path: root/src/benchmarks/dj_trace.py

import matplotlib.pyplot as plt
import numpy as np
import os
from urllib.request import urlretrieve
import sys
import re

from src.benchmark import Benchmark
from src.util import print_status

comma_sep_number_re = "(?:\d*(?:,\d*)?)*"
rss_re = "(?P<rss>" + comma_sep_number_re + ")"
time_re = "(?P<time>" + comma_sep_number_re + ")"

cycles_re = re.compile("^{} cycles$".format(time_re))
cpu_time_re = re.compile("^{} usec across.*threads$".format(time_re))

max_rss_re = re.compile("^{} Kb Max RSS".format(rss_re))
ideal_rss_re = re.compile("^{} Kb Max Ideal RSS".format(rss_re))

malloc_re = re.compile("^Avg malloc time:\s*{} in.*calls$".format(time_re))
calloc_re = re.compile("^Avg calloc time:\s*{} in.*calls$".format(time_re))
realloc_re = re.compile("^Avg realloc time:\s*{} in.*calls$".format(time_re))
free_re = re.compile("^Avg free time:\s*{} in.*calls$".format(time_re))


class Benchmark_DJ_Trace(Benchmark):
    def __init__(self):
        self.name = "dj_trace"
        self.descrition = """This benchmark uses the workload simulator written
                             by DJ Delorie to simulate workloads provided by
                             him under https://delorie.com/malloc. Those
                             workloads are generated from traces of real
                             aplications and are also used by delorie to
                             measure improvements in the glibc allocator."""

        self.cmd = "trace_run{binary_suffix} dj_workloads/{workload}.wl"
        self.measure_cmd = ""

        self.args = {
                        "workload": [
                                        "389-ds-2",
                                        "dj",
                                        "dj2",
                                        "mt_test_one_alloc",
                                        "oocalc",
                                        "qemu-virtio",
                                        "qemu-win7",
                                        "proprietary-1",
                                        "proprietary-2",
                                      ]
                    }
        self.results = {
                        "389-ds-2": {
                            "malloc": 170500018, "calloc": 161787184,
                            "realloc": 404134, "free": 314856324,
                            "threads": 41},
                        "dj": {
                            "malloc": 2000000, "calloc": 200, "realloc": 0,
                            "free": 2003140, "threads": 201},
                        "dj2": {
                            "malloc": 29263321, "calloc": 3798404,
                            "realloc": 122956, "free": 32709054,
                            "threads": 36},
                        "mt_test_one_alloc": {
                            "malloc": 524290, "calloc": 1, "realloc": 0,
                            "free": 594788, "threads": 2},
                        "oocalc": {
                            "malloc": 6731734, "calloc": 38421,
                            "realloc": 14108, "free": 6826686, "threads": 88},
                        "qemu-virtio": {
                            "malloc": 1772163, "calloc": 146634,
                            "realloc": 59813, "free": 1954732, "threads": 3},
                        "qemu-win7": {
                            "malloc": 980904, "calloc": 225420,
                            "realloc": 89880, "free": 1347825, "threads": 6},
                        "proprietary-1": {
                            "malloc": 316032131, "calloc": 5642, "realloc": 84,
                            "free": 319919727, "threads": 20},
                        "proprietary-2": {
                            "malloc": 9753948, "calloc": 4693,
                            "realloc": 117, "free": 10099261, "threads": 19},
                        }

        self.requirements = ["trace_run"]
        super().__init__()

    def prepare(self):
        super().prepare()

        def reporthook(blocknum, blocksize, totalsize):
            readsofar = blocknum * blocksize
            if totalsize > 0:
                percent = readsofar * 1e2 / totalsize
                s = "\r%5.1f%% %*d / %d" % (
                    percent, len(str(totalsize)), readsofar, totalsize)
                sys.stderr.write(s)
            else:  # total size is unknown
                sys.stderr.write("\rdownloaded %d" % (readsofar,))

        if not os.path.isdir("dj_workloads"):
            os.mkdir("dj_workloads")

