FPKM means fragment per kilobase of exon model per million reads sequenced. Fragment is an insert, for example a 2 paired-end reads account as 1 fragment.
FPKM is a double-normalization. It normalizes by transcript length and also by the total number of fragments sequenced. It is important to normalize also by the total number of fragments sequenced because different RNA-Seq libraries could have different sizes, which affects the amount of fragments mapped to each transcript.