Managing cold storage for research lab's genomics data. Currently 500TB, growing 20TB/month. Debating architecture for next 5 years.
Current Iwe need RAID-60 on-prem, but hitting MTBF concerns with 100+ drives. Considering S3-compatible object storage (MinIO cluster) for better durability.
The requirements are 11-nines durability, occasional full-dataset reads for reanalysis, POSIX mount capability for legacy pipelines. Budget: $50K initial, $5K/month operational.
RAID gives predictable performance but rebuild times terrify me. Object storage handles bit rot better but concerned about egress costs when researchers need full datasets.
Anyone architected similar scale for write-once-read-rarely data? How do you balance cost, durability, and occasional high-bandwidth access needs?