Vero Estrada will give a practice talk for her WIP paper at FAST'16.
Summary: Traditional RAID solutions were designed before the cloud. "A case for Redundant Arrays of Inexpensive Disk (RAID)" presented five settings for arrangements of independent disks to increase the performance and reliability of a system. The word “inexpensive” implies that a RAID system could be built with commodity elements. Eventually, hardware controllers became highly sophisticated and, accordingly, expensive. Often, the word inexpensive is replaced by independent. Although there is still future for cheap spinning disks in the market, the market changed substantially in the last decades. Exabyte scale redefines storage needs. How can we rethink RAID for cloud storage or HPC systems? This work explores the creation of a "cost free large stripe'' using helical entanglement codes (HEC), and well-established criteria on parity declustering from Holland and Gibson. Researchers have explored RAIDed storage at a petabyte scale. However, in general, the community remains sceptical about the future of disk arrays given their scalability limitation and the prices of proprietary hardware. Hence, this study is complemented with preliminary steps on how HEC and Ceph could be integrated as an alternative storage solution.