Integrative Tools for Hybrid Modeling, Simulation and Control of Data Flows
by
Simulation of data flows varies in flavors according to the size of the problem and the granularity of details required for the results. As the size of the problem grows, fine grained discrete event simulation (e.g. packetlevel) becomes unfeasible due to prohibitive computational demands. In place, socalled fluid flow approximations of said flows are resorted to. This shifts the modeling and simulation strategies towards the continuous systems domain implying solving sets of differential equations, departing completely from the discrete event domain, thus requiring different skills, methods and tools. Moreover, once conclusions are obtained from fluid simulations, these are supposed to be validated against a more realistic (discrete) representation of the system, because important stochastic features are lost when using fluid equivalents (e.g. saturations, packet dropping, retransmissions). Thus, we go back to the original problem (too costly or unfeasible fine grained simulations). The usual practice at this point is to jump into downscaled testing environments that try to imitate the real production environment. Yet, too often such an imitation is again too costly or even impossible.
Today I will present novel strategies and tools to bridge the gap between the fluid flow simulation stage and the real life validation stage. I will describe recently developed modeling and simulation techniques and tools that permit simulating "simultaneously" a dual representation of data flows, i.e., continuous and discrete, inheriting the best of both worlds: the performance gain of solving the system with differential equations, and the fine grained stochastic information provided by explicit discrete "probe" flows. This is achieved under a unified tool (PowerDEVS), which is in turn developed around a unified mathematical framework (DEVS, Discrete Event Systems Specification) while implementing the paradigmshifting Quantized State Systems (QSS) family of hybrid numerical solvers.
About the speaker
Rodrigo Castro received his Electronic Engineer (MAScEE) (2004) and PhD (2010) degrees from the National University of Rosario, Argentina. Since 2007 he is a lecturer at the Computer Science Department, School of Exact and Natural Sciences, University of Buenos Aires, Argentina, with a focus in Discrete Event Simulation and Theory of Communications. In 2011, Rodrigo was appointed an Assistant Researcher at the National Scientific and Technical Research Council (CONICET), Argentina.
Since 2000 Rodrigo has worked for and led several projects for the industry (e.g. Siemens, Cisco, HewlettPackard) in topics related to communication networks, performance testing, and software develoment. Since 2012 he is a postdoc visiting researcher at ETH Zurich, Switzerland and in 2013 he was awarded the DEVS Best Ph.D. Dissertation Award by the Society for Modeling and Simulation International (SCS).
Organised by: Giovanna Lehmann Miotto/PH Department
 Name
 ComputingSeminar_04sep
 Description
 Simulation of data flows varies in flavors according to the size of the problem and the granularity of details required for the results. As the size of the problem grows, fine grained discrete event simulation (e.g. packetlevel) becomes unfeasible due to prohibitive computational demands. In place, socalled fluid flow approximations of said flows are resorted to. This shifts the modeling and simulation strategies towards the continuous systems domain implying solving sets of differential equations, departing completely from the discrete event domain, thus requiring different skills, methods and tools. Moreover, once conclusions are obtained from fluid simulations, these are supposed to be validated against a more realistic (discrete) representation of the system, because important stochastic features are lost when using fluid equivalents (e.g. saturations, packet dropping, retransmissions). Thus, we go back to the original problem (too costly or unfeasible fine grained simulations). The usual practice at this point is to jump into downscaled testing environments that try to imitate the real production environment. Yet, too often such an imitation is again too costly or even impossible. Today I will present novel strategies and tools to bridge the gap between the fluid flow simulation stage and the real life validation stage. I will describe recently developed modeling and simulation techniques and tools that permit simulating ";simultaneously"; a dual representation of data flows, i.e., continuous and discrete, inheriting the best of both worlds: the performance gain of solving the system with differential equations, and the fine grained stochastic information provided by explicit discrete ";probe"; flows. This is achieved under a unified tool (PowerDEVS), which is in turn developed around a unified mathematical framework (DEVS, Discrete Event Systems Specification) while implementing the paradigmshifting Quantized State Systems (QSS) family of hybrid numerical solvers. About the speaker Rodrigo Castro received his Electronic Engineer (MAScEE) (2004) and PhD (2010) degrees from the National University of Rosario, Argentina. Since 2007 he is a lecturer at the Computer Science Department, School of Exact and Natural Sciences, University of Buenos Aires, Argentina, with a focus in Discrete Event Simulation and Theory of Communications. In 2011, Rodrigo was appointed an Assistant Researcher at the National Scientific and Technical Research Council (CONICET), Argentina. Since 2000 Rodrigo has worked for and led several projects for the industry (e.g. Siemens, Cisco, HewlettPackard) in topics related to communication networks, performance testing, and software develoment. Since 2012 he is a postdoc visiting researcher at ETH Zurich, Switzerland and in 2013 he was awarded the DEVS Best Ph.D. Dissertation Award by the Society for Modeling and Simulation International (SCS).
 Extension
 109266392
 Owner
 Miguel Marquina
 Autojoin URL

 Useful links
 Phone numbers
Share this page
Direct link
https://indico.cern.ch/e/266392Social networks
Calendaring