        download_all = None
        wl_sizes = {"dj": "14M", "oocalc": "65M", "mt_test_one_alloc": "5.7M",
                    "proprietary-1": "2.8G", "qemu-virtio": "34M",
                    "proprietary-2": "92M", "qemu-win7": "23M",
                    "389-ds-2": "3.4G", "dj2": "294M"}

        for wl in self.args["workload"]:
            file_name = wl + ".wl"
            file_path = os.path.join("dj_workloads", file_name)
            if not os.path.isfile(file_path):
                if download_all == None:
                    choice = input(("Download all missing workloads (upto 6.7GB)"
                                          " [Y/n/x] "))
                    if choice == "x":
                        break
                    else:
                        download_all = choice in ['', 'Y', 'y']

                if (not download_all and
                    input("want to download {} ({}) [Y/n] ".format(wl, wl_sizes[wl])) not in ['', 'Y', 'y']):
                    continue

                if download_all:
                    print_status("downloading {} ({}) ...".format(wl, wl_sizes[wl]))

                url = "http://www.delorie.com/malloc/" + file_name
                urlretrieve(url, file_path, reporthook)
                sys.stderr.write("\n")

        available_workloads = []
        for wl in self.args["workload"]:
            file_name = wl + ".wl"
            file_path = os.path.join("dj_workloads", file_name)
            if os.path.isfile(file_path):
                available_workloads.append(wl)

        if len(available_workloads) > 0:
            self.args["workload"] = available_workloads
            return True
        
        return False

    def process_output(self, result, stdout, stderr, allocator, perm, verbose):
        def to_int(s):
            return int(s.replace(',', ""))

        regexs = {7: malloc_re, 8: calloc_re, 9: realloc_re, 10: free_re}
        functions = {7: "malloc", 8: "calloc", 9: "realloc", 10: "free"}
        for i, l in enumerate(stdout.splitlines()):
            if i == 0:
                result["cycles"] = to_int(cycles_re.match(l).group("time"))
            elif i == 2:
                result["cputime"] = to_int(cpu_time_re.match(l).group("time"))
            elif i == 3:
                result["Max_RSS"] = to_int(max_rss_re.match(l).group("rss"))
            elif i == 4:
                result["Ideal_RSS"] = to_int(ideal_rss_re.match(l).group("rss"))
            elif i in [7, 8, 9, 10]:
                res = regexs[i].match(l)
                fname = functions[i]
                result["avg_" + fname] = to_int(res.group("time"))

    def summary(self):
        args = self.results["args"]
        allocators = self.results["allocators"]

        cpu_time_means = {allocator: {} for allocator in allocators}
        cycles_means = {allocator: {} for allocator in allocators}
        for perm in self.iterate_args(args=args):
            for i, allocator in enumerate(allocators):
                d = [x["cputime"] for x in self.results[allocator][perm]]
                # data is in milliseconds
                cpu_time_means[allocator][perm] = int(np.mean(d)/1000)

                d = [x["cycles"] for x in self.results[allocator][perm]]
                cycles_means[allocator][perm] = int(np.mean(d))

                plt.bar([i], cpu_time_means[allocator][perm], label=allocator,
                        color=allocators[allocator]["color"])

            plt.legend(loc="best")
            plt.ylabel("Zeit in ms")
            plt.title("Gesamte Laufzeit")
            plt.savefig(".".join([self.name, perm.workload, "runtime", "png"]))
            plt.clf()


        self.barplot_single_arg("{cputime}/1000",
                                ylabel='"time in ms"',
                                title='"total runtime"',
                                filepostfix="runtime")

