Storage

Key Requirements

• ✓High-performance shared storage out-of-the-box via a parallel filesystem (e.g., Weka, DDN, VAST)
• ✓Out-of-the-box managed S3-compatible object storage
• ✓Storage integration with Kubernetes for PVCs / storage class, including ReadWriteMany (RWX) PVCsKubernetes
• ✓Storage mounts correctly out-of-the-box and stays mounted (no random flaking on and off)
• ✓Read and write performance measured and within expectations
• ✓Throughput and latency measurements documented
• ✓Storage scales in both performance and capacity, and the provider can demonstrate or let customers validate that scaling
• ✓Automated backups with configurable retention policy for file systems, object storage, and databases
• ✓Cross-region replication or backup for disaster recovery (object storage CRR, geo-redundant storage, or equivalent)
• ✓Point-in-time recovery for managed databases (transaction log archival with configurable retention)
• ✓Snapshot support for persistent volumes (CSI snapshots for Kubernetes PVs, filesystem-level snapshots)KubernetesSlurm
• ✓Immutable or WORM-capable storage for ransomware protection and compliance (object lock, retention policies)
• ✓Published durability and availability SLAs for all storage tiers (e.g., 11 nines for object storage)
• ✓Backup encryption at rest and in transit with customer-managed key (CMK/BYOK) support
• ✓Centralized backup monitoring, alerting, and restore validation
• ✓Checkpoint storage durability for training workloads (replication factor, cross-AZ/cross-region guarantees)