- use persistent PstreamBuffers between iterations, restrict size
  information exchange to the processor neighbours (which is what the
  algorithm is handling there anyhow).