Currently there is not enough work on the Internet weighted fair bandwidth sharing without per-¯ow management, especially when both UDP and TCP ¯ows of different RTTs and different bandwidth targets coexist. This paper contains two contributions:
- A mechanism called SCALE±WFS, Scalable Core with Aggregation Level labEling Ð Weighted Fair bandwidth-Sharing, is presented to achieve near-optimal Max±Min weighted fairness without per-¯ow management at core routers, in the context of differentiated service (Diffserv) networks. Through extensive simulation and simple analysis, we show the performance problems with current solutions and propose SCALE±WFS to solve these problems. This scheme works effectively for different ¯ow weights, RTTs, protocols (TCP and UDP), under the scenarios of different bandwidth provisioning and multiple congested gateways, all of which we consider the basic requirements for a practical solution. 2. Apart from ªper-¯owº and completely ªcore-statelessº, SCALE represents a third management level Ð the label-based ¯ow ªaggregationº level. It reduces the information redundancy in the ªper-¯owº and surpasses the performance of the ªcore-statelessº. This is important to Diffserv networks where we try to avoid the non-scalability of per-¯ow management while keeping its good Quality of Services. In this paper, we show that SCALE±WFS is effective and scalable, robust and extensible.