grant agreement No 825184)Grant Agreement • July 6th, 2022
Contract Type FiledJuly 6th, 2022Map phase Many algorithms require synchronization at various stages. In MapReduce [38], this happens between the map and reduce phases, and it is known as shuffle. Shuffling ensures that the reduce phase starts when all the appropriate data was output in the preceding map phase. Shuffling the map output is a costly operation in MapReduce, even if the reduce phase is short. For that reason, when data is small and the reduction operation simple, it is better to skip the reduce phase and instead aggregate the map output directly in the storage layer [40]. CRUCIAL allows to easily implement this approach.