Why should I differentiate between Mr and Spark's shuffle

The shuffle phase of Mr is illustrated by a picture:

The output of the map phase is placed in the buffer, and a smaller buffer maintains the index of the key-value pair + partition number in the buffer. When the buffer is nearly full, its index is sorted and then spill to disk. When all the data is spill to disk, the fragmented files are merged, and the process of sorting and merging also occurs to reduce the amount of data transferred to the reducer. Reducer pull the final result from the mapper via an HTTP connection, and note that the desired portion is pulled along the partition. For a reducer because the data may come from multiple upstream, you still have to continue in order, and finally make a reduce operation.

Spark's shuffle will be written later.

Why should I differentiate between Mr and Spark's shuffle