Versions are stored with timestamps

Versions are ignored

Only latest version is stored

All versions are overwritten

Full table scan

Random row access using row key

Sorting entire dataset

Join operations

Table → Row → Column Family → Column

Column → Row → Table → Column Family

Row → Table → Column → Column Family

Table → Column Family → Row → Column

ZooKeeper

NameNode

DataNode

HMaster

(value, index)

(value, count)

(count, value)

(value, list)

Increases computation

Increases memory usage

Reduces unnecessary data transfer

Slows down processing

Sort entire dataset globally

Use only combiner

Store values in mapper

Use reducer to compare values

Counting words

Finding maximum value

Sorting records within each key group

Filtering data

Data is incomplete at mapper level

Mapper has no memory

Reducer is faster

Mapper cannot process numbers

(salary, dept)

(dept, salary)

(dept, (salary, 1))

(1, dept)