Quality of Service and Congestion Management

Fair and Scalable Bandwidth Distribution in the Internet


The current best effort approach to Quality of Service (QoS) in the Internet can no longer satisfy a diverse variety of customer service requirements. We are investigating methods for providing fine-grained per-flow QoS that are scalable and fair, and that do not cause network congestion and keep overall link utilization high. We have designed an efficient, fast, and scalable bandwidth distribution mechanism which fairly distributes available resources among the flows based on their resource requirements. The bandwidth distribution scheme is implemented via a message exchange protocol which incurs low overhead. We are studying the schemes together with alternative fairness mechanisms using simulations in OPNET.

Quality-of-Service for Multimedia Applications


Real-time multimedia applications communicating over a packet-switched network present new challenges to the current networking technology. One of these is a Quality-of-Service (QoS) requirement. Unlike many traditional network applications such as email and file transfer, which only require best-effort service at the network layer, multimedia applications such as video-on-demand, video conferencing, or multimedia virtual presentations demand more strict QoS including explicit delay bounds and limits on packet loss rates. There are no existing schemes to satisfy such QoS requirements. In our research, we have designed algorithms to be implemented in network switches that provide predictive loss ratio and delay bound QoS on a per-flow basis. The algorithms use dynamic resource adjustment in combination with VirtualClock rate-based control based on measured QoS performance to achieve high resource utilization and high QoS satisfaction. Extensive simulation studies have been performed using OPNET to compare the performance of our algorithms with that of more traditional schemes that do not control packet loss. In addition to higher utilization and QoS satisfaction, our algorithms also provide a fairer service to different flows when degradation in service results from inadequate admission control. We have also designed a measurement-based admission control algorithm to accompany our dynamic QoS control algorithms.

(Supported by the Federated Research Laboratory (FRL) Consortium in Advanced Telecommunications/Information Distribution Research Program (ATIRP) sponsored by the Army Research Laboratory (ARL).

Real-Time Protocols for Wireless Communication


When real-time transport of data is needed in wireless networks with random-access protocols, Quality-of-Service (QoS) guarantees must take into account that not all of the offered load can be transmitted. Popular random-access protocols such as Ethernet or its variants are undesirable for real-time settings because of the potentially unbounded access times. Other approaches known as window splitting techniques are more attractive but have not been studied before in the context of a real-time service. We combine the use of laxity, i.e., time until a packet's deadline expires, as the splitting variable with the splitting methodologies of traditional algorithms to create new splitting protocols for real-time use. We have developed many variations on such protocols including blocked access and free access algorithms, and methods for use with mixed hard real-time, soft real-time, and non real-time traffic. We have also developed analytic models for the performance evaluation of these protocols. The results of the analytic models have been validated by simulations models using OPNET. Extensive performance studies comparing the success rates and delays of the different algorithms have been performed. We have also studied scheduling algorithms for use in the LLC Layer in the context of a real-time service to higher layers.

(Supported by the Federated Research Laboratory (FRL) Consortium in Advanced Telecommunications/Information Distribution Research Program (ATIRP) sponsored by the Army Research Laboratory (ARL).

Distributed, Cooperative Congestion Control


Distributed cooperative congestion management schemes for switched high-speed networks have been studied. Of special importance is a scheme for rerouting of selected traffic at the burst level which reacts to congestion as close to its point of origin as possible, and minimizes response times. The scheme uses a progression of proactive control mechanisms on selected traffic bursts: From local rerouting, through upstream rerouting and quenching, to packet discarding. It complements existing open-loop mechanisms by allowing more liberal packet admission while preserving the underlying switching architecture.

Distributed Algorithms for Computing a Global Measure of Congestion


This project has studied distributed algorithms for computing the mean of a set of numbers distributed at various network nodes. Currently, there are no algorithms for this task. We have designed a new algorithm to compute the mean of a distributed set of numbers without knowing the number of participating nodes. This algorithm can be used to compute a Global Measure of Congestion which can help improve existing congestion control schemes.


Back to Network Management Laboratory

Adarsh Sethi's Homepage