02:52 <adgud> what is the best approach to get values of cpu in sy, us, wa and id for a period of time?

02:53 <adgud> i'm thinking of using probe scheduler but also found probe timer and count user/kernel ticks

02:58 <fche> hi adgud - not sure which is 'best' - is there a problem in trying both?

03:02 <adgud> hello fche - by best i mean the easiest while being accurate - i'm struggling to replicate vmstat-like output using systemtap and i have trouble adapting graphts.stp sample script to split cpu utilisation to kernel and user time

03:04 <fche> might look at pf2.stp -- probe timer.profile { if (user_mode()) ... }

03:05 <fche> so estimate the values, per-cpu, on a sampling basis

03:06 <adgud> but then, using timer.profile, I don't know the idle time during sample period :(

03:08 <fche> yeah, it'd be lumped into 'system'

03:10 <fche> hm but the idle threads have some identifying characteristics that may help, pid=0?

03:10 <adgud> just what i was about to ask - which process is the idle process :)

03:10 <fche> not sure it's that simple, but worth testing

03:14 <adgud> testing this right now

03:29 <adgud> hmm with such approach (http://pastebin.com/raw/YzeRwLvP) and running some load, i've got 35% system time, while vmstat shows 25%

03:33 <adgud> do you have any idea why, fche?

03:34 <adgud> i should probably account for io wait time as well...

18:38 <adgud> can anyone here help me out with obtaining cpu profiling data? detailed question is here: http://stackoverflow.com/questions/42052872/calculating-cpu-utilization-with-systemtap

18:40 <fche> adgud, will have to think about it

18:41 <fche> one big difference is that this is sampling based rather than actual accounting of the sort the kernel does at transitions

18:41 <adgud> fche I would be more than glad if anyone could just point me in the right direction

18:41 <fche> one thing I'd do is reverse-engineer a bit about how vmstat etc. get their own percentages out of procfs

18:42 <fche> and trace back to the kernel-side functions or variables where they get that stuff from

18:42 <fche> then backward as to how they are counted

18:44 <adgud> that's what I thought of as well, but I doubt I have enough knowledge of C programming and kernel internals to do it in a reasonable time

19:08 <fche> so maybe just go with what you have for now and make progress on the rest of your script. Was this only a part of what you were looking for?

19:13 <irker442> systemtap: fche systemtap.git:refs/heads/master * release-3.0-905-gd1e8b63 / runtime/vsprintf.c: runtime: suppress a gcc7 warning http://tinyurl.com/z85fbwu

20:12 <adgud> no fche, i think it is basically it - get cpu utilization

20:13 <adgud> i miss the %id component but well maybe i will find it along