From ccd03dbc23130d334dab53c1b56467cf50d80a49 Mon Sep 17 00:00:00 2001
From: KyleKlenk <kyle.c.klenk@gmail.com>
Date: Wed, 26 Apr 2023 14:33:57 -0600
Subject: [PATCH] Update README
Do not need separate config file for specifying threads. Can use $SLURM_CPUS_PER_TASK
---
{bin => .vscode}/caf-application.conf | 0
README.md | 37 ++--
build/source/actors/main.cpp | 2 +-
utils/StatisticsScripts/plots.py | 258 ++++++++++++++++++++++++++
4 files changed, 280 insertions(+), 17 deletions(-)
rename {bin => .vscode}/caf-application.conf (100%)
create mode 100644 utils/StatisticsScripts/plots.py
diff --git a/bin/caf-application.conf b/.vscode/caf-application.conf
similarity index 100%
rename from bin/caf-application.conf
rename to .vscode/caf-application.conf
diff --git a/README.md b/README.md
index da7ed4b..c1daf29 100644
--- a/README.md
+++ b/README.md
@@ -50,39 +50,44 @@ SUMMA-Actors supports four build types: Debug, Cluster, Release, and Cluster_Deb
Running SUMMA-Actors is done with the following command:
./summa_actor -g startGRU -n numGRU -c path_to_config_file
+If you are running SUMMA-Actors on a cluster, you will need to specify the number of threads when not using whole nodes.
+ This can be done with the --caf.scheduler.max-threads option
+ ./summa_actor -g startGRU -n numGRU -c path_to_config_file --caf.scheduler.max-threads $SLURM_CPUS_PER_TASK
+
The values for -g and -n are integers where -c is the full path to the configuraiton file for summa actors.
The configuration file is a json file. The contents of the JSON file are below:
- {
+{
"Distributed_Settings": {
"distributed_mode": false,
- "servers_list": [{"hostname": "simlab01"}, {"hostname": "simlab05"}],
+ "servers_list": [{"hostname": "cnic-giws-cpu-19001-04"}, {"hostname": "cnic-giws-utl-19002"}, {"hostname": "cnic-giws-utl-19003"}],
"port": 4444,
- "total_hru_count": 517315,
- "num_hru_per_batch": 1000
+ "total_hru_count": 800,
+ "num_hru_per_batch": 50
},
-
+
"Summa_Actor": {
- "max_gru_per_job": 500
+ "max_gru_per_job": 4000
},
-
+
"File_Access_Actor": {
- "num_partitions_in_output_buffer": 4,
+ "num_partitions_in_output_buffer": 8,
"num_timesteps_in_output_buffer": 500
},
-
+
"Job_Actor": {
- "file_manager_path": "/gladwell/kck540/Sundials_Settings/fileManager_actors.txt",
- "output_csv": false,
- "csv_path": ""
+ "file_manager_path": "/scratch/gwf/gwf_cmt/kck540/Summa-Actors/settings/file_manager_actors.txt",
+ "max_run_attempts": 3
},
-
+
"HRU_Actor": {
- "print_output": true,
- "output_frequency": 1000
+ "print_output": true,
+ "output_frequency": 100000,
+ "dt_init_factor": 1
}
- }
+}
+
The settings above should work for most use cases, some of the feautures we want to automate such as max_gru_per_job. However, the only field that you should have to adjust is the `file_manager_path`. This is the path to the file that manages the complete configuration of the SUMMA simulation. The summa confiuration files are explained in more depth in the follwoing (documentation)[https://summa.readthedocs.io/en/latest/input_output/SUMMA_input/]
diff --git a/build/source/actors/main.cpp b/build/source/actors/main.cpp
index 9451b50..ca5b8ee 100644
--- a/build/source/actors/main.cpp
+++ b/build/source/actors/main.cpp
@@ -155,7 +155,7 @@ void caf_main(actor_system& sys, const config& cfg) {
"fileManger is set with the \"-c\" option\n";
aout(self) << "EXAMPLE: ./summaMain -g 1 -n 10 -c location/of/config \n";
