Séminaire REGAL


Démo SPLAY - "distributed applications made simple"

Wednesday, October 29, 2008
Speaker(s) : Lorenzo Leonini

Abstract (demo Splay) Evaluating large-scale distributed applications is a highly complex, time-consuming and error-prone task. One of the main difficulties stems from the lack of appropriate tools for quickly prototyping, deploying and evaluating algorithms in real settings. Several dedicated testbeds are available that can be leveraged for better evaluations of these systems: PlanetLab, Everlab, or network emulators such as ModelNet or Emulab. Meanwhile, non dedicated testbeds such as networks of idle workstations usually found in research labs or schools, are difficult to use for distributed systems experiments, as one usually require access rights that are not easily granted by the administrators. All these testbeds are appealing as they allow real or realistic experiments to be conducted, but they are not used as systematically as they should. This calls for an integrated toolbox that allow distributed systems practitioners to straightforwardly develop, test and evaluate distributed systems.
Splay is an integrated system that facilitates the complete chain of distributed systems evaluation, from design and implementation to deployment and experiments control. Algorithms are expressed in a concise, yet very efficient, language based on Lua. Implementations in Splay are highly similar to the pseudo-code usually found in research papers. Splay eases the use of any kind of testbeds, e.g., PlanetLab, ModelNet clusters, or non-dedicated platforms such as networks of workstations. Splay controls the experiment from the resource selection to the replay of churn using traces or synthetic descriptions. Using Splay and PlanetLab, this demonstration highlights a complete evaluation chain of an epidemic protocol. Depending on the audience requirements, specific features of Splay can be presented live as well.
Biography Lorenzo "Leo" Leonini has been a PhD student in Computer Science at the University of Neuchâtel, Switzerland under the guidance of Prof. Pascal Felber since 2005. His research interest lie in large-scale distributed systems design and evaluation. He also has a principal interest in proposing systems, methods and language support that ease the task of creating, deploying and observing distributed applications.