Date:	Wed, 13 May 2015 21:05:22 +0900
From:	Namhyung Kim <namhyung@...nel.org>
To:	Andi Kleen <andi@...stfloor.org>
Cc:	Arnaldo Carvalho de Melo <acme@...nel.org>,
	Ingo Molnar <mingo@...nel.org>,
	Peter Zijlstra <a.p.zijlstra@...llo.nl>,
	Jiri Olsa <jolsa@...hat.com>,
	LKML <linux-kernel@...r.kernel.org>,
	David Ahern <dsahern@...il.com>,
	Stephane Eranian <eranian@...gle.com>,
	Minchan Kim <minchan@...nel.org>
Subject: Re: [RFC/PATCH v2] perf data: Add stat subcommand to show sample
 event stat

Hi Andi,

On Mon, May 11, 2015 at 05:44:05PM +0200, Andi Kleen wrote:
> > The sampling ratio was useful for me to determine how often the event
> > was sampled - in this case the cpu cycles event was only sampled at 12%
> 
> That's dangerous to determine without a plot. It could be that it was bimodal: 
> 100% busy and then idle. You may want to add something like the spark
> plots I submitted for stat some time ago.

Right, we cannot know the exact situation from a single number.  But
it was okay for me just to see overall status from the number.  This
is what we cannot know from the output of 'perf report' easily, so I'd
like to add this info in some way.

Anyway, I wrote a script to plot the number of samples and periods
using python's matplotlib package.  Maybe we can add it to the script
database.

Thanks,
Namhyung


/* sample-chart.py */
import os
import sys
import numpy as np
import matplotlib.pyplot as plt

sys.path.append(os.environ['PERF_EXEC_PATH'] + \
        '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')

from perf_trace_context import *
from EventClass import *

events = {}
mode = None  # 'cpu' or 'task'
nr_events = 0
first_time = 0
last_time = 0

def trace_begin():
    pass

def trace_end():
    xcnt = last_time - first_time + 1
    xpos = np.arange(xcnt)
    times = np.arange(first_time, last_time + 1)

    fig, plt_array = plt.subplots(nrows = nr_events, ncols = 2)
    fig.suptitle("Event stat", fontsize=20)

    n = 0
    for e in events:
        p1 = plt_array[n][0]
        p2 = plt_array[n][1]
        for k in events[e]:  # key = cpu or tid
            ev_stats = events[e][k]

            samples = np.zeros(xcnt)
            periods = np.zeros(xcnt)
            for t in ev_stats:
                samples[t - first_time] = ev_stats[t][0]
                periods[t - first_time] = ev_stats[t][1]

            key = "%s %d" % (mode, k)

            p1.plot(times, samples, 'o', linewidth=2, label=key)
            p2.plot(times, periods, '-', linewidth=2, label=key)

        expect = 400 * np.ones(xcnt)
        p1.plot(times, expect, '--')

        p1.set_title("Number of samples in '%s'" % e)
        p1.legend()
        p2.set_title("Event values in '%s'" % e)
        p2.legend()
        n += 1

    plt.show()

def process_event(param_dict):
    evt  = param_dict["ev_name"]
    cpu  = param_dict["sample"]["cpu"]
    tid  = param_dict["sample"]["tid"]
    time = param_dict["sample"]["time"] / 100000000  # 100 ms
    val  = param_dict["sample"]["period"]

    if evt not in events:
        global nr_events
        nr_events += 1
        events[evt] = {}

    global mode
    if mode is None:
        if cpu >= 10000000:
            mode = 'task'
        else:
            mode = 'cpu'

    key = cpu if mode == 'cpu' else tid
    if key not in events[evt]:
        events[evt][key] = {}

    global first_time, last_time
    if first_time == 0 or first_time > time:
        first_time = time
    if last_time < time:
        last_time = time

    ev_stat = events[evt][key]
    if time not in ev_stat:
        ev_stat[time] = [0, 0]  # (nr_sample, period)
    ev_stat[time][0] += 1
    ev_stat[time][1] += val

def trace_unhandled(event_name, context, event_fields_dict):
    pass
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists