REAL-TIME DECENTRALIZED NETWORK TRAFFIC MANAGEMENT USING A PARALLEL ALGORITHM

摘要

90-3-805 REAL-TIME DECENTRALIZED NETWORK TRAFFIC MANAGEMENT USING A PARALLEL ALGORITHM A decentralized, state dependent access-control and routing strategy for real-time control of circuit switched networks. To decentralize the traffic control tasks, we partition the network into subnets and assign a specific traffic controller to each one. The traffic controllers obtain periodic (approximately every 5 minutes) subnet measurements and compute the optimal traffic control policy of the subnet through an iterative and parallel dialog with the other controllers. Each controller implements the new control policy within its subnet until the next measurement epoch, thereby allowing decentralized call handling. In the formulation of the joint accesscontrol and routing problem, we allocate the incoming demand (for a given period), the optimal strategy to maximize the predicted minimum (over all network Trunk Groups) Trunk Group residual capacity. Secondly, when the projected demand can not be accommodated through this strategy, the optimal strategy rejects the extra demand at source, in an "equitable" manner. The mathematical formulation of the above routing and access-control objectives leads to an Equilibrium Programming Problem (EPP). The EPP formulation is decomposed into a number of subproblems and solved--in parallel--by the intercommunicating subnet controllers, thereby satisfying real-time control requirements. The ensemble of the subproblem solutions forms the network-wide (globally) optimal traffic management strategy for the upcoming period.