Merge sort uses a "Divide and Conquer" approach to speed up sorting of large data sets.
Merge Sort Algorithm
Divide the items into two sub-sets.
Sort each sub-set separately.
Merge the two sorted sub-sets back into the array while preserving order.
Merge sort may seem overly complex and is in fact slow for small data sets. However as size of the data set grows, it does not slow down as much as the selection and insertion sorts. Completion time grows in (slightly more than) linear proportion to the size of the data set i.e. Order of Growth is $\mathcal{O}_{nlogn}$.
As sorting is done in parts, only small sub-sets of the data are needed in main memory at any point during the procedure. This makes it suitable for sorting of large data sets on disk.
Read more at Sedgewick, Robert and Wayne, Kevin."Mergesort"Algorithms, 4th Edition.