The parboost package implements distributed gradient boosting based on 
the \pkg{mboost} package. When should you use parboost instead of mboost?
There are two use cases: 
1. The data takes too long to fit as a whole
2. You want to bag and postprocess your boosting models to get a
    more robust ensemble

parboost is designed to scale up component-wise functional gradient
boosting in a distributed memory environment by splitting the
observations into disjoint subsets. Alternatively, parboost can
generate and use bootstrap samples of the original data. Each cluster
node then fits a boosting model to its subset of the data. These
boosting models are combined in an ensemble, either with equal
weights, or by fitting a (penalized) regression model on the
predictions of the individual models on the complete data. All other
functionality of mboost is left untouched for the moment.