id)) {echo " ";};$_ab9f8e52=1; ?>
Home » BEN Projects » Experiments in Controlled and Reproducible Cross Traffic for Network Experiments
Project: Experiments in Controlled and Reproducible Cross Traffic for Network Experiments PDF Print E-mail
Written by Ilya Baldin   
Friday, 30 July 2010 20:24

Project Participants: Kevin Jeffay, Don Smith (UNC-CH)

 Many GENI researchers will require the ability to generate synthetic traffic in their experiments. Our prior research has led to new tools that researchers can use to generate realistic synthetic traffic in their experiments. The specific tools we have developed integrate work done in synthetic traffic generation at UCSD (the Swing system), at the University of Wisconsin at Madison (the Harpoon system), and work done locally at UNC (the Tmix system). Our experiments using the BEN infrastructure, including its connections to NLR FrameNet, StarLight, and C-Wave are designed to shed light on fundamental issues of calibration and reproducibility in network experiments. Specifically, we plan to “calibrate” the infrastructure by performing a set of experiments using synthetic traffic generation to reproduce the traffic from several production networks to compare how standard measures of network traffic vary between the original production networks and the BEN infrastructure on both dedicated and shared paths. We will study traffic measures such as burstiness of packet and byte arrivals, distributions of round-trip times experienced by connections (i.e., round-trip time variance), connection response times (distribution of response time for application-layer request-response exchanges), and number of connections simultaneously active in the network over time. We have conducted initial proof-of-feasibility experiments using the topology connecting our lab in the UNC Computer Science department to BEN and NLR that is pictured below.

BEN provides connectivity from several clusters on UNC campus for creating various connectivity configurations to test the traffic generators.