MapReduce-MPI WWW Site - MapReduce-MPI Documentation

MapReduce gather() method

uint64_t MapReduce::gather(int nprocs) 

This calls the gather() method of a MapReduce object, which collects the key/value pairs of a KeyValue object spread across all processors to form a new KeyValue object on a subset (nprocs) of processors. Nprocs can be 1 or any number smaller than P, the total number of processors. The gathering is done to the lowest ID processors, from 0 to nprocs-1. Processors with ID >= nprocs end up with an empty KeyValue object containing no key/value pairs. The method returns the total number of key/value pairs in the new KeyValue object, which will be the same as in the original KeyValue object.

This method can be used to collect the results of a reduce() to a single processor for output. See the collapse() and scrunch() methods for related ways to collect key/value pairs for output. A gather() may also be useful before a reduce() if the number of unique key/value pairs is small enough that you wish to perform the reduce tasks on fewer processors.

This method requires parallel point-to-point communication as processors send their key/value pairs to other processors.


Related methods: scrunch(), broadcast()