commit be48ab4d33119c4b9879f8ba89919601a1185ff5 Author: Karsten Loesing <karsten.loesing@gmx.net> Date: Fri Mar 20 23:26:02 2020 +0100 Rewrite tgen plots to use pandas and seaborn. Previous OnionPerf visualizations were written using PyLab, which is a relatively low-level visualization library based on matplotlib. PyLab is the same thing as PyPlot but with some extra shortcuts. However, using PyLab is discouraged now. This commit switches all tgen plots to use seaborn, which is a high-level data visualization library also based on matplotlib. This commit also uses pandas for underlying data structures in order to separate data tidying and data visualization. As a side effect of using pandas, graphed data is now exported to a .csv file using the same naming scheme as the produced tgen .pdf file. Implements #33258. --- onionperf/analysis.py | 6 + onionperf/visualization.py | 572 ++++++++++++++------------------------------- 2 files changed, 184 insertions(+), 394 deletions(-) diff --git a/onionperf/analysis.py b/onionperf/analysis.py index a380027..b79eac1 100644 --- a/onionperf/analysis.py +++ b/onionperf/analysis.py @@ -58,6 +58,12 @@ class Analysis(object): except: return None + def get_tgen_transfers(self, node): + try: + return self.json_db['data'][node]['tgen']['transfers'] + except: + return None + def get_tgen_transfers_summary(self, node): try: return self.json_db['data'][node]['tgen']['transfers_summary'] diff --git a/onionperf/visualization.py b/onionperf/visualization.py index a5dde54..51ea59e 100644 --- a/onionperf/visualization.py +++ b/onionperf/visualization.py @@ -8,6 +8,10 @@ import matplotlib; matplotlib.use('Agg') # for systems without X11 from matplotlib.backends.backend_pdf import PdfPages import pylab, numpy, time from abc import abstractmethod, ABCMeta +import matplotlib.pyplot as plt +import pandas as pd +import seaborn as sns +import datetime ''' pylab.rcParams.update({ @@ -140,403 +144,183 @@ class TGenVisualization(Visualization): if len(self.datasets) > 0: prefix = output_prefix + '.' if output_prefix is not None else '' ts = time.strftime("%Y-%m-%d_%H:%M:%S") + self.__extract_data_frame() + self.data.to_csv("{0}tgen.onionperf.viz.{1}.csv".format(prefix, ts)) + sns.set_context("paper") self.page = PdfPages("{0}tgen.onionperf.viz.{1}.pdf".format(prefix, ts)) - self.__plot_firstbyte() - self.__plot_byte_timeseries("time_to_first_byte") - self.__plot_lastbyte_all() - self.__plot_lastbyte_median() - self.__plot_lastbyte_mean() - self.__plot_lastbyte_max() - self.__plot_byte_timeseries("time_to_last_byte") - self.__plot_downloads() - self.__plot_downloads_timeseries() - self.__plot_errors() - self.__plot_errors_timeseries() - self.__plot_errsizes_all() - self.__plot_errsizes_median() - self.__plot_errsizes_mean() + self.__plot_firstbyte_ecdf() + self.__plot_firstbyte_time() + self.__plot_lastbyte_ecdf() + self.__plot_lastbyte_box() + self.__plot_lastbyte_bar() + self.__plot_lastbyte_time() + self.__plot_downloads_count() + self.__plot_errors_count() + self.__plot_errors_time() self.page.close() - def __plot_firstbyte(self): - f = None - - for (anal, label, lineformat) in self.datasets: - fb = [] - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "time_to_first_byte" in d: - for b in d["time_to_first_byte"]: - if f is None: f = pylab.figure() - for sec in d["time_to_first_byte"][b]: fb.extend(d["time_to_first_byte"][b][sec]) - if f is not None and len(fb) > 0: - x, y = getcdf(fb) - pylab.plot(x, y, lineformat, label=label) - - if f is not None: - pylab.xlabel("Download Time (s)") - pylab.ylabel("Cumulative Fraction") - pylab.title("time to download first byte, all clients") - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_lastbyte_all(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - lb = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "time_to_last_byte" in d: - for b in d["time_to_last_byte"]: - bytes = int(b) - if bytes not in figs: figs[bytes] = pylab.figure() - if bytes not in lb: lb[bytes] = [] - for sec in d["time_to_last_byte"][b]: lb[bytes].extend(d["time_to_last_byte"][b][sec]) - for bytes in lb: - x, y = getcdf(lb[bytes]) - pylab.figure(figs[bytes].number) - pylab.plot(x, y, lineformat, label=label) - - for bytes in sorted(figs.keys()): - pylab.figure(figs[bytes].number) - pylab.xlabel("Download Time (s)") - pylab.ylabel("Cumulative Fraction") - pylab.title("time to download {0} bytes, all downloads".format(bytes)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_lastbyte_median(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - lb = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "time_to_last_byte" in d: - for b in d["time_to_last_byte"]: - bytes = int(b) - if bytes not in figs: figs[bytes] = pylab.figure() - if bytes not in lb: lb[bytes] = [] - client_lb_list = [] - for sec in d["time_to_last_byte"][b]: client_lb_list.extend(d["time_to_last_byte"][b][sec]) - lb[bytes].append(numpy.median(client_lb_list)) - for bytes in lb: - x, y = getcdf(lb[bytes]) - pylab.figure(figs[bytes].number) - pylab.plot(x, y, lineformat, label=label) - - for bytes in sorted(figs.keys()): - pylab.figure(figs[bytes].number) - pylab.xlabel("Download Time (s)") - pylab.ylabel("Cumulative Fraction") - pylab.title("median time to download {0} bytes, each client".format(bytes)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_lastbyte_mean(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - lb = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "time_to_last_byte" in d: - for b in d["time_to_last_byte"]: - bytes = int(b) - if bytes not in figs: figs[bytes] = pylab.figure() - if bytes not in lb: lb[bytes] = [] - client_lb_list = [] - for sec in d["time_to_last_byte"][b]: client_lb_list.extend(d["time_to_last_byte"][b][sec]) - lb[bytes].append(numpy.mean(client_lb_list)) - for bytes in lb: - x, y = getcdf(lb[bytes]) - pylab.figure(figs[bytes].number) - pylab.plot(x, y, lineformat, label=label) - - for bytes in sorted(figs.keys()): - pylab.figure(figs[bytes].number) - pylab.xlabel("Download Time (s)") - pylab.ylabel("Cumulative Fraction") - pylab.title("mean time to download {0} bytes, each client".format(bytes)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_lastbyte_max(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - lb = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "time_to_last_byte" in d: - for b in d["time_to_last_byte"]: - bytes = int(b) - if bytes not in figs: figs[bytes] = pylab.figure() - if bytes not in lb: lb[bytes] = [] - client_lb_list = [] - for sec in d["time_to_last_byte"][b]: client_lb_list.extend(d["time_to_last_byte"][b][sec]) - lb[bytes].append(numpy.max(client_lb_list)) - for bytes in lb: - x, y = getcdf(lb[bytes]) - pylab.figure(figs[bytes].number) - pylab.plot(x, y, lineformat, label=label) - - for bytes in sorted(figs.keys()): - pylab.figure(figs[bytes].number) - pylab.xlabel("Download Time (s)") - pylab.ylabel("Cumulative Fraction") - pylab.title("max time to download {0} bytes, each client".format(bytes)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_byte_timeseries(self, bytekey="time_to_last_byte"): - figs = {} - - for (anal, label, lineformat) in self.datasets: - lb = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if bytekey in d: - for b in d[bytekey]: - bytes = int(b) - if bytes not in figs: figs[bytes] = pylab.figure() - if bytes not in lb: lb[bytes] = {} - for sec in d[bytekey][b]: - if sec not in lb[bytes]: lb[bytes][sec] = [] - lb[bytes][sec].extend(d[bytekey][b][sec]) - for bytes in lb: - pylab.figure(figs[bytes].number) - x = [sec for sec in lb[bytes]] - x.sort() - y = [numpy.mean(lb[bytes][sec]) for sec in x] - pylab.plot(x, y, lineformat, label=label) - - for bytes in sorted(figs.keys()): - pylab.figure(figs[bytes].number) - pylab.xlabel("Tick (s)") - pylab.ylabel("Download Time (s)") - pylab.title("mean time to download {0} of {1} bytes, all clients over time".format('first' if 'first' in bytekey else 'last', bytes)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_downloads(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - dls = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "time_to_last_byte" in d: - for b in d["time_to_last_byte"]: - bytes = int(b) - if bytes not in figs: figs[bytes] = pylab.figure() - if bytes not in dls: dls[bytes] = {} - if client not