees do not cover costs. We should be able to generate more revenue. We have paid our fair share for transit and delivery and buy bandwidth from ISPs. Issues of the current ISP settlements Net Neutrality Debate: Provide Content-based Service Differentiation ? Yes No Eyeball Transit Content Providers Network Balkanization: De-peering between ISPs Transit Transit zero-dollar peering Level 3 Cogent How to appropriately share profits amongst ISPs? Contribution of this work • Modeling of ISPs – How the revenues are generated – How different kinds of ISPs interact with one another • Profit Sharing Solution Among ISPs – Efficiency – Fairness – Uniqueness • Implications on Bilateral Settlements – Why the current settlements failed – What kind of new settlements should emerge The Network Model: Eyeball Side r=$ • Geographic Regions (r) • Per Customer Monthly Charge (ar) a$ B1 US B2 X$ • Customer Size (Xr) r=£ • Eyeball ISP (Bj) a₤ B2 UK • Revenue from a region r (arXr) B3 X₤ a$ X$+a₤ X₤ Eyeball Side Demand Assumption US • Elastic intra-region demand – Customers can switch among ISPs within a region. – New eyeballs may take customers from other eyeballs in the same region. – Customers move to other eyeballs when the original eyeball leaves the system. • Inelastic inter-region demand – Customers cannot switch to ISPs in other regions. – Constant customer size in a region. B1 B2 B3 X$ UK B3 X₤ The Network Model: Content Side • Content Items (q) {♫, ♣} {♫} • Content ISP (Ci) b♫ • Per Customer Revenue for content q (bq) • Content-side Revenue for uploading content q to region r (bqXr) C1 X$ {♣} C2 b♣ {♫,♣} C3 (b♫ +b♣)(X$+X₤) X₤ a$ X$+a₤ X₤ How to share profits amongst ISPs? How to share profit? -- the baseline case b ♫ X$ {♫} C1 US a $ X$ B1 X$ • One content and one eyeball ISP. • One region, US, and one content, ♫. Profit generated: v=(a$+b♫)X$ • Egalitarian profit sharing: 1 j(B )=j(C ) = 2 v How to share profit? -- multiple eyeballs US b ♫ X$ {♫} B1 a $ X$ C1 B2 X$ • Symmetry: symmetric eyeball ISPs get the same profit. • Efficiency: summation of all ISPs’ profit equals v. Unique solution j(C ) +2 j(B ) = v (Shapley value) 2 j(C )= v 3 • Fairness: same mutual contribution for any pair of ISPs. 1 j(C ) - 2 v = j(B ) - 0 1 j(B )= 6 v Properties of Shapley Value The Shapley Value Shapley 1953 Efficiency Symmetry Dummy Additivity Myerson 1977 Efficiency Symmetry Fairness Symmetry Strong Monotonicity Young 1985 Efficiency Shapley 1977 CoNEXT ‘07 Routing Incentive Solution Stability Interconnecting Incentive How to share profit? -- multiple eyeballs US b ♫ X$ {♫} B1 a $ X$ C1 B2 X$ n eyeball ISPs. Bn • The unique solution (Shapley value) that satisfies Efficiency Symmetry and Fairness: j(B )= 1 v, j(C ) = n v n(n+1) n+1 Results and implications of profit sharing 1 j(B )= v, j(C ) = n v n(n+1) n+1 • The more eyeballs, the more profit the content ISP gets. US {♫} B1 C1 – Elastic users move between eyeball ISPs. – Multiple eyeball ISPs provide redundancy; – The only content ISP has more leverage. Bn-1 n-1 v • When one eyeball leaves the system: j’(C )= n • The marginal profit loss of the content ISP: Dj(C )= n-1 v - n v = - 12 j(C ) n n+1 n – If n=1, the content ISP loses everything if the eyeball leaves. – The content ISP loses only 1/n2 of its original profit. Bn How to share profit? -- multiple contents {♫} C1 {♫} b ♫ X$ a $ X$ US B1 C2 X$ {♫} Cm m content ISPs. • The unique solution (Shapley value) that satisfies Efficiency Symmetry and Fairness: j(C )= 1 v, j(B ) = m v m(m+1) m+1 Results and implications of profit sharing 1 j(C )= v, j(B ) = m v m(m+1) m+1 • The more contents, the more profit the eyeball ISP gets. – Content can be obtained by any content ISP. – Multiple content ISPs provide redundancy; – The only eyeball ISP has more leverage. • The marginal profit loss of the eyeball ISP: {♫} C1 {♫} US B1 C2 {♫} Cm Dj(B )= - 12 j(B ) m – If m=1, the eyeball ISP loses everything if the content leaves. – The eyeball ISP loses only 1/m2 of its original profit. Profit share -- multiple eyeballs and contents {♫} C1 US B1 a $ X$ {♫} b ♫ X$ B2 C2 X$ {♫} Cm Bn • The unique solution (Shapley value) that satisfies Efficiency Symmetry and Fairness: m n j(B )= v, j(C ) = v n(n+m) m(n+m) Results and implications of ISP profit sharing C1 B1 C2 B2 Cm Bn m v n v j(B ) = n (n+m) , j(C ) = m (n+m) • Each ISP’s profit is – Inversely proportional to the number of ISPs of its type. – Proportional to the number of ISPs of the opposite type. • Intuition for elastic demand and supply – The more of the same kind provide redundancy. – The less of a kind can obtain more leverage. Profit share -- eyeballs, transits and contents {♫} US C1 T1 B1 a $ X$ {♫} b ♫ X$ C2 T2 B2 X$ {♫} Cm Tk m k n k m n v j(B )= S k=1 S n+m+k m=1 v j(C )= S k=1 S n+m+k n=1 v j(T )= S n=1 S n+m+k m=1 Bn m k n+m+k-1 (m ) (k ) ( m+k ) n k n+m+k-1 (n ) (k ) ( n+k ) m n n+m+k-1 (m ) (n ) ( m+n ) Profit share -- eyeballs, transits and contents • Intuition – The more of the same kind provide redundancy. – The less of a kind can obtain more leverage. m k n k m n v j(B )= S k=1 S n+m+k m=1 v j(C )= S k=1 S n+m+k n=1 v j(T )= S n=1 S n+m+k m=1 m k n+m+k-1 (m ) (k ) ( m+k ) n k n+m+k-1 (n ) (k ) ( n+k ) m n n+m+k-1 (m ) (n ) ( m+n ) Profit share -- multiple regions and items US {♫} C1 B1 T1 a $ X$ {♣} (b♫ +b♣) (X$+X₤) C2 T2 UK B2 a ₤ X₤ {♫,♣} C2 T3 • Revenue sources are separable X$ B3 X₤ a ₤ X₤ a $ X$ – Eyeball-side components: – Content-side components: b♫X$ b♣X$ b♫X₤ b♣X₤ Profit share -- multiple regions and items US {♫} C1 B1 T1 a $ X$ {♣} (b♫ +b♣) (X$+X₤) b ♣X$ C2 T2 UK B2 a ₤ X₤ {♫,♣} C2 T3 X$ B3 X₤ • A specific revenue component is shared by – Content ISPs that provide the item – Eyeball ISPs that generate the revenue – Transit ISPs that help the delivery Profit share – general topologies 1 ji(N, v) = [S ji(N \{j}, v) + v(N )1{i is veto}] |N | j≠i jC1 = 0 v = b♫ X$ {♫} C1 US C1 B1 T1 B2 T1 B2 C1 is Veto. C1 B1 X$ jC1 = 1/3v C1 B1 T1 B2 B1 jC1 = 1/3v T1 B1 B2 C1 T1 Dynamic Programming Procedure! 5 1 1 1 jC1(N, v) = [0 + v + v + v] = v 12 3 3 4 B2 Implications – the value chain B C T $ CR C1 $ $ $ $ B1 $ T2 T1 BR $ $ $ B2 $ $ $ C2 $ C3 T4 $ T3 $ $ $ B2 $B 3 Implications – the value chain B C T C1 CR $ T2 $ $ B1 T1 BR $ $ $ B2 $ $ $ C2 T4 $ C3 T3 $ $ B2 B3 • Revenue Flows – Content-side revenue (CR): Content Transit Eyeball – Eyeball-side revenue (ER): Eyeball Transit Content Implications – equivalent bilateral settlements Customer C C1 CR T $ $ B1 T1 T4 $ BR $ $ $ B2 Zero-dollar Peering C2 Customer $ T2 $ $ $ C3 B Provider $ $ B2 T3 B3 • When CR ≈ BR, bilateral implementations: – Customer/Provider: Contents & Eyeballs are customers. – Zero-dollar Peering: Transit ISPs peer with each other. – Stable structure for homogenous local ISPs 10 years ago. Implications – equivalent bilateral settlements CR $ $ $ $ B $ $ $ C T C1 $ $ $ $ $ $ $ T2 Customer T1 $ B1 BR Provider B2 $ $ $ $ $ $ C2 Paid Peering T4 $ C3 $ $ T3 B2 B3 • If CR >> BR, bilateral implementations: – Reverse Customer/Provider: Transits compensate Eyeballs. – Paid Peering: content-side compensate eyeball-side. – New settlements are needed to sustain a stable structure. Summary • Content-Transit-Eyeball ISP model – Customer demand, revenue generation. – Closed-form Shapley value for regular topologies. – Dynamic Programming for general topologies. • Implications for current bilateral settlements – Transit ISPs might need to compensate Eyeball ISPs, which creates a Reverse Customer/Provider settlement. – Paid Peering settlement might exist among Transit ISPs. • Guideline for – Government: make regulatory policy for the industry. – ISPs: negotiate stable and incentive settlements.