In a multi-node Oracle RAC (Real Application Clusters) environment, the biggest performance killer is often "Global Cache Latency." This happens when Node A needs a data block that is currently sitting in the memory (buffer cache) of Node B.

Traditionally, this requires a "three-hop" dance involving the network and the OS kernel. Exadata Smart Fusion Block Transfer eliminates this overhead by using the RoCE/InfiniBand fabric to move blocks between nodes at near-memory speeds.

1. The Problem: The Standard RAC "Ping"

In a standard (non-Exadata) cluster, if Node 1 wants a block held by Node 2:

1. Node 1 sends a request to the Global Enqueue Service (GES).

2. Node 2 receives the request, processes it through the OS kernel, and prepares the block.

3. Node 2 sends the block over the private interconnect to Node 1.

Even with fast networking, this involves CPU interrupts and context switches on both servers, adding significant latency to every "block ping."

2. The Solution: Smart Fusion

Smart Fusion Block Transfer leverages the tight integration between the Oracle Database and the Exadata hardware. It specifically optimizes the Log Write part of the block transfer.

When Node 2 sends a block to Node 1, it traditionally has to wait for the Log Writer (LGWR) to finish flushing the redo logs to disk to ensure data integrity. Smart Fusion allows Node 2 to send the block before the redo log write is even acknowledged by the storage.

How it works:

• The "Fusion" Point: The receiving node (Node 1) is informed that the block is coming.

• Parallel Processing: Instead of Node 2 waiting for the log write to complete before sending the block, it sends the block immediately.

• The Handshake: The storage cell sends the acknowledgment of the log write directly to the requesting node (Node 1).

3. The RDMA Advantage

In X8M, X9M, and newer RoCE-based Exadata models, Smart Fusion is supercharged by RDMA (Remote Direct Memory Access).

Because the nodes use RoCE, the block transfer happens memory-to-memory. Node 1 can pull the data directly from Node 2’s RAM without taxing Node 2’s CPU. This reduces the "Interconnect Latency" from hundreds of microseconds to single-digit microseconds.

4. Key Benefits for Your Workload

• Scalability: By reducing the cost of block transfers, you can add more nodes to a RAC cluster without the "diminishing returns" typically caused by interconnect congestion.

• OLTP Speed: In high-concurrency environments (like a busy e-commerce site) where multiple nodes are updating the same set of tables, Smart Fusion keeps the "cache hits" fast.

• Reduced CPU Overhead: Because the transfer is offloaded to the network hardware, more CPU cycles are available for actual SQL processing.

[Image comparing RAC interconnect latency with and without Exadata Smart Fusion]

5. Summary: Standard RAC vs. Smart Fusion

The Bottom Line

Smart Fusion Block Transfer makes a multi-node cluster feel like one giant, unified machine. It effectively removes the "physical distance" between the memory of different servers, ensuring that Oracle RAC performs at its peak on Exadata hardware.