As the implementation is SPMD, calling this function will return to every processor in the computation, the value at the required node. This requires a global syncronisation and then a broadcast of the data value from the processor that owns it to all the other processors. This will naturally impair the parallel performance of the system. Avoiding the use of the Read function is then important in any program that aims to maximise performance. However, many algorithms require this sort of global information between the various phases of computation.