# OCI Service: Autonomous Database Serverless (ADB-S) # Last verified: 2026-04 service: name: "Autonomous AI Database Serverless (ADB-S)" id: adb_serverless category: database color: "#AA643B" features: auto_scaling: true auto_indexing: true auto_tuning: true auto_patching: true transparent_application_continuity: true autonomous_data_guard: true private_endpoints: true byol_eligible: true ha_dr_options: local_ha: "Built-in TAC (Transparent Application Continuity) — ~0 perceived RTO" same_region_adg: "Autonomous Data Guard standby — synchronous, RPO 0" cross_region_adg: "Cross-region ADG — asynchronous only, RPO seconds (switchover RPO 0, manual failover RPO up to 1 min), RTO < 10 min" cross_region_auto_failover: "NOT available — cross-region requires manual switchover/failover only" limits: max_ecpu: 512 max_ocpu_legacy: 128 max_storage_tb: 384 max_auto_scale: "3x base ECPUs" auto_scale_activation: "2-3 minutes" min_ecpu: 2 min_storage_tb: 1 notes: "ECPU is now the default billing unit. 4 ECPUs ≈ 1 OCPU. OCPU billing is legacy." variants: - id: adb_s name: "Autonomous AI Database Serverless (ADB-S)" description: "Shared Exadata infrastructure, fully managed, auto-scaling" workload_types: - "Autonomous AI Transaction Processing (TP) - app-facing OLTP and mixed workloads on local database-managed data" - "Autonomous AI Lakehouse - SQL across database-local plus external/open-format data (Object Storage, Iceberg, federated catalogs)" - "Autonomous JSON Database - document-centric JSON development pattern" - "APEX Application Development - low-code application development" workload_notes: - "Since October 2025 Oracle documentation uses Autonomous AI Database naming for workload types." - "Lakehouse is not a separate database engine; it is a workload specialization with an explicit external-data plane." - "Lakehouse now also exposes TP and TPURGENT predefined service names, so do not assume warehouse-only service semantics." - id: adb_d name: "Autonomous Database on Dedicated Exadata (ADB-D)" description: "Dedicated Exadata infrastructure, full isolation" use_when: - "Strict isolation requirements (regulatory, multi-tenant)" - "Predictable performance with no noisy-neighbor risk" - "Need RAC for horizontal compute scaling" - "> 128 OCPUs sustained" when_to_use: - "Oracle Database workloads migrating to cloud" - "Need automated patching, tuning, and scaling" - "BYOL opportunity with existing Oracle DB licenses" - "JSON document workloads (AJD)" - "Rapid APEX application development" - "Data warehouse / analytics with auto-scaling" - "Need Lakehouse-style SQL over Object Storage / Apache Iceberg without copying all data into the database" when_not_to_use: - "Non-Oracle database engines (use MySQL HeatWave, PostgreSQL, NoSQL)" - "Need OS-level access or custom DB home configuration → use DBCS" - "Need RAC but not ADB-D budget → use ExaCS or DBCS VM" - "Workloads requiring specific DB patches not yet in ADB" - "Extreme IOPS requirements > what Exadata provides" - "Choose TP workload if the dominant path is OLTP write-heavy access on database-local data and external/open-table access is peripheral" sizing: adb_s: min_ecpu: 2 max_ecpu: 512 min_ocpu_legacy: 1 max_ocpu_legacy: 128 auto_scaling: "Up to 3x base ECPUs (takes 2-3 minutes to activate)" sizing_guidance: "Size base ECPUs for P75 sustained load, not P50. Auto-scaling covers peaks but has activation latency. 