A Comparative Study Between Containers and VMs with CPU Pinning, HugePages and NUMA Optimization

Cloud computing uses virtualization to offer flexible, scalable access to resources on demand. While this improves hardware utilization, it introduces CPU and memory overhead. Traditionally, Virtual Machines (VMs) have been the primary means of deploying cloud workloads. However, the advent of containerization offers a lightweight alternative that reduces latency and resource usage. This study aims to evaluate and compare the performance of VMs and containers under various optimization techniques, including Non-Uniform Memory Access (NUMA) tuning, CPU pinning, and memory management. Two experimental series are performed. The first assesses three configurations namely without optimization, with optimization across NUMA nodes, and with optimization within the same NUMA node. The second series examines CPU core affinity by comparing configurations with close vs. distant cores, both within and across NUMA nodes. Results under various stress test scenarios reveal that containers exhibit a more stable behavior, while the absence of optimization can significantly affect the overall system performance.