The guys at h2o.ai have been parallelizing machine learning algos for their open source project. I guess their major insight is to ship around different steps

The guys at h2o.ai have been parallelizing machine learning algos for their open source project. I guess their major insight is to ship around different steps of the algo to different nodes rather than ship around data. And they avoid the sort step in hadoop too.