4 ECPUs ≈ 1 OCPU." storage_min_tb: 1 storage_max_tb: 384 iops: "Proportional to OCPU count, Exadata storage servers" adb_d: min_ocpu: "2 per VM (minimum 2 VMs for HA)" max_ocpu: "Per Exadata infrastructure shape" rac: true storage: "Exadata storage dependent on infrastructure shape" gotchas: - id: adb_s_autoscale_latency severity: MEDIUM description: "Auto-scaling activation takes 2-3 minutes. Size base OCPUs for P75, not average." - id: adb_s_private_endpoint severity: HIGH description: "Always use private endpoints for production. Public endpoints allow internet access to the DB listener." - id: adb_s_service_gateway severity: MEDIUM description: "ADB-S requires a Service Gateway in the VCN for automated backups to Object Storage." - id: adb_s_byol_savings severity: LOW description: "BYOL pricing is ~50% cheaper than License Included. Always check if customer has existing Oracle DB licenses." - id: adb_d_minimum_infra severity: HIGH description: "ADB-D requires dedicated Exadata infrastructure — minimum cost is significant. Don't recommend for small workloads." - id: adb_data_guard_async severity: MEDIUM description: "Cross-region Autonomous Data Guard is async only (RPO > 0). For zero RPO, need same-region standby or ADB-D with local Data Guard." - id: adb_version severity: LOW description: "ADB now defaults to Oracle 26ai in all commercial regions. 19c is legacy. Verify customer version requirements." - id: adb_lakehouse_not_separate_engine severity: MEDIUM description: "Lakehouse is a workload specialization inside ADB Serverless, not a separate database engine or separate managed service." - id: adb_lakehouse_external_data_plane severity: MEDIUM description: "Lakehouse value appears when Object Storage / Iceberg / open-format access is structural. If the dataset is fully local and OLTP-dominant, TP is usually the cleaner fit." - id: adb_lakehouse_iceberg_limits severity: HIGH description: "Current Iceberg support has material caveats: no partitioned Iceberg tables, no credential vending, no query-time time travel, and some snapshot/schema changes require recreating the external table." - id: adb_lakehouse_tp_services severity: LOW description: "Lakehouse now supports TP and TPURGENT predefined service names. Do not model it as ADW-style query-only connectivity." implied_dependencies: - service: service_gateway reason: "Required for automated backups to Object Storage" - service: private_endpoint reason: "Production workloads must use private endpoints" - service: oci_vault reason: "Customer-managed encryption keys (recommended for production)" - service: data_safe reason: "Security assessment, audit, and data masking" competitive: vs_aws_rds: advantages: - "Autonomous operations (auto-patching, auto-tuning, auto-scaling)" - "Dramatically lower DB licensing cost with BYOL" - "Exadata performance for Oracle workloads" disadvantages: - "No multi-engine support (Oracle only)" - "Auto-scaling latency vs. Aurora's faster scale-out" vs_azure_sql: advantages: - "Oracle-native features (RAC, Data Guard, Advanced Compression)" - "BYOL cost advantage" disadvantages: - "Azure has broader SQL Server ecosystem integration" references: documentation: "https://docs.oracle.com/en/cloud/paas/autonomous-database/" pricing: "https://www.oracle.com/autonomous-database/pricing/" sizing_guide: "https://docs.oracle.com/en/cloud/paas/autonomous-database/adbsa/autonomous-database-cpu.html" workload_types: "https://docs.oracle.com/en/cloud/paas/autonomous-database/adbsa/about-autonomous-database-workloads.html" lakehouse: "https://docs.oracle.com/en-us/iaas/autonomous-database-serverless/doc/lake-lakehouse.html" iceberg: "https://docs.oracle.com/en/cloud/paas/autonomous-database/adbsa/query-external-data-apache-iceberg.html" lakehouse_tp_services: "https://docs.oracle.com/iaas/releasenotes/autonomous-database-serverless/2025-10-tp-tpurgent-predefined-db-service-names-for-lakehouse-workload.htm" data_lake_accelerator: "https://docs.oracle.com/iaas/releasenotes/autonomous-database-serverless/2025-11-data-lake-accelerator.htm" changelog: - date: "2026-04-24" contributor: { name: "Codex", team: "Field Architecture" } change: "Updated workload taxonomy to Autonomous AI naming; added Lakehouse selection guidance, Iceberg caveats, TP/TPURGENT note, and official references" - date: "2026-03-14" contributor: { name: "Diego Cabrera", team: "Field Architecture" } change: "Initial creation with sizing rules, gotchas, competitive notes, HA/DR options"