Feasibility

The issues that have not been resolved are as follows:

• If a node's memory is completely used by the particles within that mesh partition, how can PPF stop particles being sent to that node?
  • If particles are stalled within the MPI_Send of the sending nodes then we risk buffer overflows on the sending sides system buffer.
  • If we require programmers to check whether a node's memory is saturated we compromise ease of use and PPF becomes more difficult to integrate into existing code.
  • If we manage to stop sends to that node there are a number of timing issues we must deal with.