I am trying to familiarize myself with perf and run it against various programs I wrote.
When I launch it against program that is 100% single threaded, perf shows that it takes two cores on machine (task-clock event).
Here's the example output:
perf stat -a --per-core python3 test.py
Performance counter stats for 'system wide':
S0-C0 1 19004.951263 task-clock (msec) # 1.000 CPUs utilized (100.00%)
S0-C0 1 5,582 context-switches (100.00%)
S0-C0 1 19 cpu-migrations (100.00%)
S0-C0 1 3,746 page-faults
S0-C0 1 <not supported> cycles
S0-C0 1 <not supported> stalled-cycles-frontend
S0-C0 1 <not supported> stalled-cycles-backend
S0-C0 1 <not supported> instructions
S0-C0 1 <not supported> branches
S0-C0 1 <not supported> branch-misses
S0-C1 1 19004.950059 task-clock (msec) # 1.000 CPUs utilized (100.00%)
S0-C1 1 6,752 context-switches (100.00%)
S0-C1 1 25 cpu-migrations (100.00%)
S0-C1 1 935 page-faults
S0-C1 1 <not supported> cycles
S0-C1 1 <not supported> stalled-cycles-frontend
S0-C1 1 <not supported> stalled-cycles-backend
S0-C1 1 <not supported> instructions
S0-C1 1 <not supported> branches
S0-C1 1 <not supported> branch-misses
19.004688019 seconds time elapsed
It even shows that simple sleep command takes two cores on my computer and I can't explain this. I understand that OS scheduler can reassign active core for any process, but in this case CPU utilization would reflect that.
Can anyone explain this?
According to man page of perf stat subocmmand, you have -a option to profile full system:
http://man7.org/linux/man-pages/man1/perf-stat.1.html
-a, --all-cpus
system-wide collection from all CPUs (default if no target is
specified)
In this "system-wide" mode perf stat (and perf record too) will count events on (or profile for record) all CPUs in the system. When used without additional argument of command, perf will run until interrupted by Ctrl-C. With argument of command, perf will count/profile until the command works. Typical usage is
perf stat -a sleep 10 # Profile counting every CPU for 10 seconds
perf record -a sleep 10 # Profile with cycles every CPU for 10 seconds to perf.data
For getting stats of single command use single process profiling (without -a option)
perf stat python3 test.py
For profiling (perf record) you may run without -a option; or you may use -a and later do some manual filtering in perf report, focusing only on the pids/tids/dsos of your application (This can be very useful if command to profile uses some interprocess requests to other daemons to do lot of CPU work).
--per-core, -A, -C <cpulist>, --per-socket options are only for system-wide -a mode. Try --per-thread with -p pid attach to process option.
Related
When the perf stat command is used, many default events are measured. For example, when I run perf stat ls, I obtain the following output:
Performance counter stats for 'ls':
0,55 msec task-clock # 0,598 CPUs utilized
0 context-switches # 0,000 /sec
0 cpu-migrations # 0,000 /sec
99 page-faults # 179,071 K/sec
2 324 694 cycles # 4,205 GHz
1 851 372 instructions # 0,80 insn per cycle
357 918 branches # 647,403 M/sec
12 897 branch-misses # 3,60% of all branches
0,000923884 seconds time elapsed
0,000993000 seconds user
0,000000000 seconds sys
Now, let's suppose I also want to measure the cache-references and cache-misses events.
If I run perf stat -e cache-references,cache-misses, the output is:
Performance counter stats for 'ls':
101 148 cache-references
34 261 cache-misses # 33,872 % of all cache refs
0,000973384 seconds time elapsed
0,001014000 seconds user
0,000000000 seconds sys
Is there a way to add events with the -e flag, but also keep the default events shown when not using -e (without having to list all of them explicitly in the command) ?
I'm running the following command line using perf
$ perf stat ./program
[...]
<not supported> instructions
<not supported> branches
<not supported> branch-misses
<not supported> cycles
[...]
