A new polynomial-time algorithm for linear programming
Combinatorica
Convex Optimization
Maximum Pressure Policies in Stochastic Processing Networks
Operations Research
Maximizing Queueing Network Utility Subject to Stability: Greedy Primal-Dual Algorithm
Queueing Systems: Theory and Applications
IEEE/ACM Transactions on Networking (TON) - Special issue on networking and information theory
Load shedding and distributed resource control of stream processing networks
Performance Evaluation
Utility maximization in peer-to-peer systems
SIGMETRICS '08 Proceedings of the 2008 ACM SIGMETRICS international conference on Measurement and modeling of computer systems
Proceedings of the ACM SIGMETRICS international conference on Measurement and modeling of computer systems
Distributed resource allocation for synchronous fork and join processing networks
INFOCOM'10 Proceedings of the 29th conference on Information communications
Joint congestion control, routing, and MAC for stability and fairness in wireless networks
IEEE Journal on Selected Areas in Communications
Multirate multicasting with intralayer network coding
IEEE/ACM Transactions on Networking (TON)
| Hi-index | 0.00 |
Many network-based problems are naturally distributed optimization problems. Examples include routing and flow control in wireless sensor networks, congestion pricing on the Internet, and resource allocation in distributed data-processing systems. Most of the distributed solutions provided by existing work utilize algorithms based on the sub-gradient method. Although stability and asymptotic optimality of these algorithms have been extensively studied and confirmed in the literature, the practical convergence speed of these algorithms attracted little attention. Our simulation results show that sometimes it is far below what it is expected to be. This paper presents a generic formulation for distributed optimization that can model all above-mentioned problems, and proposes two classes of distributed algorithms: one using a simple coordinator and the other using network consensus. These algorithms are based on the barrier method and are closely related to the interior-point method for non-distributed optimization problems. Using new techniques, we prove that our distributed algorithms converge linearly to the optimal solution. Simulation experiments demonstrate that the actual convergence speed of these algorithms is much faster than existing distributed algorithms. We further investigate how their algorithmic parameters affect the convergence speed.