CODARcode · celiafish · Jul 18, 2019 · Jul 11, 2019 · Jul 12, 2019 · Jul 12, 2019
diff --git a/.gitignore b/.gitignore
@@ -29,3 +29,5 @@ data/
 utils/__pycache__/
 
 log/
+
+__pycache__
diff --git a/benchmark/data_handler.py b/benchmark/data_handler.py
@@ -0,0 +1,116 @@
+import struct
+import random
+import threading
+import time
+import os
+from queue import Queue
+
+class RandomMessage(object):
+    def __init__(self):
+        self.filename = None
+        # message size (default 1 MBytes)
+        self.size = 1024 * 1024
+        # the number of integer elements in a message
+        self.count = int(self.size / 4)
+
+        self.data = None
+        self.binary = None
+
+        self.generate()
+
+        # thread to process pseudo message
+        self.q = Queue()
+        self.terminate_event = threading.Event()
+        self.thread = threading.Thread(target=self._run)
+        self.thread.start()
+
+        # statistics
+        self.total_message = 0
+        self.min_start_time = None
+        self.max_end_time = None
+        self.acc_process_time = 0
+        self.status = True
+
+    def generate(self, size = None, filename = None):
+        if size is None: size = self.size
+        self.filename = filename
+
+        self.size = size - size%4
+        self.count = int(self.size/4)
+
+        if filename is None or not os.path.exists(filename):
+            self.data = [random.randint(1, 99999) for _ in range(self.count)]
+            self.binary = struct.pack('{:d}i'.format(self.count), *self.data)
+            self.save()
+        else:
+            self.load()
+
+    def save(self):
+        if self.filename is None: return
+        if self.binary is None: return
+        with open(self.filename, 'wb') as f:
+            f.write(self.binary)
+
+    def load(self):
+        if self.filename is None: return
+        with open(self.filename, 'rb') as f:
+            self.binary = f.read()
+            self.data = list(struct.unpack('{:d}i'.format(self.count), self.binary))
+
+    def is_equal(self, binary):
+        return self.binary == binary
+
+    def add_message(self, binary):
+        """This function will be accessed by multiple threads"""
+        self.q.put([time.time(), binary])
+
+    def _run(self):
+        while not self.terminate_event.isSet():
+            start_time, binary = self.q.get()
+
+            if binary is None:
+                break
+
+            # simple operation (e.g. validate data or some other operations)
+            self.status = self.status and self.is_equal(binary)
+            end_time = time.time()
+
+            # update statistics
+            self.total_message = self.total_message + 1
+
+            self.min_start_time = min(self.min_start_time, start_time) \
+                if self.min_start_time is not None else start_time
+
+            self.max_end_time = max(self.max_end_time, end_time) \
+                if self.max_end_time is not None else end_time
+
+            elapsed = end_time - start_time
+            self.acc_process_time = self.acc_process_time + elapsed
+
+            self.q.task_done()
+
+    def join(self):
+        if not self.thread.is_alive():
+            return
+
+        self.q.join() # block until all tasks are done in the queue
+        self.terminate_event.set()
+        self.thread.join()
+
+    def show_statistics(self):
+        print("=================================================")
+        print("Status                : {}".format("PASSED" if self.status else "FAILED"))
+        print("Total message received: {}".format(self.total_message))
+        if self.total_message:
+            print("Min. start time       : {:.3f} sec".format(self.min_start_time))
+            print("Max. end time         : {:.3f} sec".format(self.max_end_time))
+            print("Acc. process time     : {:.3f} sec".format(self.acc_process_time))
+            print("Total elapsed time    : {:.3f} sec".format(self.max_end_time - self.min_start_time))
+            print("Avg. process time     : {:.3f} sec".format(self.acc_process_time/self.total_message))
+            print("Throughput            : {:.3f} MBytes/sec".format(
+                self.total_message*self.size/1024/1024/(self.max_end_time - self.min_start_time)
+            ))
+            print("Throughput            : {:.3f} Messages/sec".format(
+                self.total_message/(self.max_end_time - self.min_start_time)
+            ))
+        print("=================================================")
diff --git a/benchmark/results/run_benchmark_summit_200.lsf b/benchmark/results/run_benchmark_summit_200.lsf
@@ -0,0 +1,68 @@
+#!/bin/bash
+# Begin LSF Directives
+#BSUB -P CSC299
+#BSUB -W 1:00
+#BSUB -nnodes 6
+#BSUB -J ws-test-200
+#BSUB -o ws-test-200.o.%J
+#BSUB -e ws-test-200.e.%J
+
+module load gcc/8.1.1
+module load curl/7.63.0
+module load python/3.7.0-anaconda3-5.3.0
+
+#set -x
+
+root=`pwd`
+mega=$(( 1024*1024  ))
+msz_count=$(( 100  ))
+
+for msz_mbytes in 1 2 4
+do
+
+  echo
+  echo "========== 200 with ${msz_mbytes} MBytes =========="
+
+  # prepare data and launch web server
+  addr="http://`jsrun -n 1 hostname`:5000"
+  msz_size=$(( ${msz_mbytes} * ${mega}  ))
+  msz_fn="${root}/msg_200.bin"
+  log_fn="${root}/msg_200.log"
+  echo "web server @ ${addr}"
+  jsrun -n 1 -a 1 -c 42 -g 0 -r 1 python3 ws_flask.py $addr $msz_size $msz_fn $log_fn &
+  ws_pid=$!
