GDedup: Distributed File System Level Deduplication for Genomic Big Data

摘要

During the last years, the cost of sequencing has dropped, and the amount of generated genomic sequence data has skyrocketed. As a consequence, genomic sequence data have become more expensive to store than to generate. The storage needs for genomic sequence data are also following this trend. In order to solve these new storage needs, different compression algorithms have been used. Nevertheless, typical compression ratios for genomic data range between 3 and 10. In this paper, we propose the use of GDedup, a deduplication storage system for genomics data, in order to improve data storage capacity and efficiency in distributed file systems without compromising I/O performance. GDedup can be developed by modifying existing storage system environments such as the Hadoop Distributed File System. By taking advantage of deduplication technology, we can better manage the underlying redundancy in genomic sequence data and reduce the space needed to store these files in the file systems, thus allowing for more capacity per volume. We present a study on the relation between the amount of different types of mutations in genomic data such as point mutations, substitutions, inversions, and the effect of such in the deduplication ratio for a data set of vertebrate genomes in FASTA format. The experimental results show that the deduplication ratio values are superior to the actual compression ratio values for both (file read-decompress or write-compress) I/O patterns, highlighting the potential for this technology to be effectively adapted to improve storage management of genomics data.