What are some strategies to have pig and java mapreduce jobs exchange data? E.g. we find a particular pig script in a chain is too slow and we could optimize with a custom mapreduce job we'd want pig to write the data out in a format that mapreduce could access and vice versa.
