How does Exadata use NVMe drives?
In the Exadata architecture, NVMe (Non-Volatile Memory Express) isn't just a type of drive; it is the "express lane" that connects the storage media directly to the CPU.
While traditional storage arrays use older protocols (like SAS or SATA) that were designed for slow, spinning disks, Exadata uses NVMe to unlock the true speed of modern flash memory.
In a standard server, data from a disk has to travel through a "storage controller," which acts like a slow toll booth. NVMe changes this by connecting the flash drives directly to the PCIe bus.
The Result: This provides a direct, high-bandwidth path to the CPU.
The Speed: In modern Exadata models (like the X10M), each storage server uses NVMe PCIe flash cards capable of massive throughput—delivering up to 600+ GB/s of scan throughput across a single rack.
Most Exadata deployments use the HC model, which balances speed and cost by mixing 12 large spinning disks with 4 high-speed NVMe flash cards.
In this setup, NVMe drives are used for:
Exadata Smart Flash Cache: The most frequently accessed data is automatically moved to the NVMe cards. Because of the direct PCIe connection, the database can read this data in roughly 100 microseconds.
Smart Flash Log: To speed up "Commits," Exadata writes redo logs to the NVMe flash in parallel with the disks. The database only has to wait for the faster NVMe write to finish.
For workloads that require absolute maximum performance (like massive all-flash data warehouses), the Extreme Flash model removes spinning disks entirely.
All-NVMe Tiering: In the latest X10M EF servers, Oracle uses a "tiered" NVMe approach. It combines Performance-Optimized NVMe (for caching) with Capacity-Optimized NVMe (for primary data storage).
The Benefit: You get the low latency of NVMe across 100% of your data, not just the "hot" data.
NVMe's massive parallelism (the ability to handle thousands of requests at once) is what allows Smart Scan to be so effective. When the Cell Offload Engine decides to scan a table, it can pull data from the NVMe drives using multiple "queues" simultaneously. This ensures the Storage Cell's CPUs are constantly fed with data, allowing them to filter and decompress millions of rows per second.
| Feature | Traditional SAS/SATA SSD | Exadata NVMe PCIe |
| Interface | Serial (One lane) | Parallel (Multiple PCIe lanes) |
| Command Queues | 1 queue (32 commands) | 64,000 queues (64k commands each) |
| Latency | ~200 - 500 µs | < 100 µs (Physical access) |
| Internal Path | Through a Controller | Directly to CPU/Memory |
By using NVMe as the foundation for its storage tier, Exadata ensures that the hardware can keep up with the software. Whether it's accelerating a single transaction via the Smart Flash Log or scanning terabytes of data via Smart Scan, NVMe provides the raw, parallel horsepower that makes Exadata the fastest platform for Oracle Database.