VAST Data warns that as SSD capacities reach the petabyte scale, traditional RAID configurations in parallel file systems face critical data loss risks. Simulations of a 900 PB system using 122 TB drives show that enclosure-level declustered RAID, currently used by systems like HPE Cray ClusterStor and DDN EXAScaler, results in a 1 in 11 chance of data loss over a decade.
The primary challenge is the rebuild time for high-capacity drives. Because per-drive performance remains static while capacity grows, larger drives take significantly longer to rebuild after a failure. This creates a wider window for secondary failures to occur, which can lead to catastrophic data loss in standard RAID 5 or double-parity setups.
VAST proposes system-wide erasure coding with at least triple parity to mitigate these risks. According to their simulations, implementing global declustering alongside advanced parity schemes can reduce the probability of data loss to 1 in 62 billion. The company also uses a separate, triple-replicated index on fast SSDs for metadata management.
The shift toward high-capacity drives is driven by power efficiency. A 900 PB system built with 15 TB SSDs would require over 63 000 drives and 5.6 MW of power. In contrast, using 122 TB drives reduces the drive count to approximately 8 000 and power consumption to 400 kW, allowing more energy to be allocated to compute resources.
Market trends indicate that 245 TB-class drives are already entering the market, with manufacturers such as Micron, Kioxia, SanDisk, and Solidigm advancing their roadmaps. As drive capacities continue to grow toward the petabyte level, VAST argues that current industry-standard RAID approaches will become insufficient for maintaining data integrity.




