The scheduling of a large number of tasks with various cooperation modes poses interesting theoretical problems in massively parallel systems. Such systems are fully distributed ones, preventing coherent observability. In such systems no absolute time-space reference exists. The theoretically optimal set of local references is given as the limiting factor for the incoherent observability. The failures and/or modifications inherent in the massively parallel systems may change their configurations dynamically. Thus, no optimal algorithms can be developed for the effective utilization of the resources available in such a system. The influence of the granularity level of scheduling and of the cooperation modes of the processes are investigated in order to identify areas for further study.