Application versus network level performance

This work aimed at a better understanding of how changes in network performance (and service parameters) affect the performance seen by applications. One suite of experiments involved passing packet video streams through policers with different combination of parameters, and evaluating (quantitatively and qualitatively) the evolution of application level performance. The quantitative evaluation of video quality was done using the VQM tool developed by The Institute for Telecommunications Science and was carried out over both a local testbed and the wide area QBone testbed. Testing over the QBone infrastructure was done in collaboration with members of the PennNet and Computing departments of the University of Pennsylvania and with researchers at iCAIR and the IBM T.J. Watson Research Center. Additional experiments were conducted for both voice and video application in collaboration with other researchers at UMass and UMinn. over a wide-area testbed connecting all three sites. Exploring how to accurately and easily monitor the quality of video transmissions over packet networks was also investigated in collaboration with John Apostolopoulos from HP Labs.

Another set of experiments carried out with collaborators at AT&T Research focused on the use of aggregate and “non-intrusive” performance measures for estimating the actual throughput experienced by TCP based applications. Non-intrusive refers to measures that are readily available from routine network monitoring, e.g., from routers MIBs, and do not require any flow specific awareness. The work involved the development of models (modification of existing models) that were capable of accurately predicting TCP throughput on the basis of such information, and the evaluation of their accuracy using both simulations and testbed experiments.

Penn Contributors

Publications