Performance diagnosis from the macOS Terminal: top, vm_stat, memory_pressure, fs_usage, and powermetrics

What it is

macOS ships a set of BSD-heritage and Apple-specific command-line tools that between them cover much of the ground top/vmstat/iostat/strace cover on Linux, with different names and output.

• top -o cpu sorts the periodically displayed process list by CPU instead of the default pid ordering, per its manual, which describes top as periodically displaying a sorted list of system processes.
• vm_stat [interval] displays Mach virtual memory statistics; with an interval it shows the change per period instead of only the totals since boot.
• memory_pressure is documented as a tool to apply real or simulate memory pressure on the system: -l and -p allocate real memory, -S simulates a pressure level. Run without arguments on current macOS it prints memory statistics and a system-wide free percentage instead; the manual does not document that mode, so treat its output format as unstable.
• iostat displays kernel I/O statistics on terminal, device and CPU operations, averaged over system uptime unless a repeat interval is given.
• fs_usage presents an ongoing display of system-call activity for filesystem operations; its manual states it requires root privileges because of the kernel tracing facility it uses.
• powermetrics gathers and displays CPU usage statistics split into user and supervisor time, along with interrupt and package-idle information. It must be run as root, samples every 5 seconds by default and runs until interrupted unless -n sets a sample count.
• sample <pid> [duration] suspends the target process at repeated short intervals, records every thread's call stack each time, then resumes it (default: every 1 ms for 10 seconds) — a lightweight, no-build-step alternative to attaching a full profiler. Another user's process needs sudo, and system processes protected by System Integrity Protection may refuse sampling even then.

Why it matters

None of these is a drop-in rename of its Linux counterpart: macOS iostat reports far fewer per-device fields than sysstat's Linux version, vm_stat's page-based units need multiplying by the page size printed in its header (16 KB on Apple silicon, 4 KB on Intel Macs) for bytes, and both fs_usage and powermetrics need elevated privileges because they read kernel tracing and power-management facilities that are access-controlled by design.

How to apply

• Start wide with top -o cpu (or -o mem) to find the process before narrowing.
• Use vm_stat 1 to watch page-in/page-out rates change live rather than reading one static snapshot.
• Run memory_pressure with no arguments first; use -l, -p or -S only deliberately, since applying or simulating pressure can degrade the system while active.
• Reach for sample <pid> 10 for a quick "what is this process doing" answer before setting up a heavier profiler.
• Run sudo fs_usage -w <process-name> narrowed to one process; unfiltered, system-wide output is very high volume.
• Bound powermetrics, for example sudo powermetrics -n 1 -i 1000, for a single one-second sample.

Pitfalls

• Running fs_usage or powermetrics without elevated privileges and concluding the tool found nothing, when it was simply denied access.
• Treating iostat's uptime-averaged first line as the current state instead of waiting for a second, interval-based line.
• Forgetting that sample briefly pauses the process at every sampling point, which is not appropriate for a process with hard real-time deadlines.