A Voter Model Supporting Intrusion-tolerance for Network Distance Estimation

To enhance the survivability of Network Coordinate System (NCS) in un-trusted environment, the physical meaning of anchor nodes spring force in classic model is re-explained, weight vector is taken for anchor nodes reputations instead of their prediction errors. Thus a voter model is proposed for network distance prediction and this model is categorized as a kind of method to solve a l1-loss function minimizing problem. By taking the objective functions non-differentiability into consideration, the incremental sub-gradient descending algorithm is used to minimize this function, and a proportional regulator is used to control the iterative step factor with negative feedback. The experiments show that the proposed model is more accurate than classic model in trusted environment with acceptable computing cost. Furthermore, it can also estimate network distance with moderate accuracy in serious un-trusted environment, and shows a stronger intrusion-tolerance capability than classic model.