24

I want to see if my process makes a lot of context switches. I also want to see how manpulating task groups affects the number of context switches.

  • What OS? What version? What kernel? – Mikel May 25 '12 at 16:48
  • GNU/Linux 2.6.18 – luntain May 25 '12 at 17:28
  • 5
    Check /proc/[pid]/status – Kevin May 25 '12 at 17:34
30

You can view information about your process's context switches in /proc/<pid>/status.

$ pid=307
$ grep ctxt /proc/$pid/status
voluntary_ctxt_switches:        41
nonvoluntary_ctxt_switches:     16

To see these numbers updating continuously, run

$ # Update twice a second.
$ watch -n.5 grep ctxt /proc/$pid/status

To get just the numbers, run

$ grep ctxt /proc/$pid/status | awk '{ print $2 }'
  • I tried the above watch command but the output is just 1 (for voluntary and involuntary context switch). My linux kernal version is 2.6.39-400.214.4.el6uek.x86_64. Is the output dependant on the linux version? – Andy Dufresne Mar 19 '15 at 6:49
  • Paste your output here please. – user26112 Mar 20 '15 at 19:05
  • Its the same as mentioned in the answere - "watch -n.5 grep ctxt /proc/5647/status" – Andy Dufresne Mar 21 '15 at 10:52
  • I have to use procfs but on old kernels there aren't the values inside /proc/.../status. any suggestion ? – Massimo Aug 31 '18 at 17:49
12

pidstat(1) - Report statistics for Linux tasks. According to man pidstat it's so easy as just pidstat -w …

  • I executed "watch -n0.5 pidstat -w -I -p 5876" command but the output is 0 (for both cswch/s nvcswch/s). Does this command work for linux version - 2.6.39-400.214.4.el6uek.x86_64 ? – Andy Dufresne Mar 19 '15 at 6:52
  • This command should work just fine. But beware you're using it wrong because when you don't specify report interval „tasks statistics are to be reported for the time since system startup (boot).” similar to vmstat, iostat and others. So if current statistics is needed instead of watch'ing simply run it with one second interval. – poige Mar 19 '15 at 12:46
  • If I don't do watch, how do I see the numbers updating continuously? Executing the command "pidstat -w -I -p 5876 5" the command just waits for 5 seconds and then prints the output (again as 0). It does not run continuously as I was expecting (I know that this is contradicting to what pidstat's man page says - linux.die.net/man/1/pidstat). My OS is Oracle Linux Server 6.4. – Andy Dufresne Mar 21 '15 at 10:57
  • Does ` pidstat -w -l -p SELF 1` work for you? – poige Mar 21 '15 at 16:11
4

To get a record of an entire process run, you can use the GNU time utility (don't confuse it with the bash builtin) with the -v option. Here's an example with unrelated lines of output removed:

$ `which time` -v ls
a.out  exception_finder.cpp  log.txt
    Command being timed: "ls"
               ...
    Voluntary context switches: 1
    Involuntary context switches: 2
               ...
    Exit status: 0
3

You can use, sar -w. For instance, sar -w 1 3, reports total number of context switches per second for every 1 seconds a total of 3 times.

  • 1
    That doesn't work "out of the box" on many systems, even if the command is available. Can you include in your answer how to enable data collecting for sar? – Anthon May 2 '16 at 5:40
2

Write the following script to file (ctx.sh). with ctx.sh <core> you will see all the processes running on a given core and changing nv-context switches will be highlighted. Looking at this, you will be able to identify which are the competing processes for the core.

#!/bin/bash

if [[ $# -eq 0 ]]
then
   echo "Usage:"
   echo "$0 <core>"
   exit 1
fi

if [[ -z $2 ]]
then
   watch -d -n .2 $0 $1 nw
fi

ps -Leo lastcpu:1,tid,comm | grep "^$1 " | awk '{printf $3": ";system("cut -d\" \" -f3  /proc/"$2"/task/"$2"/schedstat 2>/dev/null")}' | sort -k 1 | column -t
1

See man getrusage which will let you query the number of voluntary and involuntary context switches.

struct rusage {
           struct timeval ru_utime; /* user CPU time used */
           struct timeval ru_stime; /* system CPU time used */
           long   ru_maxrss;        /* maximum resident set size */
           long   ru_ixrss;         /* integral shared memory size */
           long   ru_idrss;         /* integral unshared data size */
           long   ru_isrss;         /* integral unshared stack size */
           long   ru_minflt;        /* page reclaims (soft page faults) */
           long   ru_majflt;        /* page faults (hard page faults) */
           long   ru_nswap;         /* swaps */
           long   ru_inblock;       /* block input operations */
           long   ru_oublock;       /* block output operations */
           long   ru_msgsnd;        /* IPC messages sent */
           long   ru_msgrcv;        /* IPC messages received */
           long   ru_nsignals;      /* signals received */
           long   ru_nvcsw;         /* voluntary context switches */
           long   ru_nivcsw;        /* involuntary context switches */
};

You can tell it to report per-thread information, like this:

struct rusage usage;
getrusage( RUSAGE_THREAD, &usage );

Just call it twice, before and after your critical section, and see if usage.ru_nivcsw value has increased or not.

0
sudo perf stat -e context-switches -I 1000 PROCESS_NAME

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.