mongodump (logical backup): BSON export; portable across MongoDB versions; supports selective restore; slow on large datasets; not ideal for sharded clusters. Use for: dev/staging, small databases (<50GB), collection-level exports, cross-version migrations. Filesystem snapshot (EBS/LVM): Block-level copy; instant capture; works at TB scale; platform-specific; requires journal on same volume for consistency. Use for: production >100GB self-managed, when RTO is critical (fast restore). Always snapshot a secondary, not the primary. Atlas Backup: Fully managed; continuous cloud backup supports PITR to any second; zero operational overhead; supports cluster-level and collection-level restore. Use for: Atlas clusters, compliance requirements (HIPAA/SOC2), when you need PITR without managing infrastructure. Rule of thumb: Atlas → Atlas Backup; self-managed >100GB → snapshots + mongodump for portability; small self-managed → mongodump.