Big Data Analytics

Hadoop • 83

• Joins of multiple data sets are convoluted and sluggish due to the lack

of indices in many (but not all) cases.

• Cluster management can be dicult—in the cluster, operations like

debugging, distributing soware, and collecting logs become more

dicult than with the old-fashioned central processing approach.

• It still requires a single master, which requires attention and can

limit scaling.

• Managing job ow is not trivial when intermediate data must be

retained, particularly if a detailed audit trail is needed.

• e optimal conguration of network nodes may be nebulous; for

example, it may not be clear how to congure the mappers, ...