M2CMeasuring, Modeling and Cost Allocation for Quality of Service |
|
|
This project focuses on the metering process needed to measure actual resource usage, to allocate costs to customers and to estimate future growth in order to be able to meet Quality of Service (perception) demands outlined in Service Level Agreements. In particular, the following two questions are investigated:
How to measure resource usage for adequate (intelligent) overprovisioning? How frequently should these measurements be done?
How can the measurements be translated into a generic and versatile traffic model? How can the resulting models be used to develop robust guidelines for overprovisioning and cost allocation?
The following are the M2C project deliverables for 2003. When ready, links to the results will be added.
D1.1 Scientific paper submitted to an international conference on the topic of Traffic Measurements for Link Dimensioning. In this paper we will quantify the differences in measured link load between fine (10 msec) and coarse-grained (5 min) measurements. The paper will show that traditional link dimensioning techniques will frequently lead to underdimensioning of links. For our analysis we will use the measurement data that was obtained last year on the campusnet and on the network of a small provider. [Published in proceedings of IFIP/IEEE DSOM2003, also: document]
D1.2 Scientific paper submitted to another international conference on a topic like, for example, the "Identification of Mice and Elephants" or the "Relation between Quality of Service and Quality of Experience". [Status: submitted]
D1.3 Visualisatie tools. We want to create some generic (web-based) visualisation tools to quickly analyse repositories of measurement data. The need to have such tools comes from previous discussions in which we tried to understand Internet traffic flows. We experienced that in these discussions always many new questions were raised; to answer these questions we often had to develop ad-hoc visualisation tools. Creating such ad-hoc tools took time; we therefore expect that the generic tools that will be developed as part of this activity will be of great help in future discussions. [available at this website; documentation]
D1.4 Analysis of unknown traffic. The measurements performed on the campusnet showed that more than one third of all campus traffic belongs to "unknown" application ports. Measurements performed (by others) on the Abeline backbone showed similar results; compared to one or two years back this percentage has increased dramatically. The purpose of this activity is to understand the nature of this unknown traffic by analysing earlier traffic flows between the same source and destination addresses. It should be noted that this investigation must be performed without looking into the user data. [document]
D1.5 Surfnet metingen. Measurements will be performed on the access lines between SURFnet and various member institutions. The result of these measurements will be a repository of measurement data that will be analysed in 2004 as part of the QuBA project. [repository available here, and an accompanying document]
D1.6 Measurement tools. Tools developed for the research presented in the other deliverables will be made available online soon. [tools available here, documention]
If possible, a white paper on the state of the art in measurements will be produced as part of the M2C project
D2.1 Overview of existing traffic models. This overview will be published as a whitepaper. [document]
D2.2 Scientific paper submitted to an international conference. In this paper we will derive a formula to predict the required link capacity to guarantee QoS on short time-scales if only long time-scale measurement data is known. To verify the formula, its outcome will be compared to the results of some measurements performed on the network of a big ADSL provider. [Status: submitted]