CTL provider online.
CUCM is picky about MAC addresses. Change a virtual NIC's MAC after installation, and the entire node's certificate chain explodes. She'd learned that the hard way during testing. Tonight, she triple-checked the port group settings: VLAN 10 for PUB, VLAN 11 for SUB1, VLAN 12 for SUB2. The Cisco switchports were pre-configured with spanning-tree portfast and switchport voice vlan . The VMs would never know they weren't physical.
CUCM's virtualized heartbeat timers are notoriously sensitive. In a physical world, a 200ms delay is a shrug. In a hypervisor, if the ESXi host gets busy, that same delay can trigger a "node isolation" event. The cluster would split-brain faster than you could say "call manager group."
She looked at the old, dead Big Yellow sitting in the corner. Then at her screen, where three clean, green VM icons showed 0% packet loss and perfect database replication.
She disabled DRS automation for the CUCM cluster. No automatic vMotion. Ever. She set an anti-affinity rule to keep Publisher and Subscribers on different physical hosts. And she wrote a big, red warning in the runbook:
The sun was rising. Her boss walked in, saw the green "All Systems Operational" dashboard, and grunted. "Good. Now document it. We're virtualizing the rest next month."
She closed her laptop, grabbed her jacket, and finally threw away that cold coffee.
Mariana leaned back. The virtualized CUCM wasn't just a backup—it was better . No more spinning disks. No more single points of failure. The UCS chassis had redundant PSUs, redundant fabric interconnects, and vMotion. If a host failed, the CUCM VMs would restart on another host in under two minutes.