How does east-west traffic impact RAC hardware?
In an Oracle Real Application Clusters (RAC) environment, East-West traffic is the lifeblood of the system. Unlike typical web traffic that moves "North-South" (from the internet to the server), East-West traffic refers to the massive volume of data moving horizontally between the database nodes themselves.
On the hardware level, this traffic doesn't just "impact" RAC—it defines the hardware requirements for the entire cluster.
In RAC, the hardware is physically connected via a dedicated, private network called the Interconnect. This is where East-West traffic lives.
Cache Fusion: When Node A needs a data block that is currently sitting in the memory (SGA) of Node B, RAC uses Cache Fusion to ship that block over the wire.
Hardware Pressure: Because moving data across a network is naturally slower than reading from local RAM, the hardware must be ultra-low latency. This is why Oracle RAC on OCI or Exadata relies on RoCE (RDMA over Converged Ethernet) or InfiniBand.
Every East-West packet sent over a traditional network stack requires the CPU to stop its database work to process network headers.
Context Switching: If the East-West traffic is handled via standard TCP/IP, the CPU cores are constantly interrupted. This can lead to a "CPU Steal" scenario where your database throughput drops because the cores are too busy being network controllers.
The Hardware Solution (RDMA): High-end RAC hardware uses Remote Direct Memory Access (RDMA). This allows one node to "reach into" the memory of another node and grab a data block without involving the remote CPU at all. This keeps the CPU dedicated 100% to SQL execution.
East-West traffic in RAC is "bursty." When a large query triggers a global cache transfer, it can saturate a network switch port instantly.
Micro-bursts: If the Spine-Leaf switches (the hardware connecting the racks) have small buffers, these bursts lead to dropped packets.
The Result: In RAC, a single dropped packet on the interconnect causes a "retransmission timeout," which manifests to the user as a "gc current block 2-way" wait event—essentially a multi-millisecond freeze in the database.
Hardware Fix: RAC-optimized switches use Deep Buffers and Priority Flow Control (PFC) to ensure that "East-West" database traffic is never dropped, even if the "North-South" public traffic is congested.
The more nodes you add to a RAC cluster, the more East-West traffic is generated. This is known as the "cluster overhead."
The Hardware Wall: If your hardware interconnect is only 10Gbps, you might find that adding a 4th or 5th node actually makes the database slower. The East-West traffic becomes so dense that the nodes spend more time talking to each other than doing work.
Modern Standard: In 2026, OCI shapes for RAC utilize 100Gbps or 200Gbps interconnects. This massive "hardware headroom" allows the cluster to scale linearly, ensuring that the East-West traffic never hits a physical bottleneck.
| Hardware Component | Impact of East-West Traffic | Why it Matters |
| Network Card (NIC) | Requires RDMA / RoCE support. | Eliminates CPU overhead during Cache Fusion. |
| Switch | Requires Spine-Leaf topology. | Ensures deterministic, 1-microsecond latency. |
| Cabling | Requires high-bandwidth (100G+). | Prevents "Cluster Congestion" as nodes are added. |
| CPU | Higher core counts needed for high concurrency. | Manages the coordination of global locks. |
"In a RAC environment, the network is the computer. East-West traffic is the invisible force that determines whether your database scales effortlessly or collapses under its own coordination overhead. When you invest in high-end Interconnect hardware, you aren't just buying speed; you're buying the ability for multiple physical servers to act as a single, unified brain."