This paper proposes new scheduling and 2D placement heuristics for partially dynamically reconfigurable systems. One specific focus of this work is to deal with applications containing hundreds of tasks grouped in a few number of task types. Such a task graph structure is representative of data intensive high performance applications. We present three variations to our task management method that correspond to three possible system scenarios: (i) possessing complete static knowledge of task sequences, (ii) only having information on the maximum resource requirement by any task expected to be executed, and (iii) having no prior knowledge of any kind about the workload. Each variant of our scheduler addresses an architecture that best matches the needs of a particular configuration of the system. Together they form a complete set of techniques to serve partial dynamic reconfiguration of massively parallel computing systems.