in dls[bytes]: dls[bytes][client] = 0 - for sec in d["time_to_last_byte"][b]: dls[bytes][client] += len(d["time_to_last_byte"][b][sec]) - for bytes in dls: - x, y = getcdf(list(dls[bytes].values()), shownpercentile=1.0) - pylab.figure(figs[bytes].number) - pylab.plot(x, y, lineformat, label=label) - - for bytes in sorted(figs.keys()): - pylab.figure(figs[bytes].number) - pylab.xlabel("Downloads Completed (\#)") - pylab.ylabel("Cumulative Fraction") - pylab.title("number of {0} byte downloads completed, each client".format(bytes)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_downloads_timeseries(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - dls = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "time_to_last_byte" in d: - for b in d["time_to_last_byte"]: - bytes = int(b) - if bytes not in figs: figs[bytes] = pylab.figure() - if bytes not in dls: dls[bytes] = {} - for sec in d["time_to_last_byte"][b]: - if sec not in dls[bytes]: dls[bytes][sec] = 0 - dls[bytes][sec] += len(d["time_to_last_byte"][b][sec]) - for bytes in dls: - pylab.figure(figs[bytes].number) - x = [sec for sec in dls[bytes]] - x.sort() - y = [dls[bytes][sec] for sec in x] - pylab.plot(x, y, lineformat, label=label) - - for bytes in sorted(figs.keys()): - pylab.figure(figs[bytes].number) - pylab.xlabel("Tick (s)") - pylab.ylabel("Downloads Completed (\#)") - pylab.title("number of {0} byte downloads completed, all clients over time".format(bytes)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_errors(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - dls = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "errors" in d: - for code in d["errors"]: - if code not in figs: figs[code] = pylab.figure() - if code not in dls: dls[code] = {} - if client not in dls[code]: dls[code][client] = 0 - for sec in d["errors"][code]: dls[code][client] += len(d["errors"][code][sec]) - for code in dls: - x, y = getcdf([dls[code][client] for client in dls[code]], shownpercentile=1.0) - pylab.figure(figs[code].number) - pylab.plot(x, y, lineformat, label=label) - - for code in sorted(figs.keys()): - pylab.figure(figs[code].number) - pylab.xlabel("Download Errors (\#)") - pylab.ylabel("Cumulative Fraction") - pylab.title("number of transfer {0} errors, each client".format(code)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_errors_timeseries(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - dls = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "errors" in d: - for code in d["errors"]: - if code not in figs: figs[code] = pylab.figure() - if code not in dls: dls[code] = {} - for sec in d["errors"][code]: - if sec not in dls[code]: dls[code][sec] = 0 - dls[code][sec] += len(d["errors"][code][sec]) - for code in dls: - pylab.figure(figs[code].number) - x = [sec for sec in dls[code]] - x.sort() - y = [dls[code][sec] for sec in x] - pylab.plot(x, y, lineformat, label=label) - - for code in sorted(figs.keys()): - pylab.figure(figs[code].number) - pylab.xlabel("Tick (s)") - pylab.ylabel("Download Errors (\#)") - pylab.title("number of transfer {0} errors, all clients over time".format(code)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_errsizes_all(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - err = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "errors" in d: - for code in d["errors"]: - if code not in figs: figs[code] = pylab.figure() - if code not in err: err[code] = [] - client_err_list = [] - for sec in d["errors"][code]: client_err_list.extend(d["errors"][code][sec]) - for b in client_err_list: err[code].append(int(b) / 1024.0) - for code in err: - x, y = getcdf(err[code]) - pylab.figure(figs[code].number) - pylab.plot(x, y, lineformat, label=label) - - for code in sorted(figs.keys()): - pylab.figure(figs[code].number) - pylab.xlabel("Data Transferred (KiB)") - pylab.ylabel("Cumulative Fraction") - pylab.title("bytes transferred before {0} error, all downloads".format(code)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_errsizes_median(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - err = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "errors" in d: - for code in d["errors"]: - if code not in figs: figs[code] = pylab.figure() - if code not in err: err[code] = [] - client_err_list = [] - for sec in d["errors"][code]: client_err_list.extend(d["errors"][code][sec]) - err[code].append(numpy.median(client_err_list) / 1024.0) - for code in err: - x, y = getcdf(err[code]) - pylab.figure(figs[code].number) - pylab.plot(x, y, lineformat, label=label) - - for code in sorted(figs.keys()): - pylab.figure(figs[code].number) - pylab.xlabel("Data Transferred (KiB)") - pylab.ylabel("Cumulative Fraction") - pylab.title("median bytes transferred before {0} error, each client".format(code)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() - - def __plot_errsizes_mean(self): - figs = {} - - for (anal, label, lineformat) in self.datasets: - err = {} - for client in anal.get_nodes(): - d = anal.get_tgen_transfers_summary(client) - if d is None: continue - if "errors" in d: - for code in d["errors"]: - if code not in figs: figs[code] = pylab.figure() - if code not in err: err[code] = [] - client_err_list = [] - for sec in d["errors"][code]: client_err_list.extend(d["errors"][code][sec]) - err[code].append(numpy.mean(client_err_list) / 1024.0) - for code in err: - x, y = getcdf(err[code]) - pylab.figure(figs[code].number) - pylab.plot(x, y, lineformat, label=label) - - for code in sorted(figs.keys()): - pylab.figure(figs[code].number) - pylab.xlabel("Data Transferred (KiB)") - pylab.ylabel("Cumulative Fraction") - pylab.title("mean bytes transferred before {0} error, each client".format(code)) - pylab.legend(loc="lower right") - self.page.savefig() - pylab.close() + def __extract_data_frame(self): + transfers = [] + for (analysis, label, lineformat) in self.datasets: + for client in analysis.get_nodes(): + tgen_transfers = analysis.get_tgen_transfers(client) + for transfer_id, transfer_data in tgen_transfers.items(): + transfer = {"transfer_id": transfer_id, "label": label, + "filesize_bytes": transfer_data["filesize_bytes"]} + if "elapsed_seconds" in transfer_data: + s = transfer_data["elapsed_seconds"] + if "command" in s: + if "first_byte" in s: + transfer["time_to_first_byte"] = s["first_byte"] - s["command"] + if "last_byte" in s: + transfer["time_to_last_byte"] = s["last_byte"] - s["command"] + if "error_code" in transfer_data and transfer_data["error_code"] != "NONE": + transfer["error_code"] = transfer_data["error_code"] + if "unix_ts_start" in transfer_data: + transfer["start"] = datetime.datetime.utcfromtimestamp(transfer_data["unix_ts_start"]) + transfers.append(transfer) + self.data = pd.DataFrame.from_records(transfers, index="transfer_id") + + def __plot_firstbyte_ecdf(self): + self.__draw_ecdf(x="time_to_first_byte", hue="label", hue_name="Data set", + data=self.data, title="Time to download first byte", + xlabel="Download time (s)", ylabel="Cumulative Fraction") + + def __plot_firstbyte_time(self): + for bytes in self.data["filesize_bytes"].unique(): + self.__draw_timeplot(x="start", y="time_to_last_byte", hue="label", hue_name="Data set", + data=self.data[self.data["filesize_bytes"]==bytes], + title="Time to download first of {0} bytes over time".format(bytes), + xlabel="Download start time", ylabel="Download time (s)") + + def __plot_lastbyte_ecdf(self): + for bytes in self.data["filesize_bytes"].unique(): + self.__draw_ecdf(x="time_to_last_byte", hue="label", hue_name="Data set", + data=self.data[self.data["filesize_bytes"]==bytes], + title="Time to download last of {0} bytes".format(bytes), + xlabel="Download time (s)", ylabel="Cumulative Fraction") + + def __plot_lastbyte_box(self): + for bytes in self.data["filesize_bytes"].unique(): + self.__draw_boxplot(x="label", y="time_to_last_byte", + data=self.data[self.data["filesize_bytes"]==bytes], + title="Time to download last of {0} bytes".format(bytes), + xlabel="Data set", ylabel="Download time (s)") + + def __plot_lastbyte_bar(self): + for bytes in self.data["filesize_bytes"].unique(): + self.__draw_barplot(x="label", y="time_to_last_byte", + data=self.data[self.data["filesize_bytes"]==bytes], + title="Mean time to download last of {0} bytes".format(bytes), + xlabel="Data set", ylabel="Downloads time (s)") + + def __plot_lastbyte_time(self): + for bytes in self.data["filesize_bytes"].unique(): + self.