Section: Software

SimGrid

Participants : Pierre-Nicolas Clauss, Fekari El Medhi, Martin Quinson, Lucas Nussbaum, Cristian Rosa.

The SimG rid framework aims at being a scientific instrument to the evaluation of algorithmic solutions for large-scale distributed experiments.

The SimG rid tool is the result of a collaboration with Henri Casanova (Univ. of Hawaii, Manoa) and Arnaud Legrand (MESCAL team, INRIA Grenoble-Rhône-Alpes, France). Simulation is a common answer to the grid specific challenges such as scale and heterogeneity. SimG rid is one of the major simulators in the Grid community.

The framework relies on a simulation kernel using a blend of analytical models and coarse-grain discrete event simulation. It proves several orders of magnitude faster than usual packet-level simulators used in the networking community (such as ns2 or GTNetS) while providing an acceptable level of accuracy [51] .

SimG rid provides several user interfaces depending on the user goal.

MSG

helps the study of distributed heuristics. This is the historical interface of SimG rid, and remains the most used one.

SimDag

eases the study of scheduling heuristics for DAGs of (parallel) tasks, which helps the work on parallel task scheduling.

SMPI

allows the emulation of MPI program on top of a simulated environment. This work is still in progress.

GRAS

(Grid Reality And Simulation) eases the development of Grid services and infrastructures [8] .

GRAS provides a C ANSI interface to build distributed services and infrastructures for the Grid. Two implementations of this API are provided: the first one (called Grid R&D Kit) lets the developers experiment, test and debug their work within the SimGrid simulator. The other implementation (called Grid Runtime Environment) allows the resulting programs to run efficiently on real systems.

The simulator thus greatly eases the research and development of Grid services (such as for example monitoring infrastructure or distributed storage systems). In addition, the Grid Runtime Environment is ported to Linux, Windows, Solaris, Mac OS X, AIX and IRIX operating systems, and to 11 hardware architectures. Services built on top of this achieve better communication performance than heterogeneous implementations of the MPI protocol.

SimG rid is freely downloadable SimGrid and its user base is rapidly growing. It grounded the experimental section of more than fifty scientific publications (two third of them from users not being part of the core team).