Performance Benchmark Report

K8s-Ceph Production Cluster Analysis • June 2025

Executive Summary

4-node Kubernetes cluster with Rook Ceph storage. Healthy, stable, resilient with 3x replication over 10GbE. Network-bound performance (not storage). Production-ready for HA workloads.

Test Environment

💻

Compute

4× Xeon D-1528

🧠

Memory

128 GB/node

💾

Storage

8× Samsung 980

🌐

Network

10GbE / 3× Rep

Performance Metrics

Random I/O

Total IOPS: 22.3K
Read: 16.7K
Write: 5.6K

Strong distributed performance. Write impacted by 3× replication.

Sequential I/O

Read: 1,150 MB/s
Write: 380 MB/s
Network Cap: 1,250 MB/s

Network-bound. NVMe drives underutilized.

Key Findings

  • Network bottleneck: 10GbE caps performance, not NVMe
  • Healthy cluster: Consistent, reliable node performance
  • Production ready: HA workload capable
  • Upgrade path: 25GbE+ would unlock NVMe potential