oci-deal-accelerator/kb/services/exadata-cloud.yaml

# OCI Service: Exadata Cloud Service (ExaCS)
# Last verified: 2026-04

service:
  name: "Exadata Cloud Service (ExaCS)"
  id: exacs
  category: database
  color: "#AA643B"

description: "Dedicated Exadata infrastructure running in OCI, providing the full performance and isolation of Exadata hardware with cloud elasticity. Supports Oracle RAC, Data Guard, and direct OS/CDB-level access for maximum DBA control."

when_to_use:
  - "Oracle RAC workloads requiring multi-node clustering"
  - "Regulatory or compliance scenarios demanding dedicated infrastructure isolation"
  - "High IOPS / low-latency database workloads (OLTP or mixed)"
  - "Sustained compute needs exceeding 128 OCPUs"
  - "Need direct OS-level or CDB-level access for custom patching, tuning, or agents"

when_not_to_use:
  - "Small workloads under 16 OCPUs — use Autonomous Database Serverless (ADB-S) or Base Database Service"
  - "No dedicated DBA team to manage patching, RAC, and infrastructure"
  - "Simple OLTP with no RAC or isolation requirements — use ADB-S instead"
  - "Dev/test environments where cost efficiency matters more than performance"

variants:
  - id: exacs_x9m
    name: "ExaCS X9M"
    description: "Exadata X9M shape — current-generation infrastructure with RDMA over Converged Ethernet and persistent memory acceleration."
    per_db_server:
      ocpus: 126
      memory_gb: 1390
      pmem_tb: 1.5
      flash_cache_tb: 25.6
    per_storage_server:
      usable_tb: 63.6
  - id: exacs_x11m
    name: "ExaCS X11M"
    description: "Exadata X11M shape — latest generation with AMD EPYC processors and XRMEM acceleration. Two storage server variants: HC (default, capacity-optimised) and EF (Extreme Flash, latency-optimised)."
    per_db_server:
      ecpus: 760
      memory_gb: 1390
      xrmem_gb: 1250
      flash_cache_tb: 27.2
      cpu: "2 × AMD EPYC 9J25 (96 cores each, 192 cores per server)"
    storage_server_variants:
      hc:
        name: "X11M High Capacity (HC)"
        description: "Default storage variant. 12 × 22 TB SAS HDDs + 27.2 TB flash cache + 1.25 TB XRMEM per SS. Best $/TB; same hot-data latency as EF (served from flash cache + XRMEM)."
        raw_tb_per_ss: 264
        flash_cache_tb_per_ss: 27.2
        xrmem_tb_per_ss: 1.25
        usable_disk_tb_per_ss: 80              # Datasheet "Total Usable Disk Capacity" per SS — post HIGH redundancy + drive-failure reserve, before compression
        max_db_size_tb_per_ss_no_local_backup: 64    # Derived: 192 TB / 3 SS at base config (datasheet Elastic Example 1)
        max_db_size_tb_per_ss_with_local_backup: 32  # Derived: 96 TB / 3 SS at base config
        total_cores_per_ss: 64
        recommended_for: "OLTP, mixed workloads, DWH where active set fits in flash cache (~80% of cases)"
      ef:
        name: "X11M Extreme Flash (EF)"
        description: "All-flash storage variant. Lower latency on cache-miss reads, but higher $/TB. Use when the working set exceeds flash cache or for latency-sensitive DWH full scans."
        usable_tb_per_ss: 52.5
        type: "Extreme Flash (EF)"
        recommended_for: "Latency-sensitive DWH with large full-table scans that miss flash cache; sub-ms consistent latency requirements"
    pricing_2026_04:
      source: "OCI Cost Estimator 2026-04-23"
      db_server_per_hour_usd: 2.9032
      storage_server_per_hour_usd: 2.9032
      base_config_monthly_usd: 10800   # (2 DB + 3 SS) × $2.9032 × 744 hr; excludes DB licenses (BYOL) and ECPU charges

sizing:
  model: "Elastic per-server (X9M+: 2-32 DB servers, 3-64 storage servers). 'Quarter/half/full rack' terminology is legacy; X9M and X11M are billed per-server."
  notes: |
    Three different storage figures appear in Oracle datasheets — they measure different things:
      • Total Usable Disk Capacity = ASM-allocatable space, post HIGH redundancy + drive-failure reserve, before compression
      • Max DB Size – No Local Backup = practical limit for DB allocation after RECO/SPARSE/DBFS reserves (this is what OCI Cost Estimator caps at)
      • Max DB Size – With Local Backup = ~half of above (FRA on disk for backups)
    Always quote both Usable Disk and Max DB Size to avoid customer confusion.
    OCI scales storage by adding individual storage servers, not in racks.

