Quality of Service for In-Home Digital Networks

(PROGRESS project EES.5653)

 

Introduction

The project is focussed on the problems surrounding resource sharing in networks of consumer devices, more precisely, of connected real-time embedded systems. We regard these networks as contained domains, meaning that their use is under complete control of its owner. The considered systems are resource constrained since the difference between worst-case and average-case performance is too large to be able to afford overdimensioning. In addition, resources cannot always be guaranteed as is the case, for example, in wireless networks. The shared resources in our project concern the network, the processing power in terminals and network devices and the memory in these. The domain we study is the domain of networked video, since this represents currently the largest challenge. Because this system must cope with overload situations, it must be possible to adjust the quality in order to avoid failure. The project addresses this graceful quality adaptation by

  1. investigating techniques to change (application) quality;
  2. to optimize quality given the circumstances by investigating control architectures that support this as well as looking at feedback from user-perception;
  3. developing methods to analyse and predict resource use such as to support decision procedures.

 

Three PhD students are working on the project, two are funded by STW and 1 is funded by TU/e.

·        Dmitri Jarnikov works on scalable video (TU/e, from October 1, 2003)

·        Sergei Kozlov addresses QoS aware transport protocols (STW, from January 1, 2003)

·        Alina Albu works on prediction of resource requirements of streaming applications (STW, from July 1, 2003)

 

The project is supervised by dr. J.J. Lukkien and dr. P.D.V.van der Stok; promotor and project responsible is prof.dr.E.H.L.Aarts.

 

 

Basic philosophy

First, taking as a start the ITU-T Recommendation E.800 – Geneva 1994 as the definition of Quality of Service:

Quality of Service: “... is the collective effect of service performances, which determine the degree of satisfaction for a user of a service”

we observe that QoS is then dependent on all system parts. It plays at different timescales and at different layers of the system. Clearly, these different timescales lead to different problems and solutions. For example, rapid bandwidth changes in the millisecond range cannot be solved by system-wide reallocation in the second range. Conversely, structural and long-lasting load changes should not just rely on algorithms and protocols that solve quick bandwidth variations.

 

Second, we think that convergence towards optimal use is in some cases better than end-to-end guarantees. Internally, the system may use strict resource assignment and enforcement. However, unforeseen external circumstances as well as fluctuations in actual resource requirements must be dealt with such that the resulting behavior converges again to optimal use. This calls for adaptibility which, because of the first point, plays at several layers.

 

We can demonstrate these issues with the simple example of a video source (encoding component) transmitting to an access point (which does transcoding) and connects through some channel to a terminal that will do decoding. We can have the following changes that the system needs to respond to:

•         Variations in terminal load

–        competitive applications, additional streams to encode/decode.

–        variations in computational effort in codec dependent on the video.

•         Variations in available bandwidth, e.g. high, bursty loss in wireless networks

The effect of these variations are that the application (or a part of it) does not get the resources it needs. Sometimes, system software does not deal with this very well. For example, some transport protocols are very sensitive to bursty loss, hence, more suitable protocols should be developed. For the rest, the response to this situation is essentially found in making the application adaptive. This can be done at several places in the chain, viz. at the transport layer, at the encoding/decoding terminals and perhaps in a combination of the two. Tradeoffs could be made between transmitting less information at the expense of more calculation etc. In order to deal with structural load changes a control structure must be able to estimate the new situation and perform admission tests. To that end, good predictive models of the application are needed.

 

Current work and plans

Our current work addresses the following questions:

·        Scalable Video. We look at the issue of changing the quality of the transferred video based on quality variations of the network and variations in processing capacity. Particularly, we address layered video using SNR scalabilty. We develop a feedback-control loop in order to optimize the behavior.

·        Adaptative transport protocols. Quick variations in network quality should be tackled by the transport protocol. We look at selective and adaptive retransmission strategies and at smart frame-dropping.

·        Predicting resource requirements. When a set of components is activated on a particular node their resource consumption should be known.

