The Open Science Grid relies on several network facing services to deliver resources to its users. The major services are the Compute Elements, Storage Elements, Workload Management Systems and Information Systems. All of these services are exposed to traffic coming from all over the world in an unmanaged way, so it is very important to know how they behave at different levels of load. In this paper we present the methodology and the results of scalability and reliability tests performed by OSG on some of the above services. The major services being tested are the Condor batch system, the GT2, GRAM5 and CREAM CEs, and the BeStMan SRM SE.