Section: New Results

Application Domains

In the context of the FP7 STREP PLAY and French SocEDA ANR research projects we have developed a middleware dubbed EventCloud (Section  5.5 ). This last aims to store and retrieve Resource Description Framework (RDF) data but also to relay them to interested parties through a publish/subscribe layer that allows the formulation of content-based subscriptions. Content-based subscriptions are automatically deduced from more complex rules deployed onto a Complex Event Processing engine, the aim of these CEP rules being to trigger new (complex) events after detecting interesting situations [24] . The EventCloud architecture relies on a CAN structured P2P overlay network we initially designed and implemented for the former SOA4ALL FP7-IP project [44] .

This year we continued to improve the performances of the EventCloud middleware and its usability as a standalone component but also as a component integrated within the previous projects' platform. Concretely, we proposed a new publish/subscribe matching algorithm for RDF events made of several related RDF triples, that was thoroughly presented in [31] and [22] . To further improve performance, we pursue some efforts to finalize the usage of the newest multi-active object library (cf. Section 6.1.1 ). Also, to handle more efficiently multicast messaging, we replaced our initial and naive solution with the optimal one presented in Section  6.2.1 . Finally, we proposed a solution for managing multiple EventCloud instances on various cloud platforms, especially for the integration of our middleware in the PLAY and SocEDA platforms (whose latest assessment can be found in [32] ). Details about EventCloud management are provided in [29] .

Since RDF resources have the property to be poorly balanced, we are also investigating new algorithms that decrease load imbalance for events and data.

Distributed Network Simulation

NetStep[16] , is a prototype we developped for the distributed simulation of very large scale network simulations, such as the simulation of peer-to-peer applications. We use simulation micro-steps as a means for optimizing the overlap of communications and computations, without changing the original event-driven model. As a consequence, NetStep allows for the reuse of unmodified existing sequential simulators for building large-scale distributed simulations: the overall simulation is divided both in time and space, into a large number of simulation micro-steps, each of which being executed by a legacy sequential simulator. By choosing the time-step smaller than the minimal look-ahead due to communications, we avoid the need for synchronization between logical processes (LPs) during the simulation. Instead, the simulated communications become inputs and outputs of the simulation micro-steps, and are routed in parallel between LPs by a NetStep dedicated entity. Our prototype is based on the SimGrid sequential simulator.

Discrete Time Representation

The representation of time in simulations is a long standing issue, for which many solutions and formalisms have been proposed. However, once the formalism is chosen, the implementation of the time representation is still a non trivial problem: Integer values have a limited range and require the selection of a minimal fixed step that does not support well the multi-scale models; Floating Points numbers have numerous limitations and hidden effects such as rounding due to quantization; those issues result in inaccuracies or even timing errors. In collaboration with our partner in the DISSIMINET Associated Team, we have started a new research on this topic. This research will be released in the form of a new Discrete Event Simulation engine library for the DEVS formalism, designed to fully exploit the 2011 C++ standard; it is candidate for inclusion in the BoostC++ Libraries.