return;
- }
+ }
auto summa = sys.spawn(summa_actor, cfg.startGRU, cfg.countGRU, summa_actor_settings,
file_access_actor_settings, job_actor_settings, hru_actor_settings, self);
diff --git a/utils/StatisticsScripts/plots.py b/utils/StatisticsScripts/plots.py
new file mode 100644
index 0000000..6522089
--- /dev/null
+++ b/utils/StatisticsScripts/plots.py
@@ -0,0 +1,258 @@
+import matplotlib as mpl
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+from pylab import cm
+# from mpl_toolkits.axes_grid.anchored_artists import AnchoredText
+
+# MATPLOTLIBRC stuff
+import matplotlib as mpl
+# mpl.use('AGG') # for systems not running a GUI
+
+font = {'family' : 'serif',
+ 'weight' : 'normal',
+ 'size' : 20
+}
+mpl.rc('font', **font)
+
+def time_convert(seconds):
+ seconds = seconds % (24 * 3600)
+ hours = seconds / 3600
+ return round(hours, 2)
+
+def add_failures(failed_start_hru):
+ failure_data = pd.read_csv('SummaOriginal_rerun_failed_Jul-09.csv')
+
+ failed_wall_clock = failure_data.loc[failure_data["startHRU"] == failed_start_hru, "Wall-Clock Time"]
+
+
+ return(time_convert(failed_wall_clock.values[0]))
+
+
+def summa_worst_case():
+ # wall clock times of the processes
+ batches = ["1", "2", "3", "4", \
+ "5", "6", "7", "8"]
+
+ plt_1 = plt.figure(figsize=(12, 8))
+
+ results = [0.87, 0.75, 2.27, 0.87, 0.88, 0.75, 0.76, 1.48]
+
+ cmap = plt.get_cmap('viridis').copy()
+ original_color = cmap(0.5)
+
+
+ plt.bar(batches,results, color=original_color, alpha=0.6, edgecolor=original_color)
+ # plt.title('SUMMA Job CPU Distribution', fontsize=20)
+ plt.xlabel('Sub-task'
+# , fontsize=16
+ )
+ plt.ylabel('Hours'
+ # , fontsize=16
+ )
+ # plt.xticks(fontsize=14)
+ # plt.yticks(fontsize=14)
+ plt.savefig("SUMMA-worst-case.png",bbox_inches="tight")
+
+
+def summa_best_case():
+ batches = []
+
+def cpuComparsion(data_set_1, data_set_2):
+
+ df1 = pd.DataFrame(data_set_1)
+ df2 = pd.DataFrame(data_set_2)
+ SummaActors = df1["CPU Efficiency"].values
+ SummaOriginal = df2["CPU Efficiency"].values
+ print("Average CPU-Efficiency SUMMA-Actors =",sum(SummaActors) / len(SummaActors))
+ print("Average CPU-Efficiency SUMMA =",sum(SummaOriginal) / len(SummaOriginal))
+
+ summa_count = 0
+ actors_count = 0
+ for x in range(0, len(SummaOriginal)):
+ if SummaActors[x] > SummaOriginal[x]:
+ actors_count += 1
+ else:
+ summa_count += 1
+
+ print("TOTAL ACTORS COUNT =", actors_count)
+ print("TOTAL SUMMA COUNT =", summa_count)
+
+ min_summa_actors = min(SummaActors)
+ max_summa_actors = max(SummaActors)
+
+ min_summa_original = min(SummaOriginal)
+ max_summa_original = max(SummaOriginal)
+ string_text = "SUMMA-Actors Range: [{}% - {}%]\nSUMMA Range: [{}% - {}%]".format(min_summa_actors, max_summa_actors, min_summa_original, max_summa_original)
+ # text = AnchoredText(string_text, prop=dict(size=12), frameon=True, loc=2)
+
+ # Create plot
+ nbins = 50
+
+ # Setup colors
+ cmap = plt.get_cmap('viridis').copy()
+ actors_color = cmap(0.02)
+ print("Actors color is: ", actors_color)
+ original_color = cmap(0.5)
+
+ # Get the binning range of the historgram plots (this is to help us ensure the bins are consistent)
+ hist, actors_bins, _ = plt.hist(SummaActors, bins = nbins, color = "red")
+ hist, original_bins, _ = plt.hist(SummaOriginal, bins = nbins, color = "pink")
+
+ # This needs to come after the histograms above so they do not get added to our plot.
