Work out nohup job duration instantly with clear inputs, formula shown and shareable results.
A long-running detached job faces a compounding risk of interruption: at 2 percent per hour, a 90-hour run has only about a 16 percent chance of finishing untouched. Checkpointing bounds the loss to half the checkpoint interval on average, which is what turns a long job from all-or-nothing into resumable work.
Long job survival
runtime = data / throughput; P(complete) = (1 - hourly failure rate)^runtime hours; expected lost work = checkpoint interval / 2
For a fire-and-forget job it works, but a systemd unit or a tmux session gives restart handling, logging and the ability to reattach. nohup only detaches from the terminal.
Frequently enough that redoing the work costs less than the checkpoint overhead. Balancing the two usually lands between five and thirty minutes.