Planning waves against a measured copy rate

A wave is the set of machines that fit in one maintenance window. Building one needs two things from you: the windows you actually have, and a measured copy rate.

The copy rate is measured, never assumed. Copy one representative machine end to end and divide. Not the array's rated throughput, not the link speed — what a real copy actually achieved on this path, on a night like the one you are planning.

Both this form and the server refuse a missing or zero rate, and that refusal is deliberate. Every downtime figure in the plan is derived from that one number, so a guess there is a guess in every row, and the overrun lands in somebody else's outage notice at 3am.

Machines are placed largest first. A long copy that will not fit anywhere should be discovered while there is still time to split its disks, negotiate a longer window, or decide it moves on its own night — not at the end, when only scraps of capacity are left.

Read the unplaced list. It is as important as the waves. Every machine on it needs a decision before the migration starts rather than during it, and each one says why it did not fit: blocked findings, no measured throughput, or simply longer than any remaining window.

Plans are saved. Handoff usually happens on a different day than planning, and often by a different person — a plan that only existed in the tab that built it would not be one anybody could hand to whoever is working Saturday.