A rack-by-rack plan for migrating off a proprietary NOS

Most failed network migrations try to change too much at once. Our approach moves one rack at a time and keeps the old and new fabrics interoperating until the last switch is replaced.

The sequence

1. Translate configurations automatically and review the diff by hand.
2. Rebuild the fabric in a lab with the same topology, then replay production traffic.
3. Cut over one rack during a change window, with the old image staged for rollback.
4. Repeat, speeding up as confidence and automation improve.

What to measure

Track convergence time, BGP session stability and application latency before and after each cutover. If any of them regress, pause and fix before moving to the next rack.