A high load average, low “free” memory, or busy disk is a clue—not a root cause. Use interval-based Linux tools to separate CPU saturation, memory reclaim, swap, storage latency, network load, process behavior, and cgroup pressure before restarting or killing anything.
ncdu turns a recursive disk-usage scan into a fast terminal browser, but its delete key deserves the same respect as `rm`. Start read-only, stay on the intended filesystem, export remote scans for offline analysis, and confirm why `df`, `du`, and ncdu may report different numbers.
Use scp for a straightforward one-off copy, rsync over SSH for repeatable or resumable directory synchronization, SFTP for interactive/file-oriented workflows, and a tar stream for carefully controlled archive pipelines. The protocol encrypts transport; you still need host verification, least privilege, safe destination semantics, and end-to-end validation.