Mapreduce is used for large datasets that need to be processed. If one machine can't process all the data, you'll have to split the workload across many machines, and handle haradware failure, network partitions, and software failure.
Mapreduce only requires you to define two steps: map and
reduce. Map applies some computation to each element in the
set, and reduce aggregates the set into one value.
Every value must be a key-value pair. All keys with the
same key are given to the same machine, and sorted by the
machine. Then the library reduces according to logic you gave it, and
outputs one result per key, or for the whole dataset.
Mapreduce handles machines failing and network partitions for you, which allows you to focus abstractly on the computation, rather than how it's done.
Prev: [publish-subscribe-pattern](publish-subscribe-pattern.html) Next: [security-and-https](security-and-https.html)