Ceph Data Durability Configuration Change (Planning)

Transitioning from 3x replication to 4+2 erasure coding (EC):

• ~2× improvement in usable storage capacity
• Comparable large-write performance
• Increased CPU overhead during writes

Management Nodes in production

3 management nodes now in production. System components are being migrated (via weighted affinity) to isolate them from user workloads (HTCondor, Jupyter notebook servers, etc.), improving cluster stability through fault isolation and more effective resource management.