+ fig, ax = plt.subplots(1,1, figsize=(12,8))
+
+ # Get the proper bins
+ lower = min(actors_bins[0],original_bins[0])
+ upper = max(actors_bins[-1],original_bins[-1])
+ linbins = np.linspace(lower,upper,nbins)
+
+
+ sa_hist = ax.hist(SummaActors, bins=linbins, alpha=0.75, label="SUMMA-Actors", edgecolor=actors_color, linewidth=0.5, color=actors_color)
+ so_hist = ax.hist(SummaOriginal, bins=linbins, alpha=0.6, label="SUMMA", edgecolor=original_color, linewidth=0.5, color=original_color)
+
+ # Set Lables
+ ax.set_xlabel('CPU Efficiency as Percentage')
+ ax.set_ylabel('Count')
+
+ props = dict(boxstyle='round', facecolor="cornflowerblue", alpha=0.5, edgecolor='black')
+ ax.text(0.02, 0.8, string_text, transform=ax.transAxes,
+ verticalalignment='top', bbox=props)
+
+
+
+ # plot legend
+ handles, labels = plt.gca().get_legend_handles_labels()
+ order = [0,1]
+ ax.legend([handles[idx] for idx in order], [labels[idx] for idx in order], loc=0, edgecolor='black')
+
+ plt.savefig("CPU-Efficiency.png", bbox_inches="tight")
+
+
+def wallClockComparison(data_set_1, data_set_2):
+ nbins = 50
+
+ data_set_1_values = data_set_1["job_duration"].values
+ data_set_1_reads = list(map(time_convert, data_set_1["read_duration"].values))
+ data_set_1_writes = list(map(time_convert, data_set_1["write_duration"].values))
+
+ data_set_2_values = data_set_2["job_duration"].values
+ data_set_2_reads = list(map(time_convert, data_set_2["read_duration"].values))
+ data_set_2_writes = list(map(time_convert, data_set_2["write_duration"].values))
+
+
+ print("Total Time For SUMMA-Actors = ", sum(data_set_1_values))
+ print(" Total Read Time = ", sum(data_set_1_reads))
+ print(" Total Write Time = ", sum(data_set_1_writes))
+ print(" Total IO Time = ", sum(data_set_1_reads) + sum(data_set_1_writes))
+ print(" Total CPU Time = ", sum(data_set_1_values) - (sum(data_set_1_reads) + sum(data_set_1_writes)))
+ print("Total Time For SUMMA = ", sum(data_set_2_values))
+ print(" Total Read Time = ", sum(data_set_2_reads))
+ print(" Total Write Time = ", sum(data_set_2_writes))
+ print(" Total IO Time = ", sum(data_set_2_reads) + sum(data_set_2_writes))
+ print(" Total CPU Time = ", sum(data_set_2_values) - (sum(data_set_2_reads) + sum(data_set_2_writes)))
+
+ summa_count = 0
+ actors_count = 0
+ for x in range(0, len(data_set_1_values)):
+ if data_set_1_values[x] < data_set_2_values[x]:
+ actors_count += 1
+ else:
+ summa_count += 1
+
+ print("TOTAL ACTORS COUNT =", actors_count)
+ print("TOTAL SUMMA COUNT =", summa_count)
+
+
+ histActors, sa_bins, _ = plt.hist(data_set_1_values, bins=50, edgecolor='black', linewidth=1.0)
+ # Get the binning range from a default histogram plot
+ histOriginal, so_bins, _ = plt.hist(data_set_2_values, bins=50, edgecolor='black', linewidth=1.0)
+
+ fig2, ax2 = plt.subplots(1, 1, figsize=(12, 8))
+
+
+ cmap = plt.get_cmap('viridis').copy()
+ sa_color = cmap(0.02)
+ so_color = cmap(0.5)
+
+ lower = min(sa_bins[0], so_bins[0])
+ upper = max(sa_bins[-1], so_bins[-1])
