Our paper on ``Federated learning of predictive models from federated Electronic Health Records''...
Admission Control in High Speed Multimedia Networks
Funding Agency: National Science Foundation, Directorate for Computer and Information Science and Engineering, Division of Advanced Networking Infrastructure and Research (ANIR), Networking Research Program.
Award Number: NCR-9706148.
Principal Investigators: Yannis Paschalidis, Boston University.
Quality of Service (QoS) provisioning in multimedia high speed networks is one of the fundamental problems that need to be addressed before real-time applications can be reliably supported. Such applications include, among many others, internet telephony, web access of multimedia information, interactive TV, video-on-demand, and videoconferencing. They have important uses in a wide range of activities from education and health care to entertainment. Their full deployment will clearly have a huge impact on the quality of our lives and the society as a whole.
Technology to accommodate these applications exists (e.g., the Asynchronous Transfer Mode – ATM – protocol, the Internet enhanced with differentiated services capabilities); the challenge is how to manage the network resources (bandwidth) to provide several QoS grades that such a diverse set of applications requires.
To provide QoS the network should prevent congestion which causes packet losses due to buffer overflows and excessive delays. One class of mechanisms that have been proposed is based on worst-case analysis and provides deterministic QoS guarantees, that is, ensuring no packet losses and no large delays with certainty. Although such an approach is useful when no statistical description of the offered traffic is available, it can lead to substantial underutilization of the network resources. To realize statistical multiplexing gains the so called effective bandwidth mechanism has been proposed. Briefly, effective bandwidth is a number between the peak and average rate of a connection such that when connections are allocated their effective bandwidth in an appropriately dimensioned buffer, the buffer overflow probability stays below a given small level (say on the order of 10-7). Real-time applications can tolerate such small frequencies of congestion phenomena.
The effective bandwidth is a single class scheme: all connections are multiplexed into one buffer and, thus, they face the same QoS. The objective of our work is to develop an admission control approach that provides class-specific QoS guarantees. We quantify QoS by the probability of excessive delays and the loss probability. It is desirable to keep them at very small levels (on the order of 10-7). Determining such probabilities for non trivial traffic models is a particularly hard problem, thus, it is natural to focus on asymptotic regimes and determine their exponential decay rate. To this end, large deviations theory is our main analytical tool, since it provides a methodology to obtain asymptotic expressions for the tails of arbitrary distributions.