__draw_timeplot(x="start", y="time_to_last_byte", hue="label", hue_name="Data set", + data=self.data[self.data["filesize_bytes"] == bytes], + title="Time to download last of {0} bytes over time".format(bytes), + xlabel="Download start time", ylabel="Download time (s)") + + def __plot_downloads_count(self): + for bytes in self.data["filesize_bytes"].unique(): + self.__draw_countplot(x="label", + data=self.data[self.data["filesize_bytes"] == bytes], + xlabel="Data set", ylabel="Downloads completed (#)", + title="Number of downloads of {0} bytes completed".format(bytes)) + + def __plot_errors_count(self): + if "error_code" in self.data.columns: + self.__draw_countplot(x="error_code", hue="label", hue_name="Data set", data=self.data, + xlabel="Error code", ylabel="Downloads failed (#)", + title="Number of downloads failed") + + def __plot_errors_time(self): + if "error_code" in self.data.columns: + self.__draw_stripplot(x="start", y="error_code", hue="label", hue_name="Data set", + data = self.data, + xlabel="Download start time", ylabel="Error code", + title="Downloads failed over time") + + def __draw_ecdf(self, x, hue, hue_name, data, title, xlabel, ylabel): + data = data.dropna(subset=[x]) + p0 = data[x].quantile(q=0.0, interpolation="lower") + p99 = data[x].quantile(q=0.99, interpolation="higher") + ranks = data.groupby(hue)[x].rank(pct=True) + ranks.name = "rank_pct" + result = pd.concat([data[[hue, x]], ranks], axis=1) + result = result.append(pd.DataFrame({hue: data[hue].unique(), + x: p0 - (p99 - p0) * 0.05, "rank_pct": 0.0}), + ignore_index=True, sort=False) + result = result.append(pd.DataFrame({hue: data[hue].unique(), + x: p99 + (p99 - p0) * 0.05, "rank_pct": 1.0}), + ignore_index=True, sort=False) + result = result.rename(columns={hue: hue_name}) + plt.figure() + g = sns.lineplot(data=result, x=x, y="rank_pct", + hue=hue_name, drawstyle="steps-post") + g.set(title=title, xlabel=xlabel, ylabel=ylabel, + xlim=(p0 - (p99 - p0) * 0.03, p99 + (p99 - p0) * 0.03)) + sns.despine() + self.page.savefig() + plt.close() + + def __draw_timeplot(self, x, y, hue, hue_name, data, title, xlabel, ylabel): + plt.figure() + data = data.dropna(subset=[y]) + data = data.rename(columns={hue: hue_name}) + xmin = data[x].min() + xmax = data[x].max() + ymax = data[y].max() + g = sns.scatterplot(data=data, x=x, y=y, hue=hue_name, alpha=0.5) + g.set(title=title, xlabel=xlabel, ylabel=ylabel, + xlim=(xmin - 0.03 * (xmax - xmin), xmax + 0.03 * (xmax - xmin)), + ylim=(-0.05 * ymax, ymax * 1.05)) + plt.xticks(rotation=10) + sns.despine() + self.page.savefig() + plt.close() + + def __draw_boxplot(self, x, y, data, title, xlabel, ylabel): + plt.figure() + data = data.dropna(subset=[y]) + g = sns.boxplot(data=data, x=x, y=y, sym="") + g.set(title=title, xlabel=xlabel, ylabel=ylabel, ylim=(0, None)) + sns.despine() + self.page.savefig() + plt.close() + + def __draw_barplot(self, x, y, data, title, xlabel, ylabel): + plt.figure() + data = data.dropna(subset=[y]) + g = sns.barplot(data=data, x=x, y=y, ci=None) + g.set(title=title, xlabel=xlabel, ylabel=ylabel) + sns.despine() + self.page.savefig() + plt.close() + + def __draw_countplot(self, x, data, title, xlabel, ylabel, hue=None, hue_name=None): + plt.figure() + if hue is not None: + data = data.rename(columns={hue: hue_name}) + g = sns.countplot(data=data.dropna(subset=[x]), x=x, hue=hue_name) + g.set(xlabel=xlabel, ylabel=ylabel, title=title) + sns.despine() + self.page.savefig() + plt.close() + + def __draw_stripplot(self, x, y, hue, hue_name, data, title, xlabel, ylabel): + plt.figure() + data = data.rename(columns={hue: hue_name}) + xmin = data[x].min() + xmax = data[x].max() + data = data.dropna(subset=[y]) + g = sns.stripplot(data=data, x=x, y=y, hue=hue_name) + g.set(title=title, xlabel=xlabel, ylabel=ylabel, + xlim=(xmin - 0.03 * (xmax - xmin), xmax + 0.03 * (xmax - xmin))) + plt.xticks(rotation=10) + sns.despine() + self.page.savefig() + plt.close() # helper - compute the window_size moving average over the data in interval def movingaverage(interval, window_size):