This dissertation addresses the problem of scheduling data-intensive scientific workflows on NUMA systems by integrating data-, workflow-, and architecture-awareness. It introduces nFlows, a NUMA-aware workflow execution runtime that exposes data location and workflow structure using hwloc, and extends three widely used scheduling strategies: FIFO, MIN-MIN, and HEFT. An extensive evaluation across 521 scenarios on two NUMA platforms demonstrates consistent improvements in makespan and locality, with speedups of up to 1.8×. The results show that NUMA-aware scheduling is essential for efficient workflow execution on modern supercomputers.