Uploaded November 2025 | Updated September 2026, 3 weeks ago
Why We Have 17 Clusters Instead of One (and Why You Probably Shouldn’t Do That) - Ondřej Kukla & Petr Tlapa, CDN77
We’ll share an unconventional architecture we adopted for a project where reliability and the ability to scale by two to three orders of magnitude were critical. Instead of building a single massive Ceph RGW cluster, we split storage across seventeen independent clusters, with a load balancer using consistent hashing to distribute requests seamlessly.
This approach allowed us to achieve extreme scalability and strong fault isolation, ensuring that the failure of a single cluster would not disrupt the entire storage stack and just it’s part. However, running multiple clusters comes with significant challenges, from operational overhead to the heightened need for robust monitoring and cross-cluster coordination.
Why We Have 17 Clusters Instead of One (and Why You Probably Shouldn’t Do That) - Ondřej Kukla & Petr Tlapa, CDN77
We’ll share an unconventional architecture we adopted for a project where reliability and the ability to scale by two to three orders of magnitude were critical. Instead of building a single massive Ceph RGW cluster, we split storage across seventeen independent clusters, with a load balancer using consistent hashing to distribute requests seamlessly.
This approach allowed us to achieve extreme scalability and strong fault isolation, ensuring that the failure of a single cluster would not disrupt the entire storage stack and just it’s part. However, running multiple clusters comes with significant challenges, from operational overhead to the heightened need for robust monitoring and cross-cluster coordination.
