Section: New Results
Experimental Evaluation
Automating ns-3 Experimentation in Multi-Host Scenarios
Participants: Alina Ludmila Quereilhac, Damien Saucez, Thierry Turletti and Walid Dabbous
ns-3 is a flexible simulator whose capabilities go beyond running purely synthetic simulations in a local desktop. Due to its ability to run unmodified Linux applications, to execute in real time mode, and to exchange traffic with live networks, ns-3 can be combined with live hosts to run distributed simulations or to transparently integrate live and simulated networks. Nevertheless, setting up ns-3 multi-host experiment scenarios might require considerable manual work and advanced system administration skills. The NEPI experiment management framework is capable of automating deployment, execution, and result collection of experiment scenarios that combine ns-3 with multiple hosts in various ways, reducing the burden of manual scenario set up. We proved that this approach can be used to seamlessly running parallel simulations on a cluster of hosts, running distributed simulation spanning multiple hosts, and integrating live and simulated networks. This work has been published in [18] and has been awarded as the best paper of the workshop.
DiG: Emulating Data Centers and Cloud Architectures in a Grid Network
Participants: Hardik Soni, Thierry Turletti, Damien Saucez
We are witnessing a considerable amount of research work related to data-center and cloud infrastructures but evaluations are often limited to small scale scenarios as very few researchers have access to a real infrastructure to confront their ideas to reality. We have designed an experiment automation tool, called DiG (Data-centers in the Grid), which explicitly allocates physical resources in grids to emulate data-center and cloud networks. DiG allows one to utilize grid infrastructures to evaluate research ideas pertaining to data-centers and cloud environments at massive scale and with real traffic workload. We have automated the procedure of building target network topologies while respecting effective performance capacity of available physical resources in the grid against the demand of links and hosts in the experiment. We demonstrate a showcase where DiG automatically builds a large data-center topology composed of hundreds of servers executing various Hadoop intensive workloads (see our demo abstract at IEEE NFV/SDN 2015 in [24] ).