STM Journals Peer-reviewed research from Consortium e-Learning Network

Recent Trends in Parallel Computing Review Article

Memory Reclamation Behavior Under cgroups v2 Versus Legacy cgroups: A Measurement Framework for Containerized Workloads

  1. Nidhi Chahal Electronics and Communication Engineering, Chandigarh Engineering College-CGC
  2. Tinu Anand Kumar Electronics and Communication Engineering, Chandigarh Engineering College-CGC
  3. Reshika Electronics and Communication Engineering, Chandigarh Engineering College-CGC

Abstract

Linux control groups (cgroups) are the kernel mechanism that lets container runtimes such as Docker, containerd, and CRI-O enforce per-container CPU, memory, and I/O limits, and the transition from the legacy cgroups v1 hierarchy to the unified cgroups v2 hierarchy changed several aspects of how the memory controller signals and responds to reclaim pressure most notably through the introduction of memory.low/memory.high soft-limit semantics and system-wide Pressure Stall Information (PSI). Kernel documentation and a small number of practitioner benchmarks describe these interface-level changes, and separate networking-focused studies have measured cgroups v2's lower CPU-instruction overhead relative to v1, but we are not aware of published work that measures, under controlled and reproducible conditions, how the two hierarchies actually behave when several memory-constrained containers compete for host memory on the same node—the scenario that determines achievable container density in practice. This paper makes two contributions. First, it synthesizes the documented and practitioner-reported differences between the v1 and v2 memory controllers into a specific set of hypotheses about reclaim behavior under contention (soft-limit responsiveness, page-cache versus anonymous-memory eviction balance, and OOM-kill fairness across co-located containers). Second, it specifies a controlled, reproducible benchmark protocol using a memory-pressure workload generator, a fixed host configuration, and a defined density-scaling procedure for testing these hypotheses directly, and it reports the results section as an execution plan with the intended tables rather than fabricated benchmark numbers, since the benchmark has not yet been run on physical or virtualized hardware at the time of writing. The protocol is designed so that any reader with access to a Linux host supporting both hierarchies can execute it and populate the results independently.

Keywords

References (10)

  1. Billimoria KN. Linux kernel programming: a comprehensive and practical guide to kernel internals, writing modules, and kernel synchronization. Birmingham (UK): Packt Publishing; 2024 Feb 29.
  2. Jain SM. Cgroups. In: Linux containers and virtualization: utilizing Rust for Linux containers. Berkeley (CA): Apress; 2023. p. 47–81.
  3. Wiedner F, Daichendt A, Andre J, Carle G. Control groups added latency in NFVs: an update needed? In: 2023 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN); 2023 Nov 7. p. 40–45.
  4. Anderson J, Hu H, Agarwal U, Lowery C, Li H, Apon A. Performance considerations of network functions virtualization using containers. In: 2016 International Conference on Computing, Networking and Communications (ICNC); 2016 Feb 15. p. 1–7.
  5. Wang Q, Kar S, Mishra P, Linduff C, Izard R, Anjam K, et al. Container resource allocation versus performance of data-intensive applications on different cloud servers [Preprint]. 2023. arXiv:2311.07818.
  6. Brigoli R, Masood F. MLOps with Red Hat OpenShift: a cloud-native approach to machine learning operations. Birmingham (UK): Packt Publishing; 2024 Jan 31.
  7. Takkalapally D. Engineering for scale: performance testing and capacity planning in OpenShift and Pivotal Cloud Foundry. Int J Comput Sci Eng Innov. 2026 Feb 24:1–6.
  8. Gedia D, Perigo L. Performance evaluation of SDN-VNF in virtual machine and container. In: 2018 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN); 2018 Nov 27. p. 1–7.
  9. Dziurzanski P, Indrusiak LS. Value-based allocation of Docker containers. In: 2018 26th Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP); 2018 Mar 21. p. 358–362.
  10. Micallef Z, Rajenthiram K, Gerostathopoulos I. A systematic review of MLOps tools: tool adoption, lifecycle coverage, and critical insights [Preprint]. 2026. arXiv:2604.16371.
Support