+  while [ ! -f ${msz_fn} ]
+  do
+    echo "wait pseudo-message"
+    sleep 1
+  done  
+  echo "pseudo-message is ready!"
+
+  # start sending pseudo-messages
+  s_time="$(date -u +%s.%N)"
+  jsrun -n 5 -a 40 -c 40 -g 0 -r 1 python3 send_message.py "${addr}/messages" $msz_fn $msz_count
+  e_time="$(date -u +%s.%N)"
+
+  # print out statistics
+  total_ranks=$(( 5 * 40  ))
+  elapsed="$(bc -l <<<"$e_time-$s_time")"
+  out1=$(bc -l <<<"${msz_mbytes}*${msz_count}*${total_ranks}/(${e_time}-${s_time})")
+  out2=$(bc -l <<<"${msz_count}*${total_ranks}/(${e_time}-${s_time})")
+  echo 
+  echo "# Ranks             : $total_ranks"
+  echo "Message size        : $msz_mbytes MBytes"
+  echo "# Message (per rank): $msz_count"
+  echo "Elapsed time        : $elapsed sec"
+  echo "Throughput          : $out1 MBytes/sec"
+  echo "Throughput          : $out2 Messages/sec"
+  echo 
+
+  # clean for the next run
+  jsrun -n 1 -c 1 curl --silent --output /dev/null -X POST "${addr}/shutdown"
+  rm -f ${msz_fn} ${log_fn}
+  wait $ws_pid
+  echo "======================================================"
+  echo
+
+done
+
diff --git a/benchmark/results/ws-test-200.e.512751 b/benchmark/results/ws-test-200.e.512751
@@ -0,0 +1,7 @@
+
+Lmod is automatically replacing "xl/16.1.1-3" with "gcc/8.1.1".
+
+
+Due to MODULEPATH changes, the following have been reloaded:
+  1) spectrum-mpi/10.3.0.1-20190611
+
diff --git a/benchmark/results/ws-test-200.o.512751 b/benchmark/results/ws-test-200.o.512751
@@ -0,0 +1,109 @@
+
+========== 200 with 1 MBytes ==========
+web server @ http://h19n15:5000
+wait pseudo-message
+wait pseudo-message
+pseudo-message is ready!
+
+# Ranks             : 200
+Message size        : 1 MBytes
+# Message (per rank): 100
+Elapsed time        : 171.537479508 sec
+Throughput          : 116.59259572522318211046 MBytes/sec
+Throughput          : 116.59259572522318211046 Messages/sec
+
+=================================================
+Status                : PASSED
+Total message received: 20000
+Min. start time       : 1563380683.435 sec
+Max. end time         : 1563380853.165 sec
+Acc. process time     : 5.965 sec
+Total elapsed time    : 169.731 sec
+Avg. process time     : 0.000 sec
+Throughput            : 117.834 MBytes/sec
+Throughput            : 117.834 Messages/sec
+=================================================
+======================================================
+
+
+========== 200 with 2 MBytes ==========
+web server @ http://h19n15:5000
+wait pseudo-message
+wait pseudo-message
+pseudo-message is ready!
+
+# Ranks             : 200
+Message size        : 2 MBytes
+# Message (per rank): 100
+Elapsed time        : 340.276443815 sec
+Throughput          : 117.55148123549517294405 MBytes/sec
+Throughput          : 58.77574061774758647202 Messages/sec
+
+=================================================
+Status                : PASSED
+Total message received: 20000
+Min. start time       : 1563380858.040 sec
+Max. end time         : 1563381197.162 sec
+Acc. process time     : 7.876 sec
+Total elapsed time    : 339.121 sec
+Avg. process time     : 0.000 sec
+Throughput            : 117.952 MBytes/sec
+Throughput            : 58.976 Messages/sec
+=================================================
+======================================================
+
+
+========== 200 with 4 MBytes ==========
+web server @ http://h19n15:5000
+wait pseudo-message
+wait pseudo-message
+wait pseudo-message
+pseudo-message is ready!
