Understanding NUMA: Features and Role in High-Performance Virtualization

Modern telecom and cloud infrastructures rely on high-performance servers that must process massive amounts of data with minimal latency. One of the key hardware designs enabling this is NUMA (Non-Uniform Memory Access). For anyone working with NFV or virtualized workloads, understanding NUMA and its features is essential.

What is NUMA?

NUMA is a computer memory design used in multiprocessor servers. In NUMA systems, memory is divided into regions (NUMA nodes) that are directly attached to specific CPU sockets. While each CPU can access all memory, access to its local memory is faster than accessing memory attached to another CPU.

This design reduces memory bottlenecks, especially in servers with multiple processors.

Key Features of NUMA

1. NUMA Affinity

NUMA affinity ensures that a VM’s virtual CPUs (vCPUs) and its memory allocations are kept within the same NUMA node.

• Benefit: Reduces memory latency and improves throughput for demanding VNFs like packet gateways or core network functions.

• Risk: Poor placement may cause cross-socket memory access, leading to degraded performance.

2. I/O NUMA

I/O NUMA extends the concept of memory affinity to network and storage interfaces. In modern servers, PCIe devices (like NICs or GPUs) are attached to specific NUMA nodes. Placing a VM or VNF close to the NUMA node where its I/O device resides minimizes latency for data transfer.

• Example: A VNF handling high packet throughput performs better if its vCPUs and memory are aligned with the NUMA node connected to the NIC.

3. CPU Pinning with NUMA Awareness

NUMA works best when combined with CPU pinning, where vCPUs are bound to physical cores within the same NUMA node. This ensures consistent performance and avoids “cross-node” scheduling overhead.

4. Memory Locality Optimization

NUMA allows memory allocation policies, such as:

• Local allocation: Prefer local memory of the NUMA node.

• Interleaved allocation: Distribute memory evenly across nodes to balance load.
  These policies give operators flexibility to tune performance based on workload requirements.

Why NUMA Matters in NFV

In telecom environments, where VNFs and CNFs must process packets at line rate, NUMA-awareness is crucial. Poor NUMA configuration can lead to unpredictable latency, jitter, and throughput issues. Properly leveraging NUMA affinity, I/O NUMA, and CPU pinning ensures that workloads get the best possible performance out of the hardware.

Conclusion

NUMA is not just a server hardware design — it’s a performance enabler for virtualization. By understanding features like NUMA affinity, I/O NUMA, and memory locality policies, operators can maximize efficiency, minimize latency, and deliver the reliability modern telecom networks demand.