Big data starts when one machine is not enough.
HDFS cuts every file into fixed-size blocks and copies each block to several DataNodes. MapReduce, Pig, Hive and Spark all read data in these blocks. Pick a file size, a block size and a replication factor to see the blocks and the disk space they use.
128
128
128
128
30
542 MB = 4 × 128 MB + 30 MB → 5 blocks × RF 3 = 15 replicas, 1626 MB of raw disk