1. Exploiting super peers for large- scale peer-to-peer Wi-Fi roaming Efstratios G. Dimopoulos, Pantelis A. Frangoudis and George.C.Polyzos

2. Motivation Very high Wi-Fi density in cities The case for Skyhook Residential Wi-Fi hotspots with excess capacity How to exploit this user-provided infrastructure? We need a Wi-Fi sharing scheme! Can community based Wi-Fi access complement cellular? Very high Wi-Fi density in cities The case for Skyhook Residential Wi-Fi hotspots with excess capacity How to exploit this user-provided infrastructure? We need a Wi-Fi sharing scheme! Can community based Wi-Fi access complement cellular? 2

3. Design options Centralized Permanent IDs Full view of transactions Easy to detect misuse  FON Centralized Permanent IDs Full view of transactions Easy to detect misuse  FON 3 Decentralized Free/disposable IDs Enhances privacy Should discourage misuse  Our approach Decentralized Free/disposable IDs Enhances privacy Should discourage misuse  Our approach

4. 4 Our approach Design principle Users form a club that relies on indirect service reciprocity Distinct characteristics Fully decentralized No user registration Designed with off-the-shelf equipment in mind Does not assume altruists Design principle Users form a club that relies on indirect service reciprocity Distinct characteristics Fully decentralized No user registration Designed with off-the-shelf equipment in mind Does not assume altruists

5. 5 Entities Peer: provides service via home AP, consumes when mobile Peer ID: uncertified public/private key pair Accounting unit: digital receipt Signed by roaming user Proof of transaction Receipt repositories Peer: provides service via home AP, consumes when mobile Peer ID: uncertified public/private key pair Accounting unit: digital receipt Signed by roaming user Proof of transaction Receipt repositories

6. 6 Receipts and the reciprocity algorithm Consuming member Certificate Timestamp Member Signature (Signed with member private key) Weight (amount of bytes relayed) Contributor Public Key Receipt generation AP periodically requests fresh receipt Roamer sends signed receipt Storage Receipt repositories Input to the reciprocity algorithm Algorithm output Indirect Normalized Debt (IND) Translated to QoS Receipt generation AP periodically requests fresh receipt Roamer sends signed receipt Storage Receipt repositories Input to the reciprocity algorithm Algorithm output Indirect Normalized Debt (IND) Translated to QoS

7. 7 Can it scale?

8. 8 The locality of visits Visits to foreign areas are rare IND ≈0 Receipts are unvalued in foreign areas Visits to foreign areas are rare IND ≈0 Receipts are unvalued in foreign areas

9. 9 A Super-Peer-assisted architecture At least one Super Peer per Area Super Peers: Globally known Trusted Without extra computational capabilities

10. 10 An algorithm for large-scale roaming - Specification The algorithm should run for all transactions (not only for roaming ones) Low Complexity As few Super Peers as possible Super peers should be used only when necessary Incentive based Normal users To contribute service to Super Peers To contribute service to roamers Super Peers To mediate other transactions The algorithm should run for all transactions (not only for roaming ones) Low Complexity As few Super Peers as possible Super peers should be used only when necessary Incentive based Normal users To contribute service to Super Peers To contribute service to roamers Super Peers To mediate other transactions

11. 11 Example A user visits a foreign area He asks service from an AP and informs the AP about the SP of his home area The team server runs the reciprocity algorithm According to the result he should not contribute service. So, he asks the SP of his home location to find a guarantor, in order to provide service to the user Simultaneously asks from the user’s home location SP to calculate the same quantity and the waits for the answer. The VSP runs the reciprocity algorithm for the prospective consumer, in order to calculate the quantity (IND) that he is able to guarantee. The HSP runs the reciprocity algorithm in order to calculate the IND for the prospective consumer Informs the VSP for the calculated IND VSP calculates: 1. The final IND for the prospective consumer. 0,2xIND (VSP) + 0,8xIND (HSP) 2.The guarantor SP for this transaction. Informs the HSP (guarantor) and the Team Server of the provider about the IND calculated AP asks receipts from the SP for his own use and also from the consumer on behalf of the SP The consumer signs receipts to the SP and the SP signs receipts to the AP

12. 12 Everyone is happy! Roaming users have consumed service The AP has gained the valuable receipts of the SP The SP helped a member of his area and paid off his debt Roaming users have consumed service The AP has gained the valuable receipts of the SP The SP helped a member of his area and paid off his debt

13. 13 Simulations Input Parameters Server Repository Size Client Repository Size Users Number Areas Number Area Population Roaming probability Number of stay rounds in the foreign area(stop over rounds) Contribution of the super peers to IND Number of super peers per area Input Parameters Server Repository Size Client Repository Size Users Number Areas Number Area Population Roaming probability Number of stay rounds in the foreign area(stop over rounds) Contribution of the super peers to IND Number of super peers per area Output Parameters SW Hit Ratio Requests to the super peers Super peers guarantees Output Parameters SW Hit Ratio Requests to the super peers Super peers guarantees

14. 14 Number of Regions Effect Input Parameters Patience=20 (rounds) Server Repository Size=2000 (receipts) Client Repository Size=300 (receipts) Number of peers=1000 (2x500 - 4x250 - 8x125 - 10x1000 - 20x50) Roaming Start Round=5 Roaming Probability p=0.1 Stop Over Rounds=1 Super peers Participation=80% consumer part. - 20% provider part. Super Peers per Region=1

15. 15 Server Repository Size Effect Input Parameters Patience=20(rounds) Server Repository Size=1000 (250) 3000 Client Repository Size=250 Number of peers=1000 (4x250) Roaming Start Round=5 Roaming Probability p=0.1 Stop Over Rounds=1 Super peers Participation=80% consumer part. - 20% provider part. Super Peers per Region=1

16. 16 Participations of super peers in the IND result Input Parameters Patience = 20 (rounds) Repository size = 2000 (receipts) Receipts to merge = 300 Number of Peers = 1000 ( 4x250) - (20x50) Roaming Start Round = 25 Roaming Probability p = 0.1 Stop Over Rounds = 1 super peers participation=variable Super Peers per Region = 1

17. 17 The effect of the number of super peers per region Input Parameters Patience = 5 (rounds) Repository size = 500 (receipts) Receipts to merge = 100 Number of Peers = 1000 ( 8x125) Roaming Start Round = 1 Roaming Probability p = 0.1 Stop Over Rounds = 1 super peers participation (Home - Visited)=80% - 20% Super Peers per Region = 1,2,3

18. 18 Scale Effect Input Parameters Patience = 20 (rounds) Repository size = 1500 (receipts) Receipts to merge = 250 Number of Peers = (4x250) (10x250) (20x250) (28X250) Roaming Start Round = 1 Roaming Probability p = 0.1 Stop Over Rounds = 2 super peers participation (Home - Visited)=80% - 20% Super Peers per Region = 1

19. 19 THE END Thank you!