Topic: P2P Trading in Social Networks: The Value of Staying Connected The purpose of this paper is to propose a P2P incentive paradigm named Networked Asynchronous Bilateral Trading (NABT)

Outline  General introduction of Global currency, Bartering and NABT  NABT model Basic concepts of NABT Modeling of NABT Credit transfer routing in NABT  Simulation analysis Credit limit setting Simulation setup Simulation results

General introduction of Global currency, Bartering and NABT  Global currency ( 货币 ) An efficient incentive mimics real-world currency and transactions Disadvantage:  Requires high degree of coordination: central bank rely on a well-functioning legal system that enforces contracts and resolves disputes  Need to built reputation against inflation ( 通胀 )

General introduction of Global currency, Bartering and NABT  Bartering( 以货易货 ) Peers bilaterally and synchronously trade resources. Advantage: simplicity, no need for currency, banks, dispute-resolution mechanisms or bank regulation. Disadvantage: synchronous; difficult to trade long-tail contents between peers.

General introduction of Global currency, Bartering and NABT  NABT: Peers belong to a social network, with each peer having a set of friends. Each pair of friends maintains a credit balance and a credit limit.  Advantage: Asynchronism over time and over nodes. Provide service differentiation, exploit trading intermediaries, discourage free-riders.

Basic concepts of NABT  Social network: each peer has a set of friends. Each peer has certain resources.  Credit limits: Each peer i sets a credit limit Cij for each friend j. Cij quantifies the trust between i and j and depends on their past trading history.

Basic concepts of NABT  Credit balance: bij denotes the amount of credits user j owes user i. bij = -bji and – Cji <= bij <= Cij.  Asynchronous trading: i charges j a certain amount of credits. i ’ s credit balance with j increases correspondingly. Service stops if j ’ s resulting debt bij exceeds the credit limit Cij.

Basic concepts of NABT  Trading via intermediaries: peers can use their mutual friends as a credit intermediary to get service.

Modeling of NABT  Gs = (U, F) U: users connected in social network. F: friends relations between users.

Modeling of NABT  G D = (U, D) D: service demand profile among users. For a service demand d = (k, l) ∈ D, k is the provider and l is the consumer. K charge l a cost of h(k, l)for providing the service.

Modeling of NABT  H = [H(k, l)] H is a |U|×|U| demand matrix as H(k, l) = h(k, l) if (k, l) ∈ D, otherwise H(k, l) = 0.  Initially, bij(0) = bji(0) = 0. after i serves j once, balance becomes bij(1) = h(i, j). Then j serves i, balance becomes bij(2) = h(i, j) – h(j, i). – Cji <= bij <= Cij, else i can ’ t get service from j.

Modeling of NABT  Specifically, if l wants to obtain service from k, l first tries to find a path from k to l, p(k,l) = {k = r 0 - >r 1 -> … ->r m = l}. L then transfer credit h(k,l) in the reverse direction: r n pays h(k,l) to r n-1. after transfers, the credit balance on each node is updated.

Credit transfer routing in NABT  Xd : the amount of credit j take form i for service demand d.

Credit transfer routing in NABT  The resulted credit balance between user I and j after all credit transers is bij = ∑ d ∈ D (Xd - Xd ).  So, we have to make sure the resulted credit balances on all links after executing all demands satisfy: -Cji - Xd ) <= Cij.

Simulation analysis  Trading with global currency (GCT): If peer i wants to download from peer j, i pays credit directly to j.  Synchronous trading (ST): Two peers can trade if and only if they can supply files to each other simultaneously.  Two-hop NABT The peer checks whether there are potential supplying peers within the two hops, and whether the paths to these supplying peers can pass sufficient pair wise credits.

Credit limit setting  Additive Increase Multiplication Decrease (AIMD) algorithm: the credit limit on a social link increases by an amount of α after each fulfilled transaction utilizing the link; the credit limit decreases by a factor of β whenever the link is involved in an unfulfilled or disputed transaction. To avoid credit limit explosion, maximum credit limit can also be set.

Simulation setup  Assume file popularity follows a Zipf distribution with the parameter ρ = 0.27; 37% peers with upload bandwidth of 1.2M, 63% of 400K; 10% with 1000 files to share, the others with 50; Each peer generates file download requests by Poisson process every 10 minutes on average.  Request can ’ t be scheduled in 60 minutes would be drop. Each peer can at most manage 100 unscheduled requests. if reaches 100, any newly generated requests will be simply blocked.

Simulation setup  Supplier maintains a serve queue and services requests in FIFO fashion.  If more than one supplier found, select a supplier can serve request with the shortest waiting time.

Simulation results  Request success ratio and waiting time

Simulation results  ST: Lack of common interest in the ST system limits the chance of trading and cause low success ratio and long time to find a trading partner with mutual interest.  GCT: Peers with less credit or low serving capacity result in low success ratio.  NABT: peers not well connected spend more time before their requests be scheduled.

Simulation results  Contribution and relay time vs requests success ratio

Simulation results  Requests success ratio vs Hops and cooperation attitude