One-stop performance analysis using atop

Posted May 20, 2010 20:52 UTC (Thu) by oak (guest, #2786)
Parent article: One-stop performance analysis using atop

htop is also pretty nice, easily configurable and although ncurses based, supports mouse & colors. It shows only the current state, not history (and no disk or network usage), but from TIME+ column one can check whether process CPU ticks value increase.

Latest version allows stracing selected process (it should use "-f" arg for strace though, so that all threads are tracked) or using lsof.

The problem with these fancier top programs (e.g. compared to simple Busybox top) is that they take more CPU and are slower to start. If your system is sometimes _really_ slow, but only temporarily, it matters a lot how fast "top" starts and can update its screen.

to post comments

One-stop performance analysis using atop

Posted Jun 14, 2010 0:47 UTC (Mon) by Tobu (subscriber, #24111) [Link]

The problem with these fancier top programs (e.g. compared to simple Busybox top) is that they take more CPU and are slower to start. If your system is sometimes _really_ slow, but only temporarily, it matters a lot how fast "top" starts and can update its screen.

The nice thing with atop is that it has per-process, historical info. Assuming you use the debian package which sets up cronjobs and logrotate, you can run sudo atop -r, use t and T to navigate the day in 10s slices, find the time the system slowed down, and pinpoint the problem process and the scarce resource. Thanks to process accounting, data is logged reliably under load.