Module & NUMA Configuration

See Configuration Overview for the core_binded_node/core_binded_cores fields themselves. This page covers how to fill them in deliberately, using your node’s real NUMA topology.

Inspecting your node’s CPU and NUMA layout

Before setting core_binded_node/core_binded_cores by hand, find your node’s real CPU and NUMA layout. Two common tools: lscpu prints the NUMA node count and, for each node, the CPU ids that belong to it:

lscpu

NUMA node(s):          2
NUMA node0 CPU(s):     0-7
NUMA node1 CPU(s):     8-15

numactl --hardware (package numactl) gives the same CPU breakdown plus each NUMA node’s local memory:

numactl --hardware

available: 2 nodes (0-1)
node 0 cpus: 0 1 2 3 4 5 6 7
node 0 size: 131072 MB
node 1 cpus: 8 9 10 11 12 13 14 15
node 1 size: 131072 MB

Use this data the same way you’d use probed numa_specs below: pick a NUMA node, use its CPU list (or a subset) as core_binded_cores, and set core_binded_node to match.

Reading probed NUMA data

If you created your config with --probe-nodes (see What Changes on Your Nodes), each node’s config.numa_specs is filled in with something like:

"modules": [
  {
    "type": "rdb",
    "id": 11,
    ...,
    "config": {
      "numa_specs": {
        "node0": { "cpu": "0,1,2,3,4,5,6,7", "mem_mb": 131072 },
        "node1": { "cpu": "8,9,10,11,12,13,14,15", "mem_mb": 131072 }
      }
    }
  }
]

Each key is a NUMA node, with the CPU ids that belong to it and its local memory. This is recorded for reference; RDT doesn’t automatically bind modules to NUMA nodes for you, you decide the binding based on this data.

Setting core binding manually

To pin a module to a set of cores, set both fields in its config:

"modules": [
  {
    "type": "rdb",
    "id": 11,
    ...,
    "config": {
      "core_binded_node": "0",
      "core_binded_cores": "0-5"
    }
  }
]

core_binded_node identifies which NUMA node this binding corresponds to (for your own reference; it’s not cross-checked against numa_specs).
core_binded_cores accepts individual CPU ids, ranges, or a mix, comma-separated (e.g. 0-5, 0,1,2, or 0-3,8,10-11). Unlike the flat comma-separated list in numa_specs.cpu, ranges are written with a hyphen here.
If num_threads is left unset, it’s derived automatically from the core count. If you set it explicitly, it must not exceed the number of cores in core_binded_cores.

A typical pattern: pick a NUMA node from numa_specs, use a subset (or all) of its CPU list as core_binded_cores, and set core_binded_node to that NUMA node’s id.

Self-Hosted Deployment

Manual Deployment

Regatta Deployment Tool

Module & NUMA Configuration

Inspecting your node’s CPU and NUMA layout

Reading probed NUMA data

Setting core binding manually

​Inspecting your node’s CPU and NUMA layout

​Reading probed NUMA data

​Setting core binding manually

Inspecting your node’s CPU and NUMA layout

Reading probed NUMA data

Setting core binding manually