+ linbins = np.linspace(lower,upper,nbins)
+ binsForCDF = np.linspace(lower,upper,517)
+
+ pdf1 = histActors / sum(histActors)
+ # print(pdf1)
+ pdf2 = histOriginal / sum(histOriginal)
+ # print(pdf2)
+ newList1 = []
+ newList2 = []
+ newList1[:] = [x / 70 for x in data_set_1_values]
+ newList2[:] = [x / 70 for x in data_set_2_values]
+
+ # Create the bar graph
+ sa_n, sa_bins, patch = ax2.hist(data_set_1_values, bins=linbins, alpha=0.75, label="SUMMA-Actors", edgecolor=sa_color, linewidth=0.5, color=sa_color)
+ so_n, so_bins, patch2 = ax2.hist(data_set_2_values, bins=linbins, alpha=0.6, edgecolor=so_color, linewidth=0.5, label="SUMMA", color=so_color)
+
+ ax2.set_title("Wall-Clock Time Comparison Of Jobs On Each CPU", fontsize=20)
+ ax2.set_xlabel('Wall-Clock Time (Hours)'
+ # ,fontsize=16
+ )
+ ax2.set_ylabel('Count'
+ # ,fontsize=16
+ )
+
+ # Merge graphs
+ ax3 = ax2.twinx()
+
+ ax3.hist(data_set_1_values, bins=sa_bins, linewidth=1.5, histtype='step', cumulative=True, label="SUMMA-Actors", color="#CC00CC")
+ ax3.hist(data_set_2_values, bins=so_bins, linewidth=1.5, histtype='step', cumulative=True, label="SUMMA", color="#00CCCC")
+ ax3.set_ylabel('Total Number of Bins', labelpad=15)
+
+ # Get rid of line that goes straigh down at the end of this plot
+ axpolygons = [poly for poly in ax3.get_children() if isinstance(poly, mpl.patches.Polygon)]
+ for poly in axpolygons:
+ poly.set_xy(poly.get_xy()[:-1])
+
+
+ handles1, labels1 = ax2.get_legend_handles_labels()
+ handles2, labels2 = ax3.get_legend_handles_labels()
+
+ ax2.legend((*handles1, *handles2), (*len(labels1)*[''], *labels2),
+ loc='right', ncol=2, handlelength=3, fontsize=16)
+ # ax2.legend(())
+ plt.savefig("WallClock.png",bbox_inches="tight")
+
+def boxPlot(data_set_1, data_set_2):
+
+ data_set_1_reads = data_set_1["read_duration"].values
+ data_set_1_writes = data_set_1["write_duration"].values
+
+ data_set_2_reads = data_set_2["read_duration"].values
+ data_set_2_writes = data_set_2["write_duration"].values
+
+
+ fig, ax = plt.subplots(1, 1, figsize=(14, 8))
+
+ ax.set_title("Box Plot Of Read And Write Times", fontsize=20)
+ ax.set_xlabel('Read/Write', fontsize=14)
+ ax.set_ylabel('Time (Seconds)', fontsize=14)
+
+ ax.boxplot([data_set_1_reads, data_set_2_reads, data_set_1_writes, data_set_2_writes], labels=["SUMMA-Actors Read", "SUMMA Read", "SUMMA-Actors Write", "SUMMA Write"], flierprops=dict(markerfacecolor='r', marker='D'), vert=False)
+ # set size of tick labels
+ ax.tick_params(axis='both', which='major', labelsize=14)
+
+ plt.savefig("BoxPlot.png",bbox_inches="tight")
+
+
+
+
+# Assemble Da"a
+data_set_1 = pd.read_csv("/scratch/gwf/gwf_cmt/kck540/Single_CPU_Test/actors/logs/_log_summaryActors_sorted.csv")
+data_set_2 = pd.read_csv("/scratch/gwf/gwf_cmt/kck540/Single_CPU_Test/non-actors/logs/_log_summaryOriginal_sorted.csv")
+
+
+# data_set 1 and 2 are used for the paper
+
+
+wallClockComparison(data_set_1, data_set_2)
+# boxPlot(data_set_1, data_set_2)
--
GitLab