 

Demonstrator

There are regular demonstrations of the achievements in the involved groups, viz., SAN at TU/e and OASIS at Philips Research. There have been cooperations with product divisions in Philips resulting in more robust demonstrators and validation of methods. A demonstration has been given on the IST event of the European Union in the week of November 15, 2004. A demonstration of the work by Sergei is planned on the exposition of CSE in Las Vegas, January 2005.

 

For the project we aim at a combined demonstrator according to the following picture.

By extending to multiple streams we can increase the challenge on the control structure to include admission control.

 

Meetings, contacts and conference visits

·        Weekly meetings on Friday  both with the SAN group (the weekly colloquium series) as well as with staff members for supervision purposes.

·        Usergroup meeting on May 28, 2004.

·        Posters on the Progress workshop of October 2003 (best poster presentation) and October 2004.

·        prof. dr. Gerhard Fohler is a regular visitor advising on real-time issues.

·        dr. Kurt Wallnau from Carnegie Mellon (SEI) was visiting TU/e in October 2004; discussions of Alina with his staff have started.

·        Presence at the EU/IST conference on November 15-17, 2004.

 

Relevant output so far

·        First progress report in October 2003

·        Review on May 28, 2004

1.      Second progress report in May 2004

2.      Agenda and slides from the meeting on 28-5-2004.

3.      Alina’s presentation

4.      Sergei’s presentation

5.      Minutes

·        Review on December 10, 2004

1.      Third progress report in November 2004

2.      Agenda and slides from the meeting on 10-12-2004

 

·        Poster: S. Kozlov, Peter v.d. Stok, Johan Lukkien, TCP-MM – a real-time transport protocol
for multimedia in in-home networks, 4th PROGRESS Symposium on Embedded Systems, October 2003.

·        Poster: M.A. Weffers-Albu, P.v.d.Stok, J.J. Lukkien, "A Characterization of Streaming Apllications Execution", 5th PROGRESS Symposium on Embedded Systems, October 2004

 

·        Draft report: M.A. Weffers-Albu, P.v.d.Stok, J.J. Lukkien, "Quality of Service Overview"

·        Report: H. de Groot (editor), I. Nitescu, I.C. Kang, D. Jarnikov, P. D.V. van der Stok, Robust scalable video over wireless networks (KISS demonstrator). PR-TN-2004/00160 (Philips Restricted)

 

·        Presentations by Alina on QoS papers

1.      Prediction-based policy adaptation for QoS management in wireless networks - 23.10.2003

2.      Cooperative Run-Time Management of Adaptive Applications and Distributed Resources. - 7.11.2003

3.      An Architecture for QoS guarrantees and routing wireless/mobile networks. - 03.12.2003

·        Presentation by Alina: overview of NCS calculation method, at SAN meeting on 12-11-2004

·        Presentation by Sergei: TCP-FCW – transport protocol for real-time transmissions on high-loss networks, at Philips Research on 23-2-2004.

 

·        Paper: M.A. Weffers-Albu, P.v.d.Stok, J.J. Lukkien, "NCS Calculation Method for Streaming Applications", Proceedings of the 5th PROGRESS Symposium on Embedded Systems.

·        Paper: M.A. Weffers-Albu, P.v.d.Stok, J.J. Lukkien, "A Characterization of Streaming Apllications Execution", submitted. 

·        Paper: Sergei Kozlov, Peter van der Stok, Johan Lukkien, Adaptive scheduling of MPEG video frames during real-time wireless video streaming, submitted.

·        Paper: Dmitri Jarnikov, Peter van der Stok, Clemens C. Wust, Predictive Control of Video Quality under Fluctuating Bandwidth Conditions. Published at ICME2004.

·        Paper: Dmitri Jarnikov, Sergei Sawitzki, Peter van der Stok, Wireless transmission of scalable video with optimal layering. Submitted to ICC2005.

·        Paper: Dmitri Jarnikov, Peter van der Stok, Johan Lukkien, Timely wireless streaming based on a scalability scheme using legacy MPEG2 decoders. Submitted to RTAS2005

 

 

Patents

 

ID number

Submission date

Title

693304

2004-Jul-28

Enhancement layer GOP

693454

2004-Jul-20

Scalable video layer identification