  base_configuration:
    description: "Minimum elastic configuration — what you get on day-1 before scaling."
    db_servers: 2
    storage_servers: 3

  x11m_elastic_base:
    description: "Datasheet 'Elastic Configuration Example 1' (2 DB + 3 SS, all-elastic SKUs)."
    source: "ADB-D X11M Datasheet v1.0, Jan 2025, Table 1"
    db_servers: 2
    storage_servers: 3
    total_ecpus: 1520               # 2 × 760
    total_memory_gb: 2780           # 2 × 1,390
    total_xrmem_tb: 3.75            # 3 × 1.25 (XRMEM lives on storage servers)
    total_flash_tb: 81.6            # 3 × 27.2
    hc_variant:
      total_usable_disk_tb: 240             # Datasheet authoritative
      max_db_size_no_local_backup_tb: 192   # OCI Cost Estimator caps allocation here
      max_db_size_with_local_backup_tb: 96
      max_sql_flash_bandwidth_gbps: 300
      max_sql_xrmem_bandwidth_gbps: 1500
      max_sql_disk_bandwidth_gbps: 5.4
      max_sql_read_iops: 5_600_000
      max_sql_write_iops: 3_000_000
      max_sql_disk_iops: 7_800
      max_data_load_rate_tb_per_hr: 7.5
    ef_variant:
      usable_tb: 157.5        # 3 × 52.5
    sample_elastic_growth_examples:
      compute_heavy: { db_servers: 8, storage_servers: 8, total_ecpus: 6080, max_db_size_tb: 512 }
      storage_heavy: { db_servers: 2, storage_servers: 14, total_ecpus: 1520, max_db_size_tb: 896 }

  x11m_base_system:
    description: "Hardware-generation-agnostic 'Base System' SKU — fixed-spec, NOT expandable, single VM Cluster only. Different product from elastic; entry-price option for small footprints."
    source: "ADB-D X11M Datasheet v1.0, Jan 2025, Table 1, footnote 1"
    db_servers: 2
    storage_servers: 3
    total_ecpus: 192
    total_memory_gb: 720
    total_flash_tb: 38.4
    total_usable_disk_tb: 73
    max_db_size_no_local_backup_tb: 58
    max_db_size_with_local_backup_tb: 29
    network: "10 GbE (vs 50 GbE on elastic)"
    notes: "Base System is NOT the same as 'Elastic Example 1 with 2 DB + 3 SS'. The elastic config has ~8× more ECPUs, ~4× more memory, ~3× more storage at same physical-server count, and 50 GbE networking. Base System is for cost-optimised entry; if customer expects to scale, go elastic from day 1."

  x9m_base:
    cpu_ocpus: 252            # 2 × 126
    memory_tb: 2.71           # 2 × 1,390 GB
    pmem_tb: 3.0              # 2 × 1.5 (persistent memory tier)
    flash_cache_tb_db: 51.2   # 2 × 25.6
    hc_variant:
      usable_tb_per_ss: 63.6
      base_usable_tb_high_redundancy: 190.8  # 3 × 63.6
    notes: "X9M HC drives are 14 TB (vs 22 TB on X11M)."

  legacy_x8_fixed_shapes:
    quarter_rack:
      storage_tb: 149
    half_rack:
      storage_tb: 299
    full_rack:
      storage_tb: 598
    notes: "X8 shapes are fixed-size (not elastic). Storage values are usable disk, not flash cache."

  ocpu_scaling: "OCPUs/ECPUs can be scaled online in increments without downtime. You pay for enabled cores, not the total available in the rack."
  sizing_guidance: "Start with the elastic base (2 DB + 3 SS) for most production workloads. Add storage servers when capacity or IOPS exceed base. Add DB servers when CPU saturates. Choose HC over EF unless workload is latency-sensitive on cache-miss reads."

gotchas:
  - id: exacs_minimum_cost
    severity: HIGH
    description: "Minimum infrastructure cost is approximately $6,000+/month for a quarter rack, regardless of OCPU usage. Ensure sustained workload justifies the commitment."

  - id: exacs_provisioning_time
    severity: MEDIUM
    description: "Exadata infrastructure provisioning takes 4-8 hours. Plan deployments accordingly — this is not an instant resource like a Compute VM."

  - id: exacs_patching_window
    severity: MEDIUM
    description: "Patching requires a scheduled maintenance window. Database nodes are patched in a rolling fashion but plan for reduced capacity during the process."

  - id: exacs_gi_db_patch_separate
    severity: MEDIUM
    description: "Grid Infrastructure (GI) and Database Home patches are applied separately and may have different schedules. Coordinate both to avoid version skew."

  - id: exacs_overprovisioning_risk
    severity: HIGH
    description: "Customers frequently over-provision by choosing half or full racks when a quarter rack with OCPU scaling would suffice. Always validate actual IOPS and CPU requirements before sizing up."

  - id: exacs_network_placement
    severity: LOW
    description: "ExaCS requires placement in a private subnet with sufficient IP addresses. Each DB node and each virtual IP consumes an IP; plan your subnet CIDR accordingly."

  - id: exacs_x11m_hc_vs_ef
    severity: MEDIUM
    description: "X11M ships in two storage variants — HC (default, 264 TB raw / SS) and EF (all-flash, 52.5 TB usable / SS). Most customers should choose HC: same hot-data latency (served from 27.2 TB flash cache per SS) at far better $/TB. Only pick EF when the working set genuinely exceeds flash cache and you need sub-ms consistent latency on cache-miss reads. Many quotes default to EF without justification — confirm with the customer."

implied_dependencies:
  - service: vcn
    reason: "ExaCS infrastructure must be deployed in a VCN"
  - service: private_subnet
    reason: "DB nodes and SCAN listeners require private subnet placement"
  - service: service_gateway
    reason: "Access to OCI object storage for backups and patching from private subnets"
  - service: oci_vault
    reason: "TDE wallet management and encryption key storage"

competitive_notes:
  vs_aws: "AWS has no equivalent dedicated Oracle Exadata infrastructure offering. RDS Oracle and RDS Custom lack RAC support and Exadata storage acceleration. Customers needing true Exadata performance have no AWS-native path."

references:
  documentation: "https://docs.oracle.com/en-us/iaas/Content/Database/Concepts/exaoverview.htm"
  pricing: "https://www.oracle.com/cloud/price-list/#exadata"

changelog:
  - date: "2026-03-14"
    contributor: { name: "Diego Cabrera", team: "Field Architecture" }
    change: "Initial creation with shapes, sizing, gotchas, dependencies"
  - date: "2026-04-23"
    contributor: { name: "Diego Velásquez", team: "Field Architecture" }
    change: "Added X11M HC variant (default) alongside existing EF; replaced misleading rack_configurations table that mixed X9M data with elastic per-server model; added X11M and X9M base configs; added pricing verified against OCI Cost Estimator (DB+SS @ $2.9032/hr); added gotcha on HC vs EF default choice."
  - date: "2026-04-23"
    contributor: { name: "Diego Velásquez", team: "Field Architecture" }
    change: "Replaced earlier estimator-derived storage figures (63 TB/SS, 190 TB at base) with authoritative numbers from ADB-D X11M Datasheet v1.0 Jan 2025 Table 1. X11M HC SS = 80 TB Total Usable Disk (HIGH redundancy + drive-failure reserve). 3-SS Elastic base = 240 TB Total Usable Disk OR 192 TB Max DB Size (no local backup, what the Estimator caps at) OR 96 TB Max DB Size (with local backup). Also added separate Base System SKU (fixed, non-expandable, 73 TB usable disk) — distinct product from elastic 2 DB + 3 SS. Added performance metrics (IOPS, GB/s) for elastic base."