The run-time environment is a key component of a parallel programming and execution environment. It features services to the processes of parallel applications. The most fundamental of these services are allowing processes to communicate with one another and forwarding IOs and signals. In this thesis I present results concerning run-time environments for large-scale systems and the services provided to large-scale applications. The first part studies the run-time environment's performance for large-scale systems and proposes an approach centered on topologies to support applications efficiently. The second part studies an additional feature provided by the run-time environment on systems made of a large number of components: fault tolerance. The third and the fourth part concern a specific kind of large-scale systems: computational grids. These systems present particularities that are specific to the usage of remote, heterogeneous resources. Problematic related to connectivity throughout computational grids are presented in the third part of this thesis, as well as an extension of the run-time environment designed to face them. The fourth part presents a method to design and program applications for computational grids and examples of typical applications that achieve a good performance.
