Module numa 

NUMA-Aware Scheduling

Optimizes process placement for Non-Uniform Memory Access (NUMA) architectures.

NUMA Background

Modern multi-socket systems have NUMA characteristics where memory access latency depends on which CPU socket is accessing which memory node. Local memory access is faster than remote access (typical ratio: 1.0x vs 1.5-2.0x).

Optimization Strategy

1. Memory affinity: Schedule processes on CPUs close to their memory
2. Load balancing: Balance load within NUMA nodes before cross-node migration
3. Page migration: Move pages to local node when access patterns change
4. Interleaving: Distribute memory across nodes for bandwidth-intensive workloads

ACPI Topology Parsing

On x86_64, NUMA topology is discovered from ACPI tables:

• SRAT (System Resource Affinity Table): CPU-to-domain and memory-to-domain mappings
• SLIT (System Locality Information Table): inter-node distance matrix
• MADT: CPU enumeration including offline CPUs

Structs

CpuInfo

CPU information from MADT parsing.

NodeLoad

Per-node load statistics

NumaNode

Represents a discovered NUMA node.

NumaScheduler

NUMA scheduler

NumaTopology

NUMA topology information

SlitEntry

Parsed SLIT (System Locality Information Table) data.

Enums

SratEntry

SRAT sub-table entry types.

Functions

build_topology

Build a NumaTopology from parsed SRAT and SLIT data.

get_distance

Get the distance between two NUMA nodes from a parsed SLIT entry.

get_numa_scheduler

Get global NUMA scheduler

init

Initialize NUMA-aware scheduling

parse_madt_topology

Parse MADT to extract CPU topology.

parse_slit

Parse the raw SLIT table bytes.

parse_srat

Parse the raw SRAT table bytes into structured entries.

Type Aliases

CpuId

CPU identifier

NodeId

NUMA node identifier