$ cat /proc/sys/kernel/perf_event_paranoid
3
How do I enable these events? Mainly instructions and cycles?
I already changed the value to -1 in the /proc/sys/kernel/perf_event_paranoid file, but it keeps the events off, when I reboot, it goes back to the original value which is 3.
I want to check how much resources took each command in my script
example code
#!/bin/bash
mkdir /tmp/{1..100}
touch /tmp/{1..100}
I want to know what resource was taken by mkdir and what by touch
You can use the command perf, and it's really complete.
Intall it with those 2 commands:
sudo apt install linux-tools-common
sudo apt install linux-tools-5.4.0-47-generic
test#test:~/work$ sudo perf stat ls
perf.data
Performance counter stats for 'ls':
0.83 msec task-clock # 0.793 CPUs utilized
2 context-switches # 0.002 M/sec
0 cpu-migrations # 0.000 K/sec
113 page-faults # 0.136 M/sec
<not supported> cycles
<not supported> instructions
<not supported> branches
<not supported> branch-misses
0.001051057 seconds time elapsed
0.000989000 seconds user
0.000000000 seconds sys
This my repository
test#test:~$ cat /etc/apt/sources.list
deb http://fr.archive.ubuntu.com/ubuntu focal-security universe
deb http://fr.archive.ubuntu.com/ubuntu focal-security multiverse
My distribution
test#test:~$ uname -a
Linux test 5.4.0-47-generic #51-Ubuntu SMP Fri Sep 4 19:50:52 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
I have a program which parses the output of the linux command perf. It requires the use of option -x, (the field separator option. I want to extract elapsed time (not task-time or cpu-clock) using perf. However when I use the -x option, the elapsed time is not present in the output and I cannot find a corresponding perf event. Here are the sample outputs
perf stat ls
============
Performance counter stats for 'ls':
0.934889 task-clock (msec) # 0.740 CPUs utilized
6 context-switches # 0.006 M/sec
0 cpu-migrations # 0.000 K/sec
261 page-faults # 0.279 M/sec
1,937,910 cycles # 2.073 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,616,944 instructions # 0.83 insns per cycle
317,016 branches # 339.095 M/sec
12,439 branch-misses # 3.92% of all branches
0.001262625 seconds time elapsed //here we have it
Now with field separator option
perf stat -x, ls
================
2.359807,task-clock
6,context-switches
0,cpu-migrations
261,page-faults
1863028,cycles
<not supported>,stalled-cycles-frontend
<not supported>,stalled-cycles-backend
1670644,instructions
325047,branches
12251,branch-misses
Any help is appreciated
# perf stat ls 2>&1 >/dev/null | tail -n 2 | sed 's/ \+//' | sed 's/ /,/'
0.002272536,seconds time elapsed
Starting with kernel 5.2-rc1, a new event called duration_time is exposed by perf statto solve exactly this problem. The value of this event is exactly equal to the time elapsed value, but the unit is nanoseconds instead of seconds.
I want to measure stalled cycles for my application using perf.
When I try: perf stat -B dd if=/dev/zero of=/dev/null count=1000000
1000000+0 records in
1000000+0 records out
512000000 bytes (512 MB) copied, 0.218456 s, 2.3 GB/s
Performance counter stats for 'dd if=/dev/zero of=/dev/null count=1000000':
218.420011 task-clock # 0.995 CPUs utilized
25 context-switches # 0.000 M/sec
1 CPU-migrations # 0.000 M/sec
255 page-faults # 0.001 M/sec
821,183,099 cycles # 3.760 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,526,427,190 instructions # 1.86 insns per cycle
292,281,624 branches # 1338.163 M/sec
1,013,837 branch-misses # 0.35% of all branches
0.219551862 seconds time elapsed
As you can see, I'm getting for stalled-cycles* events. I couldn't find a solution or explanation for this online.
My kernel version is 3.2.0-59, perf version is 3.2.54, and my CPU is an i7-3770.