NUMA architecture stands for NUMA, a design used in modern multi-CPU servers to improve performance and scalability.

🧠 Simple Idea

In a NUMA system:

• Each CPU has its own local memory
• All memory is shared, but:
  • Accessing local memory = fast
  • Accessing remote memory (other CPU) = slower

👉 That’s why it’s called Non-Uniform memory access.

🏗️ How NUMA Works

In servers like Dell PowerEdge:

• Each CPU socket forms a NUMA node
• Each node contains:
  • CPU cores
  • Memory (RAM)
• Nodes are connected by high-speed links (Intel UPI / AMD Infinity Fabric)

📊 Example

2-Socket Server:

• CPU 1 → 128 GB RAM (Node 0)
• CPU 2 → 128 GB RAM (Node 1)

Scenario:

• A process running on CPU 1:
  • Uses Node 0 memory → ✅ Fast
  • Uses Node 1 memory → ⚠️ Slower

⚡ Why NUMA is Important

NUMA improves:

• Performance scalability
• Memory bandwidth
• CPU efficiency

👉 Essential for:

• Virtualization
• Databases (Oracle, SQL Server)
• AI / analytics workloads

🧩 NUMA in Virtualization

With hypervisors like:

• VMware ESXi
• Microsoft Hyper-V

NUMA helps:

• Place VMs close to memory
• Reduce latency
• Improve VM performance

⚙️ Key NUMA Concepts

🔹 NUMA Node

• CPU + its local memory

🔹 Local vs Remote Access

• Local = fast
• Remote = slower

🔹 NUMA Awareness

• OS or hypervisor optimizes workload placement

🚀 Best Practices

• Keep workloads within one NUMA node when possible
• Balance RAM across CPUs
• Enable NUMA in BIOS (do NOT disable)
• Use NUMA-aware applications

⚠️ Common Mistakes

• Disabling NUMA (huge performance loss)
• Uneven memory configuration
• Oversized VMs spanning multiple NUMA nodes

🧠 Simple Definition

NUMA is a server architecture where each CPU has its own local memory, and accessing that memory is faster than accessing memory attached to another CPU.