MAP REDUCE
==========

Introduction
------------

A modest implementation.

The framework is composed in 4 operations, 'inputify', 'map', 'reduce'
and 'ouputify'.

In a first step the 'inputify' operation takes any document to split
it in chunks of 'struct input_split'. Basically a set of 'struct
input_split' is a simple linked-list.

An internal process called 'distribute' will be responsible to equally
share to the map processes, part of the inputs.

The 'map' function is executed in parallel depending the number of
threads requested. The key/value pair results of map should be 'emit'
to the in-memory storage. That one is thread-safe and appends the
values based on the key.

  The in-memory storage concept

             ____ ____ ____ ____ 
  storage = | k1 | k2 | .. | kn |
	     ‾|‾‾ ‾‾‾‾ ‾‾‾‾ ‾‾‾‾
	     _|__ 
            | v1 |
	     ‾|‾‾
     	     _|__ 
            | v2 |
	     ‾|‾‾


The 'reduce' process receives a reference of the storage, computes the
result and returns an user defined data-structure which will be then
passed to the 'outputify' function responsible of managing the result.


Hacking note
------------

Trying to follow:

  https://www.kernel.org/doc/html/v4.10/process/coding-style.html


Known issues
------------

- The in-memory storage is based on a array and take 0(n) to find a
  key
- The emit is globally locked by a mutex
- It's something wrote in few days, the design may be not optimal at
  all :/
- ...