Exploiting Zero-Entropy Data for Efficient Deduplication