+
+# Ranks             : 200
+Message size        : 4 MBytes
+# Message (per rank): 100
+Elapsed time        : 679.628615884 sec
+Throughput          : 117.71134724211570908910 MBytes/sec
+Throughput          : 29.42783681052892727227 Messages/sec
+
+=================================================
+Status                : PASSED
+Total message received: 20000
+Min. start time       : 1563381203.659 sec
+Max. end time         : 1563381881.362 sec
+Acc. process time     : 11.812 sec
+Total elapsed time    : 677.703 sec
+Avg. process time     : 0.001 sec
+Throughput            : 118.046 MBytes/sec
+Throughput            : 29.511 Messages/sec
+=================================================
+======================================================
+
+
+------------------------------------------------------------
+Sender: LSF System <lsfadmin@batch4>
+Subject: Job 512751: <ws-test-200> in cluster <summit> Done
+
+Job <ws-test-200> was submitted from host <login2> by user <sungsoo> in cluster <summit> at Wed Jul 17 12:12:03 2019
+Job was executed on host(s) <1*batch4>, in queue <batch>, as user <sungsoo> in cluster <summit> at Wed Jul 17 12:24:34 2019
+                            <42*h19n15>
+                            <42*h19n16>
+                            <42*h19n17>
+                            <42*h19n18>
+                            <42*h20n01>
+                            <42*h20n02>
+</ccs/home/sungsoo> was used as the home directory.
+</gpfs/alpine/world-shared/csc299/sungsoo/ChimbukoVisualization/benchmark> was used as the working directory.
+Started at Wed Jul 17 12:24:34 2019
+Terminated at Wed Jul 17 12:44:43 2019
+Results reported at Wed Jul 17 12:44:43 2019
+
+The output (if any) is above this job summary.
+
+
+
+PS:
+
+Read file <ws-test-200.e.512751> for stderr output of this job.
+
diff --git a/benchmark/run_benchmark.sh b/benchmark/run_benchmark.sh
@@ -0,0 +1,55 @@
+#!/usr/bin/env bash
+
+# test condition
+#root=`pwd`
+#nranks=10
+#msz_mbytes=$(( 50 ))
+#export msz_count=$(( 100 ))
+#export filename="${root}/message.bin"
+
+#export addr="http://`hostname`:5000"
+export addr="http://127.0.0.1:5000"
+echo "web server @ ${addr}"
+
+mega=$(( 1024*1024 ))
+msz_size=$(( ${msz_mbytes} * ${mega} ))
+
+# run a web server
+python3 ws_flask.py $addr $msz_size $filename &
+ws_pid=$!
+while [ ! -f ${filename} ]
+do
+    echo "wait pseudo-message"
+    sleep 10
+done
+echo "pseudo-message is ready!"
+#ls -al
+
+sleep 1
+# start send pseudo-messages
+start_time="$(date -u +%s.%N)"
+#mpirun -n $nranks ./send_message.sh
+mpirun -n $nranks python3 send_message.py "${addr}/messages" $filename $msz_count
+end_time="$(date -u +%s.%N)"
+
+elapsed="$(bc -l <<<"$end_time-$start_time")"
+throughput=$(bc -l <<<"${msz_mbytes}*${msz_count}*${nranks}/(${end_time}-${start_time})")
+throughput2=$(bc -l <<<"${msz_count}*${nranks}/(${end_time}-${start_time})")
+echo "================================================="
+echo "From sender perspective ...."
+echo "# Ranks             : $nranks"
+echo "Message size        : $msz_mbytes MBytes"
+echo "# Message (per rank): $msz_count "
+echo "Total elapsed time  : $elapsed sec"
+echo "Throughput          : $throughput MBytes/sec"
+echo "Throughput          : $throughput2 Messages/sec"
+echo "================================================="
+
+# at this point all message was sent, shutdown web server
+#curl -X POST http://127.0.0.1:5000/shutdown
+curl -X POST "${addr}/shutdown"
+echo
+
+wait $ws_pid
+#kill -9 $ws_pid
+rm -f ${root}/log.txt ${filename}
Original file line number	Diff line number	Diff line change
Expand Up		@@ -29,3 +29,5 @@ data/
		utils/__pycache__/

		log/

		__pycache__