한국정보처리학회:학술대회논문집 (Annual Conference of KIPS)

한국정보처리학회 (Korea Information Processing Society)
권호 표지

Performance Analysis of Metric-Based Scaling in Kubernetes Environments: A Comparative Study of CPU Utilization and Custom Metric Approaches

  • Jin-Cheol Jung (Dept. of Software Convergence, Kyung-Hee University)
  • 발행 : 2024.10.31
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

This study compares CPU-based and custom metric-based scaling methods in Kubernetes, showing that custom metrics tailored to application needs can enhance scalability and efficiency. Findings reveal that, at certain scaling thresholds under dynamic network traffic, custom metrics reduce average latency by 85% to 87% compared to CPU-based scaling.

키워드

참고문헌

  1. Anjaly Parayil et al. "Towards Cloud Efficiency with Large-scale Workload Characterization", arXiv:2405.07250v1, 2024
  2. Cilium, https://cilium.io/.
  3. Vegeta, https://github.com/tsenart/vegeta
  4. Mpstat, https://linux.die.net/man/1/mpstat
  5. Prometheus adapter, https://github.com/kubernetessigs/prometheus-adapter