A Simulation-based Evaluation of a Hybrid Storage System combining - PowerPoint PPT Presentation

. . A Simulation-based Evaluation of a Hybrid Storage System combining P2P, F2F, and Cloud storage with a Distributed Reputation System Anders Skoglund andsk668@student.liu.se November 04, 2013

. Cloud storage P2P storage F2F storage Hybrid storage + reputation system A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System . Storage methods

. . Storage methods Cloud storage Advantages Disadvantages A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System + Scalable + High availability + Contractual accountability − Cost − Possible single point of failure

. . Storage methods P2P storage Advantages Disadvantages A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System + Large number of peers + Scalable + No single point of failure + Cooperative / low cost − Semi-anonymous peers − No accountability − Peers can’t be trusted

. . Storage methods F2F storage Advantages Disadvantages A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System + No single point of failure + Cooperative / low cost + Social accountability + Known and trusted peers − Few peers − Possible availability issues

. . Storage methods Hybrid storage + reputation system Advantages Disadvantages A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System + Scalable + Trusted friend peers + Predictable peer behavior + No single point of failure + Cooperative / low cost − ?

. . System Overview Components DHT Distributed search Block distributor Reputation system A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System

. . Reputation systems Simple reputation systems A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System

. . A. Skoglund A Simulation-based Evaluation of a Hybrid Storage System Distributed reputation systems Reputation systems 1 1 1.0 1.0 1.0 1.0 1 1 1 1 1 1 1 0 1 0.5 1.0 1.0 0.5 1 1 1 0 0 0 1 1 1 0.3 0.1 0.0 0.3 1 1 0 0 0 0 0 1 1 1 1 1.0 1.0 1.0 1.0 1 1

. repeat Threshold used to stop the algorithm once it converges. 1: function ComputeTrust ( C ) 2: . 3: 4: 5: Weight given to pre-trusted peers when computing global trust. p 6: 7: 8: 9: 10: end function A. Skoglund : p : Vector of relative trust values for all pre-trusted peers. Reputation systems EigenTrust Algorithm 1 BasicEigenTrust C : Local trust matrix. a p : A Simulation-based Evaluation of a Hybrid Storage System t : Vector of global trust values for all peers. # » # » ϵ t 0 ← # # » » k ← 0 t k +1 ← (1 − a ) C T # t k + a # # » » » t k +1 − # δ ← || # » » t k || k ← k + 1 until δ < ϵ return # » t k

. 9: 13: 12: 11: d 10: repeat end for d 8: . 6: 5: 4: 3: 14: 15: 1: function ComputeTrust ( C ) 21: A. Skoglund 24: end function end for 23: d 22: end for end for 20: j 19: 18: 17: 16: 2: 7: Threshold used to stop the algorithm once it converges. Vector of relative trust values for all pre-trusted peers. Peers that have reported local trust values about a daughter peer d . : A d Vector of global trust values for all peers. : t : : p Local trust matrix. : C Algorithm 2 SecureEigenTrust Distributed EigenTrust Reputation systems B d : Peers that a daughter peer d has reported local trust values about. Daughter peers of the score manager. Weight given to pre-trusted peers when computing global trust. : a All score managers for the peer i . : M i A Simulation-based Evaluation of a Hybrid Storage System : D # » # » ϵ SendLocalTrust( c dj , M j ) for each d ∈ D do SendGlobalTrust( t k +1 , M j ) A d ← ServedByDaughter( d ) B d ← HasServedDaughter( d ) for each j ∈ A d do k ← 0 M j ← Hash( j ) for each j ∈ A d do c jd ← RecieveLocalTrust( M j ) c jd ← QueryLocalTrust(Hash( j )) t k +1 ← RecieveGlobalTrust( M j ) t k +1 k ← k + 1 ← (1 − a ) ∑ n j + ap d j =1 c jd t k until | t k +1 − t k d | < ϵ for each j ∈ B d do M j ← Hash( j )

. otherwise A. Skoglund i . A Simulation-based Evaluation of a Hybrid Storage System Reputation systems EigenTrust s ij = sat ( i , j ) − unsat ( i , j ) max ( s ij , 0) { if ∑ j max ( s ij ) ̸ = 0 ∑ j max ( s ij , 0) , c ij = p j , + Simple + Well analyzed { 1 if i ∈ P | P | , p i = + Scalable 0 , if i / ∈ P − Very simple trust model − Relative trust values ∑ t j = (1 − a ) c ij t i + ap j

. otherwise A. Skoglund algorithms as EigenTrust Can use the same distributed than EigenTrust Computes more useful trust values t i . A Simulation-based Evaluation of a Hybrid Storage System Reputation systems Modified EigenTrust { sat ( i , j ) if sat ( i , j ) + unsat ( i , j ) ̸ = 0 sat ( i , j )+ unsat ( i , j ) , s ij = 0 , c ij = s ij {∑ if j / ∈ P i c ij w ij , t j = 1 , if j ∈ P w ij = ∑ k ∈ A j t k