Single parallel systems are surely the less deployed in actual computing grids. They are usually composed of a frontal node where clients log in, and from which they can log on numerous nodes and execute their parallel jobs, without any kind of reservation (time and space). Some problems occur with such a use of parallel resources: multiple parallel tasks can share a single processor, hence delaying the execution of all applications using it; during the deployment, the application must at least check the connectivity of the resources; if performance is wanted, some monitoring has to be performed by the application.