        # Function Times
        func_times_means = {allocator: {} for allocator in allocators}
        xa = np.arange(0, 6, 1.5)
        for perm in self.iterate_args(args=args):
            for i, allocator in enumerate(allocators):
                x_vals = [x+i/len(allocators) for x in xa]

                func_times_means[allocator][perm] = [0,0,0,0]

                func_times_means[allocator][perm][0] = np.mean([x["avg_malloc"] for x in self.results[allocator][perm]])
                func_times_means[allocator][perm][1] = np.mean([x["avg_calloc"] for x in self.results[allocator][perm]])
                func_times_means[allocator][perm][2] = np.mean([x["avg_realloc"] for x in self.results[allocator][perm]])
                func_times_means[allocator][perm][3] = np.mean([x["avg_free"] for x in self.results[allocator][perm]])

                plt.bar(x_vals, func_times_means[allocator][perm], width=0.25,
                        align="center", label=allocator,
                        color=allocators[allocator]["color"])

            plt.legend(loc="best")
            plt.xticks(xa + 1/len(allocators)*2,
                       ["malloc\n" + str(self.results[perm.workload]["malloc"]) + "\ncalls",
                        "calloc\n" + str(self.results[perm.workload]["calloc"]) + "\ncalls",
                        "realloc\n" + str(self.results[perm.workload]["realloc"]) + "\ncalls",
                        "free\n" + str(self.results[perm.workload]["free"]) + "\ncalls"])
            plt.ylabel("Durchschnittliche Zeit in cycles")
            plt.title("Durchscnittliche Laufzeiten der API Funktionen")
            plt.savefig(".".join([self.name, perm.workload, "apitimes", "png"]))
            plt.clf()

        # Memusage
        # hack ideal rss in data set
        allocators["Ideal_RSS"] = {"color": "C" + str(len(allocators) + 1)}
        self.results["stats"]["Ideal_RSS"] = {}
        for perm in self.iterate_args(args=args):
            ideal_rss = self.results[list(allocators.keys())[0]][perm][0]["Ideal_RSS"]/1000
            self.results["stats"]["Ideal_RSS"][perm] = {"mean": {"Max_RSS": ideal_rss}}


        self.barplot_single_arg("{Max_RSS}/1000",
                                ylabel='"Max RSS in MB"',
                                title='"Highwatermark of Vm (VmHWM)"',
                                filepostfix="newrss")

        del(allocators["Ideal_RSS"])
        del(self.results["stats"]["Ideal_RSS"])

        rss_means = {allocator: {} for allocator in allocators}
        for perm in self.iterate_args(args=args):
            for i, allocator in enumerate(allocators):
                d = [x["Max_RSS"] for x in self.results[allocator][perm]]
                # data is in kB
                rss_means[allocator][perm] = np.mean(d)/1000

                plt.bar([i], rss_means[allocator][perm], label=allocator,
                        color=allocators[allocator]["color"])

            # add ideal rss
            y_val = self.results[list(allocators.keys())[0]][perm][0]["Ideal_RSS"]/1000
            plt.bar([len(allocators)], y_val, label="Ideal RSS")

            plt.legend(loc="best")
            plt.ylabel("Max RSS in MB")
            plt.title("Maximal benötigter Speicher (VmHWM)")
            plt.savefig(".".join([self.name, perm.workload, "rss", "png"]))
            plt.clf()

        self.export_to_csv("Max_RSS")
        self.export_to_csv("cputime")

        self.export_to_dataref("Max_RSS")
        self.export_to_dataref("cputime")

        # Tables
        for perm in self.iterate_args(args=args):
            # collect data
            d = {allocator: {} for allocator in allocators}
            for i, allocator in enumerate(allocators):
                d[allocator]["time"] = [x["cputime"] for x in self.results[allocator][perm]]
                d[allocator]["rss"] = [x["Max_RSS"] for x in self.results[allocator][perm]]

            times = {allocator: np.mean(d[allocator]["time"]) for allocator in allocators}
            tmin = min(times)
            tmax = max(times)

            rss = {allocator: np.mean(d[allocator]["rss"]) for allocator in allocators}
            rssmin = min(rss)
            rssmax = max(rss)

            fname = ".".join([self.name, perm.workload, "table.tex"])
            with open(fname, "w") as f:
                print("\\documentclass{standalone}", file=f)
                print("\\usepackage{xcolor}", file=f)
                print("\\begin{document}", file=f)
                print("\\begin{tabular}{| l | l | l |}", file=f)
                print("& Zeit (ms) / $\\sigma$ (\\%) & VmHWM (KB) / $\\sigma$ (\\%) \\\\", file=f)
                print("\\hline", file=f)

