What is the architecture of Exadata Full Rack?
An Exadata Full Rack is the ultimate expression of Oracle’s "Engineered System" philosophy. It is a massive, pre-integrated unit of compute, storage, and networking designed to handle the world’s largest and most mission-critical database workloads.
While you can start with a "Quarter Rack" or "Half Rack," the Full Rack represents a complete, balanced ecosystem of hardware that fills a standard 42U cabinet.
A standard Exadata Full Rack typically follows a "balanced" configuration. Using the latest generations (like the X10M) as a reference, a Full Rack consists of:
8 Database Servers: High-performance compute nodes (the "Brains") where your Oracle instances run.
14 Storage Servers: Intelligent storage cells (the "Brawn") that handle Smart Scans and data persistence.
RoCE Network Fabric: Redundant 100Gbps switches that connect every node in the rack.
Power & Management: Redundant Power Distribution Units (PDUs) and an Ethernet switch for administrative management and monitoring.
The Full Rack architecture is built on the principle of redundancy at every level. Because it has 8 compute nodes and 14 storage nodes, the system is designed to keep running even if multiple components fail simultaneously.
Grid Redundancy: Data is mirrored across the 14 storage cells (using High or Normal redundancy). If a whole storage server dies, the database doesn't even "blink"—it just pulls the mirrored data from other cells.
RAC Connectivity: With 8 database servers, you can run an 8-node Oracle RAC cluster. If one server needs maintenance or crashes, the other 7 take over the user sessions instantly.
Dual Network Paths: Every server has dual connections to the RoCE fabric. If a cable or a switch fails, the traffic reroutes in microseconds.
The Full Rack is not a "closed" system. It uses a Scale-Out architecture.
If you outgrow a single Full Rack, you don't replace it. You simply connect a second rack using the RoCE expansion ports. The Database and Storage layers see the new rack as additional resources, expanding your "pool" of CPU and Disk seamlessly. This allows companies to scale from a few terabytes to hundreds of petabytes.
When you put all these components together in a Full Rack, the numbers become staggering:
| Feature | Approximate Full Rack Capacity (X10M) |
| Total CPU Cores | ~1,500+ Cores (Database + Storage) |
| SQL Read Throughput | Up to 1 Terabyte per second |
| IOPS | Up to 25+ Million |
| Storage Capacity | Multiple Petabytes (HC Model) |
| Networking | 100Gbps RoCE Internal Fabric |
A Full Rack is generally used for two main scenarios:
Massive Consolidation: Instead of having 500 separate servers scattered across a data center, a company moves all of them into one Exadata Full Rack. I/O Resource Management (IORM) ensures they all play nice together.
Giant Data Warehousing: For applications like global banking, telecommunications, or real-time analytics, where the database is tens or hundreds of terabytes and thousands of users are querying it simultaneously.
The Exadata Full Rack is the "Gold Standard" for database architecture. It combines massive parallel processing with a high-speed, low-latency fabric, ensuring that as your data grows, your performance doesn't degrade. It is essentially a private cloud in a box, engineered to run the Oracle Database faster and more reliably than any other platform on Earth.