[MS 리서치] Admission Control and Rate Adaptation

Admission Control and Rate Adaptation

Overview

Demo_Scenarios

Papers

Talks

Overview

Communication networks carry a mixture of elastic traffic, such as Web downloads, and inelastic real-time traffic. For the latter, quality degrades appreciably below a specified throughput and so bandwidth sharing is inappropriate. It is preferable instead to accept such traffic only when it is possible to guarantee that it receives a specified bandwidth for a specified duration. Rather than do this in a centralized manner, we study a decentralized solution, where a call or flow probes the network and decides whether to admit itself on the basis of these measurements. More generally, a flow may choose among a small set of rates on the basis of network measurements.

Specific scenarios were are considering are

• A home network, where a request is made for an interactive videoconference or  retrieving stored  multimedia (eg from a jukebox) and we wish to know if is possible to grant the request with neither a  central controller nor a reservation protocol.  The end-system have to quickly make a yes-no decision, using existing infrastructure.     We are also considering future scenarios where rudimentary priorities may come from the use of 802.1p/q switches.
• The WAN / Corporate Intranet, where we wish to run a delay sensitive application  (eg VOIP, video conference) and want to have in essence a distributed light-weight signalling solution.   For some of this work we assume that routers/ firewalls can signal incipient congestion by setting the ECN flag.
• Creation of a  "lower than best effort" service, e.g. for background transfers, where we need have minimal effect on foreground flows.

As a by-product of this work, we are looking at what statistical techniques can say about inferring properties of a network from measurements (eg RTT, link capacity, spare capacity, characterising cross-traffic etc).

People within Microsoft Research Cambridge

• Ayalvadi Ganesh
• Dinan Gunawardena
• Peter Key
• Laurent Massoulie

Demo Scenarios

 

The image above (presented at WinHEC 2003) shows the typical scenario in the home where admission control probing can help ensure a good user experience for networked multimedia. This is work done in collaboration with the Windows Home Networking team during 2003.

We have extended our work to investigate adding congestion awareness to networked applications such as those used for conferencing and media streaming:

This is work done in collaboration with the Windows Microsoft Live Communications Group and in consultation with the Windows Digital Media Division team during 2002.

Papers

Network Aware Applications: A Background Transfer Service, Peter Key, Laurent Massouli and Bing Wang, Forty-First Allerton Conference on Communication, Control, and Computing, Allerton, October 2003 Network characteristics: modelling, measurements and admission control, Dinan Gunawardena, Peter Key and Laurent Massouli, IWQoS 2003, Monterey, June 2003. A network flow model for mixtures of file transfers and streaming traffic, Peter Key, Laurent Massouli, Alan Bain and Frank Kelly. Microsoft Research Technical Report, MSR-TR-2003-37. ITC 18, Berlin, 2003. Probing strategies for distributed admission control in large and small scale systems, Peter Key and Laurent Massouli. PDF(procedings version) , IEEE Infocom,San Francisco 2003). Service-Differentiation for Delay-Sensitive Applications: An Optimisation-Based Approach, with Laurent Massouli and Jonathan Shapiro, Proceedings of IFIP Performance 2002 conference, Rome September 2002.  Final version in Performance Evaluation, 2002, Vol 49, pp 471-489.  see also short version.  Modelling the Performance of In-Call Probing for Multi-Level Adaptive Applications, with Alan Bain (Stats Lab, Cambridge), Microsoft Research Technical Report, MSR-TR-2002-06, Jan 2002, PDF , Postscript Feedback and bandwidth sharing in networks, A. J Ganesh, P. B.  Key and L. Massouli. Proc. 39th Annual Allerton Conference on Communication, Control and Computing, 2001. Resource Allocation with Persistent and Transient Flows S. Deb, A. J. Ganesh and P. B. Key, MSR Technical Report, MSR-TR-2001-114. Modelling the Performance of Distributed Admission Control for Adaptive Applications Alan Bain and P. B. Key, MAMA Workshop, June 2001. Distributed Admission Control, F. P. Kelly, P. B. Key and S. Zachary. J. Sel. Areas Comm., 2000.

Talks

Distributed Admission Control and Congestion Pricing, P. B. Key, Workshop on Pricing and Quality of Service, Paris, 1999. PowerPoint.