Add benchmarking tool

Author: evax

.gitignore

@@ -3,6 +3,7 @@ ebin/*.beam
 priv/*.so
 c_src/*.o
 deps
+graphs
 perf/*.beam
 .eunit
 doc/edoc-info

Makefile

@@ -34,3 +34,10 @@ test: compile
 
 docs:
 	@./rebar doc
+
+bench: perf
+	@echo 'Running benchmarks, this could take some time...'
+	@mkdir -p graphs
+	@./perf/perfgraphs.py
+	@mv -f *.png graphs/
+

README.md

@@ -44,6 +44,12 @@ Run the test suite
 
     $ make test
 
+Run the benchmarks (requires [python](http://www.python.org) and [matplotlib](http://matplotlib.sourceforge.net/))
+
+    $ make bench
+
+This will run performance tests and output png graphs in the graphs directory.
+
 Please note that to behave properly on your system ZeroMQ might require [some tuning](http://www.zeromq.org/docs:tuning-zeromq).
 
 Architecture

perf/perfgraphs.py

@@ -0,0 +1,93 @@
+#! /usr/bin/env python
+import subprocess
+import matplotlib.pyplot as plt
+
+def do_test(localcmd, remotecmd):
+    local = subprocess.Popen(localcmd.split(), stdout=subprocess.PIPE)
+    remote = subprocess.Popen(remotecmd.split(), stdout=subprocess.PIPE,
+                              close_fds=True)
+    local.wait()
+    remote.wait()
+    return (local.stdout.read(), remote.stdout.read())
+
+def thr_run(iterations, msgsize=1, msgcount=100000, active=False):
+    localcmd = "escript perf/local_thr"+(active and "_active" or "")+".erl" \
+               " tcp://lo:1234 "+str(msgsize)+" "+str(msgcount)
+    remotecmd = "escript perf/remote_thr.erl" \
+                " tcp://localhost:1234 "+str(msgsize)+" "+str(msgcount)
+    results = []
+    for i in xrange(iterations):
+        out = do_test(localcmd, remotecmd)[0]
+        results.append(float(out.split('\n')[2].split(' ')[2]))
+    return results
+
+def lat_run(iterations, msgsize=1, msgcount=100000, active=False):
+    localcmd = "escript perf/local_lat"+(active and "_active" or "")+".erl" \
+               " tcp://lo:1234 "+str(msgsize)+" "+str(msgcount)
+    remotecmd = "escript perf/remote_lat"+(active and "_active" or "")+".erl" \
+                " tcp://localhost:1234 "+str(msgsize)+" "+str(msgcount)
+    results = []
+    for i in xrange(iterations):
+        out = do_test(localcmd, remotecmd)[1]
+        results.append(float(out.split('\n')[2].split(' ')[2]))
+    return results
+
+def average(l):
+    return [sum(l)/float(len(l))]*len(l)
+
+class TestRun(object):
+    def __init__(self, iterations, msgsize, msgcount):
+        self.iterations = iterations
+        self.msgsize = msgsize
+        self.msgcount = msgcount
+
+    def plot_all(self):
+        self.plot_thr()
+        self.plot_lat()
+
+    def title(self):
+        return "%d messages in of %d byte(s)" % \
+                    (self.msgcount, self.msgsize)
+
+    def plot_thr(self):
+        results = thr_run(self.iterations, self.msgsize,
+                          self.msgcount, False)
+        results_active = thr_run(self.iterations, self.msgsize,
+                                 self.msgcount, True)
+        plt.clf()
+        plt.title("Throughput: %s" % self.title())
+        plt.ylabel("Msg/s")
+        plt.plot(results, 'bo-', label="passive")
+        plt.plot(average(results), 'b-', label="passive average",
+                                         linewidth=2)
+        plt.plot(results_active, 'ro-', label="active")
+        plt.plot(average(results_active), 'r-', label="active average",
+                                                linewidth=2)
+        plt.legend()
+        plt.savefig("thr_%d_%d_%d.png" % (self.iterations, self.msgsize,
+                                          self.msgcount))
+
+    def plot_lat(self):
+        results = lat_run(self.iterations, self.msgsize,
+                          self.msgcount, False)
+        results_active = lat_run(self.iterations, self.msgsize,
+                                 self.msgcount, True)
+        plt.clf()
+        plt.title("Latency: %s" % self.title())
+        plt.ylabel("Average latency (us)")
+        plt.plot(results, 'bo-', label="passive")
+        plt.plot(average(results), 'b-', label="passive average",
+                                         linewidth=2)
+        plt.plot(results_active, 'ro-', label="active")
+        plt.plot(average(results_active), 'r-', label="active average",
+                                                linewidth=2)
+        plt.legend()
+        plt.savefig("lat_%d_%d_%d.png" % (self.iterations, self.msgsize,
+                                          self.msgcount))
+
+if __name__ == "__main__":
+    TestRun(100, 1, 100000).plot_thr()
+    TestRun(50, 1024, 100000).plot_thr()
+    TestRun(40, 1, 10000).plot_lat()
+    TestRun(20, 1024, 10000).plot_lat()
+