What is the architecture of Exadata storage grid?
To understand why Oracle Exadata is often called the "world's fastest database machine," you have to look under the hood at the Exadata Storage Grid.
Unlike traditional storage that simply acts as a "dumb" repository for bits and bytes, the Exadata Storage Grid is a massively parallel, scale-out architecture that brings intelligence to the data layer.
The grid is composed of multiple Storage Servers (also known as Storage Cells). Each cell is a self-contained unit of power:
CPU: Multiple Intel Xeon processors dedicated to running storage software.
Memory: Massive amounts of RAM for caching.
Flash: High-performance NVMe Flash (Exadata Smart Flash Cache).
Capacity: High-capacity SAS disks or high-performance Extreme Flash drives.
The secret sauce is the Exadata Storage Server Software (Cellsrv). This software runs on every storage cell and allows the storage to "understand" Oracle Database blocks.
In a traditional SAN, the database asks for a block, and the storage sends the whole block back. In the Exadata Storage Grid, the database says, "Give me the names of customers who spent over $1,000," and the storage cell filters the data at the disk level, returning only the relevant rows.
The Database Servers and the Storage Grid are connected by a high-bandwidth, low-latency 100Gbps RoCE fabric.
RDMA (Remote Direct Memory Access): This allows the database to pull data directly from the storage cell’s memory without involving the OS kernel or CPU on either side.
This eliminates the "bottleneck" typically found in standard Fibre Channel or iSCSI networks.
The Storage Grid organizes physical hardware into logical layers to ensure data is both protected and accessible:
| Layer | Description |
| LUNs | Physical disks (HDD or Flash) mapped to the software. |
| Cell Disks | Abstracted layers over LUNs that manage partitioning and interleaving. |
| Grid Disks | The actual slices of storage presented to Oracle ASM. |
| ASM Diskgroups | The final layer where data is striped and mirrored across all cells in the grid. |
The architecture enables three specific technologies that define Exadata:
Smart Scan: Offloads SQL processing (filtering and column projection) to the storage CPUs.
Hybrid Columnar Compression (HCC): Compresses data by grouping it into "Compression Units" to reduce the I/O footprint significantly.
Storage Indexing: A memory-resident structure in the storage cell that tracks the Min/Max values of columns, allowing the grid to skip entire regions of data that don't match a query's WHERE clause.
The Exadata Storage Grid isn't just a place to store data; it’s a distributed compute engine. By moving the processing to the data—rather than moving the data to the processing—it solves the I/O bottleneck that plagues traditional enterprise architectures.
Would you like me to dive deeper into how Smart Scan specifically interacts with this architecture to speed up big data queries?