                for allocator in allocators:
                    print(allocator.replace("_", "\\_"), end=" & ", file=f)

                    s = "\\textcolor{{{}}}{{{}}} / {}"

                    t = d[allocator]["time"]
                    m = times[allocator]
                    if m == tmin:
                        color = "green"
                    elif m == tmax:
                        color = "red"
                    else:
                        color = "black"
                    print(s.format(color, m, np.std(t)/m), end=" & ", file=f)

                    t = d[allocator]["rss"]
                    m = rss[allocator]
                    if m == rssmin:
                        color = "green"
                    elif m == rssmax:
                        color = "red"
                    else:
                        color = "black"
                    print(s.format(color, m, np.std(t)/m if m else 0), "\\\\", file=f)

                print("\end{tabular}", file=f)
                print("\\end{document}", file=f)

        # Create summary similar to DJ's at
        # https://sourceware.org/ml/libc-alpha/2017-01/msg00452.html
        with open(self.name + "_plain.txt", "w") as f:
            # Absolutes
            fmt = "{:<20} {:>15} {:>7} {:>7} {:>7} {:>7} {:>7}"
            for i, allocator in enumerate(allocators):
                print("{0} {1} {0}".format("-" * 10, allocator), file=f)
                print(fmt.format("Workload", "Total", "malloc", "calloc",
                                 "realloc", "free", "RSS"), file=f)

                for perm in self.iterate_args(args=args):
                    cycles = cycles_means[allocator][perm]
                    times = [int(t) for t in func_times_means[allocator][perm]]
                    rss = int(rss_means[allocator][perm])
                    print(fmt.format(perm.workload, cycles, times[0], times[1],
                                     times[2], times[3], rss), file=f)

                print(file=f)

            # Changes. First allocator in allocators is the reference
            fmt_changes = "{:<20} {:>14.0f}% {:>6.0f}% {:>6.0f}% {:>6.0f}% {:>6.0f}% {:>6.0f}%"
            for i, allocator in enumerate(list(allocators)[1:]):
                print("{0} Changes {1} {0}".format("-" * 10, allocator), file=f)
                print(fmt.format("Workload", "Total", "malloc", "calloc",
                                 "realloc", "free", "RSS"), file=f)

                ref_alloc = list(allocators)[0]
                cycles_change_means = []
                times_change_means = []
                rss_change_means = []
                for perm in self.iterate_args(args=args):

                    normal_cycles = cycles_means[ref_alloc][perm]
                    if normal_cycles:
                        cycles = np.round(cycles_means[allocator][perm] / normal_cycles * 100)
                    else:
                        cycles = 0
                    cycles_change_means.append(cycles)

                    normal_times = func_times_means[ref_alloc][perm]
                    times = [0, 0, 0, 0]
                    for i in range(0, len(times)):
                        t = func_times_means[allocator][perm][i]
                        nt = normal_times[i]
                        if nt != 0:
                            times[i] = np.round(t/nt * 100)
                    times_change_means.append(times)

                    normal_rss = rss_means[ref_alloc][perm]
                    if normal_rss:
                        rss = np.round(rss_means[allocator][perm] / normal_rss * 100)
                    else:
                        rss = 0
                    rss_change_means.append(rss)

                    print(fmt_changes.format(perm.workload, cycles, times[0],
                                             times[1], times[2], times[3], rss),
                          file=f)
                print(file=f)
                tmeans = [0,0,0,0]
                for i in range(0, len(times)):
                    tmeans[i] = np.mean([times[i] for times in times_change_means])
                print(fmt_changes.format("Mean:", np.mean(cycles_change_means),
                                         tmeans[0], tmeans[1], tmeans[2],
                                         tmeans[3], np.mean(rss_change_means)),
                      '\n', file=f)


dj_trace = Benchmark_DJ_Trace()