Advanced · Module 9 · Lesson 7 of 13

NUMA Architecture, Memory Interconnects, and CPU Pinning

Architect NUMA-aware multi-threaded systems, manage local vs. remote memory interconnect latency, configure memory allocation policies with libnuma, and bind execution threads with pthread_setaffinity_np.

ProAbout 16 minutes

By the end you can

Differentiate Uniform Memory Access (UMA) from Non-Uniform Memory Access (NUMA) topologies
Trace cross-socket interconnect penalties across Intel UPI and AMD Infinity Fabric
Diagnose the Linux kernel first-touch allocation policy and eliminate remote memory access bottlenecks
Pin worker threads to dedicated CPU cores using pthread_setaffinity_np and CPU_SET

Builds on Cache-Conscious Data Structures: Alignment, False Sharing, and Structure of Arrays.

01 · A question first

Why does a multi-threaded C service run at 10 million operations per second on a single-socket workstation, yet drop to 4 million operations per second on an enterprise dual-socket server with twice as many cores? The answer lies in NUMA node boundaries and cross-socket memory interconnect penalties.

Make a guess before you read on.

02 · The idea

Members only

Unlock the full lesson

This lesson is part of the membership. Join to read it in full, along with every other premium deep dive.

What's inside

Start with the problem
How the idea works
Production Implementation: NUMA-Aware Worker Pool
The libnuma Memory Allocation Policies
Why it is correct
What it costs
Pitfalls and variations
Where it is used