What Is Subnet Manager?_
The Subnet Manager (SM) is a critical software component in an InfiniBand network that initializes and manages the fabric. It discovers all switches, HCAs, and cables, computes optimal routing paths, and configures forwarding tables. Without an active SM, no data traffic can flow across the fabric.
Technical Details
The SM runs on a designated node (often a management server or one of the GPU servers) and uses subnet management packets (SMPs) to query and set switch forwarding tables. It supports multiple routing algorithms (e.g., LID-based, adaptive) and can run in a master-standby configuration for redundancy. The SM must be restarted or fail over if the fabric topology changes (e.g., after a cable swap).
How Leviathan Systems Works with Subnet Manager
During commissioning, we verify that the Subnet Manager is running and correctly discovers all InfiniBand switches and GPU nodes. If the SM fails or misconfigures routes, we see link flapping or no fabric connectivity, requiring a restart or topology rescan.
Related service
Network Testing & Commissioning →