Grid platforms federate large numbers of resources across several organizations. While their promises are great, these platforms have proven challenging to use because of inherent heterogeneity and dynamic characteristics. Therefore, grid application development is possible only if robust distributed services infrastructures, e.g. for resource and data discovery, resource monitoring or application deployment, are available. These infrastructures, which are large-scaled distributed loosely-coupled applications, are very difficult to design, develop and tune.This paper presents the Grid Reality And Simulation (GRAS) framework that allows grid developers to first implement and experiment with such an infrastructure in simulation, benefiting from a controlled and fast environment. The infrastructure can then be deployed in situ without code modification. We first detail the design goals and the implementation of GRAS, and contrast them to the state of the art. We then present a case study to highlight the fundamentals of GRAS and illustrate its ease-of-use. In addition, we quantify the complexity of a code example using either GRAS or several other communication solutions. We also conduct tests over LAN and WAN networks to assess the performance. We find that the code using GRAS is simpler and shorter than any other solution while achieving better performance than most of the other solutions.