viewing CPU usage by thread

Hello,

I'm looking to view a list of threads/processes and how much CPU usage they're each taking up. My screen saver has recently turned choppy for some strange reason and I'm wondering if a running thread hogging CPU time is the culprit. Similar to process list in task manager in windows.

How would I view this in linux? I guess the ps command, but I can't figure out how to format the output.

Thanks,
Jeff

If you're using kde, you can press control-escape which should start up ksysguard (or you can press alt-f2 and type in ksysguard). Gnome has a similar app called gnome-system-monitor.

If you're comfortable with the terminal, try using top. top is usually installed by default on most non-minimal distros. top refreshes the list of processes every few seconds, and sorts them by CPU usage. You can press h to get help on what the various keys do, H will show threads, or you can press q to quit back to the prompt.

There's a nicer version of top called htop, which I only discovered yesterday. It's slightly prettier and has a few extra options. It's top on steroids I guess.

ps can of course be used. Here's my attempt at a useful set of command line options:

ps axm -o pid,lwp,%cpu,time,command k -%cpu |head -n 20

The command line arguments to ps are a bit of a nightmare because it tries to be compatibile with several different standards and old implementations of ps in different unixes...
The options mean:

• axm : all process (ax), show thread details (m)
• -o pid,lwp,%cpu,time,command : show columns as specified.
• k -%cpu : sort by %cpu column, descending.[/i]

The output of ps is sent to head -n 20, which just passes through the first 20 lines, and junks the rest - just so you can see the top CPU users... you probably won't want to see the whole list of processes on the system.

Oh, by the way, LWP stands for "light weight process". That's a thread to you and I.

Must admit I prefer top. It accepts a configuration file that enables you to tailor it for a particular situation.
I run it in batch mode, and also use "head" to trim the output, then write it to a file. It'll take a delay and iteration count as well.

As usual, multiple solutions available to suit everybody.