Scheduling nonuniform traffic in a packet-switching system with small propagation delay
IEEE/ACM Transactions on Networking (TON)
Proceedings of the 11th Colloquium on Automata, Languages and Programming
A methodology for estimating interdomain web traffic demand
Proceedings of the 4th ACM SIGCOMM conference on Internet measurement
Enabling distributed throughput maximization in wireless mesh networks: a partitioning approach
Proceedings of the 12th annual international conference on Mobile computing and networking
A local greedy scheduling scheme with provable performance guarantee
Proceedings of the 9th ACM international symposium on Mobile ad hoc networking and computing
Achieving 100% throughput in an input-queued switch
INFOCOM'96 Proceedings of the Fifteenth annual joint conference of the IEEE computer and communications societies conference on The conference on computer communications - Volume 1
Scheduling and performance limits of networks with constantly changing topology
IEEE Transactions on Information Theory
| Hi-index | 0.00 |
This paper takes a philosophically new approach to throughput-optimal scheduling queueing systems with interference. All existing popular approaches (e.g. max-weight, greedy, "pick-and-compare" etc.) focus on the weights of individual queues. We take an alternative approach, by focusing instead on the aggregate queues of bottlenecks. A bottleneck is a set of mutually-interfering queues; a schedule drains a bottleneck if it removes a packet from any one of its queues. We consider (the standard) switch scheduling problem, where the bottlenecks are the nodes. We establish the following phasetransition (1) ensuring only that the very heaviest nodes are drained is not enough for throughput optimality, but (2) ensuring scheduling for all nodes with weight within (1 - α) of the heaviest is enough for throughput optimality, for any α 0. The proof uses a new Lyapunov function: the weight of the critical bottleneck. Our alternate node-focused view also enables the development of new algorithms for scheduling. We show (a) how any policy can be made throughput-optimal by doing a small number of extra operations, (b) a new algorithm - Maximum Vertex-weighted Matching (MVM) - has (empirical) delay performance better than the current state of the art, and lower complexity than Max-(edge) weighted Matching, and (c) a class o f throughputoptimal policies that trade off between complexity and delay.