Section: New Results

Large-scale Simulation Platform: Techniques and methodologies

Participants : O. Dalle, E. Mancini.

In the domain of simulation techniques and methodologies, this year, we conducted research in the three following areas:

Simulation in the Cloud

In recent years, numerous applications have been deployed into mobile devices. However, until now, there have been no attempts to run simulations on handheld devices. In the context of the DISSIMINET Associate Team, we work in collaboration with our partners at the Carleton University to investigate different architectures for running and managing simulations on handheld devices, and putting the simulation services in the Cloud[24] . We propose a hybrid simulation and visualization approach, where a dedicated mobile application is running on the client side and the RISE simulation server is hosted in the Cloud.

Simulation Methodology

In the context of the ANR INFRA SONGS project, we are involved (as coordinators) in a Work-Package called “Open Science” whose aim is to investigate and contribute means to ensure the long term visibility and reproductibility of simulation results obtained using the SimGrid simulation platform. Our preliminary work in this direction consisted in identifying the issues, trends and potential solutions to ensure the long-term reproducibility of simulations[16] .

Peer-to-peer Simulation

In order to evaluate the performance and estimate the resource usage of peer-to-peer backup systems, it is important to analyze the time they spend in storing, retrieving and keeping the redundancy of the stored files. The analysis of such systems is difficult due to the random behavior of the peers and the variations of network conditions. In the context of the ANR USS-SIMGRID and INFRA-SONGS projects, we investigated means for reproducing such varying conditions in a controlled way. We worked on the design of a general simulation meta-model for peer-to-peer backup systems and a tool-chain, based on SimGrid, to help in their analysis[20] . We validated the meta-model and tool-chain through the analysis of a common scenario, and verified that they can be used, for example, for retrieving the relations between the storage size, the saved data fragment sizes and the induced network workload. We also started to investigate a new simulation technique for very-large scale distributed simulation of peer-to-peer systems based on the decomposition of a simulation into many micro-simulation steps[31] in order to optimize the overlap between communications and computations.