Building a Data Classification System with Mahout - Data Just Right: Introduction to Large-Scale Data and Analytics - page 141

Mahout. However, MapReduce-based frameworks can also create a performance
bottleneck due to heavy disk access. The UC Berkeley AMPlab, which produces the
in-memory Spark distributed-computing framework, also sponsors a project known
as MLbase. The components of the MLbase project aim to make distributed machine
learning as accessible as possible and could be a viable and performant alternative to
Mahout for some use cases.