In this paper, we study the problem of real-time scheduling of parallel tasks represented by a Directed Acyclic Graph (DAG) on multiprocessor architectures. We focus on Global Earliest Deadline First scheduling of sporadic DAG tasksets with constrained-deadlines on a system of homogeneous processors. Our contributions consist in analyzing DAG tasks by considering their internal structures and providing a tighter bound on the workload and interference analysis. This approach consists in assigning a local offset and deadline for each subtask in the DAG. We derive an improved sufficient schedulability test w.r.t. an existing test proposed in the state of the art. Then we discuss the sustainability of this test.