Congestion Management for Packet-Based Communication Networks

• Applicant: Professor Dr. Michael Menth
• Subject Area: Security and Dependability, Operating-, Communication- and Distributed Systems
• Term: from 2016 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 310918795

In this project, new concepts for congestion management in packet-based communication networks, in particular for IP technology, are developed and investigated. Congestion management detects congestion in a network and resolves it. Challenges are delay limitation, high resource utilization, and fair bandwidth partition in case of congestion.The Internet Engineering Task Force (IETF) currently develops two different protocols to signal congestion from the network to its edge to facilitate efficient congestion management. In this context, the idea of congestion policing (CP) is discussed. CP implies that packets of a traffic aggregate should be dropped if that aggregate causes more congestion than a pre-assigned allowance. However, the exact definition of congestion for this purpose is still unclear and CP has been treated only qualitatively. There are only vague algorithmic approaches and none of them has been investigated. Therefore, this project develops and investigates CP as novel approach for congestion management. Event-based simulation and mathematical modelling and analysis will be applied as research methods. As the problem is difficult, the project will tackle the problem in three steps that are described in the following.Active queue management (AQM) mechanisms reduce queueing delay in communication networks by avoiding standing queues on outgoing interfaces. Currently, new AQM algorithms are discussed in the IETF. In this project, a novel CP-based AQM mechanism should be developed, evaluated, and compared with the other AQM algorithms. In a second step, this CP-based AQM will be extended in such a way that it shares the bandwidth of a link fairly among traffic aggregates, e.g., of multiple users, in a similar way as scheduling algorithms like weighted fair queueing do. While these approaches utilize only congestion information from a local queue, this information will be substituted by signaled congestion feedback from the network in a third step. The objective is to have a congestion policer at the network edge that detects congestion in the network from signalled feedback and controls the congestion in such a way that the bandwidth of distant congested links is fairly shared among traffic aggregates.This novel congestion management approach is scalable because it leaves the core network stateless unlike scheduling algorithms that require per-aggregate information on controlled links. The developed congestion management algorithm is primarily conceived for trusted networks like data centers or the backhaul of mobile networks. It will be evaluated for these use cases and compared with other approaches. The results of the project will be reported to the IETF to contribute to an appropriate definition of congestion, appropriate congestion feedback protocols, and to efficient congestion management mechanisms.

• DFG Programme: Research Grants