Never Underestimate Memory Architecture - NUMA and SMT Performance Impact in Cloud Environments

Explore how NUMA (Non-Uniform Memory Access) and SMT (Symmetric Multi-Threading) technologies significantly impact application performance in modern cloud environments in this 30-minute conference talk from CNCF. Demystify these critical memory architecture concepts by understanding what NUMA and Hyperthreading are, how they function, and why they matter for cloud-native applications. Discover how NUMA can cause memory access to be significantly slower when connected to different CPUs, and learn why Hyperthreading makes single CPU cores appear as two without doubling performance. Examine Kubernetes integration with NUMA architecture and understand the current limitations of this interaction. Analyze real-world performance measurements from AWS and Google Cloud platforms to see concrete examples of how these technologies affect application behavior. Gain practical visibility tools by learning to use Prometheus metrics and Linux commands to monitor and assess your servers' NUMA and SMT configurations. Acquire the knowledge and measurement tools necessary to evaluate the impact of memory architecture on your workload performance in cloud-native environments.