Add script for run and upload the benchmark result to bq

15 files changed, 748 insertions, 208 deletions

diff --git a/benchmarks/Makefile.am b/benchmarks/Makefile.am
index 3de0e380..eba256f2 100644
--- a/benchmarks/Makefile.am
+++ b/benchmarks/Makefile.am
@@ -28,11 +28,11 @@ make_tmp_dir:
 if USE_EXTERNAL_PROTOC
 
 protoc_middleman: make_tmp_dir $(benchmarks_protoc_inputs)
-	$(PROTOC) -I$(srcdir) -I$(top_srcdir) --cpp_out=. --java_out=./tmp $(benchmarks_protoc_inputs) $(benchmarks_protoc_inputs_benchmark_wrapper)
+	$(PROTOC) -I$(srcdir) -I$(top_srcdir) --cpp_out=./cpp --java_out=./tmp $(benchmarks_protoc_inputs) $(benchmarks_protoc_inputs_benchmark_wrapper)
 	touch protoc_middleman
 
 protoc_middleman2: make_tmp_dir $(benchmarks_protoc_inputs_proto2)
-	$(PROTOC) -I$(srcdir) -I$(top_srcdir) --cpp_out=. --java_out=./tmp $(benchmarks_protoc_inputs_proto2)
+	$(PROTOC) -I$(srcdir) -I$(top_srcdir) --cpp_out=./cpp --java_out=./tmp $(benchmarks_protoc_inputs_proto2)
 	touch protoc_middleman2
 
 else
@@ -41,60 +41,60 @@ else
 # relative to srcdir, which may not be the same as the current directory when
 # building out-of-tree.
 protoc_middleman: make_tmp_dir $(top_srcdir)/src/protoc$(EXEEXT) $(benchmarks_protoc_inputs) $(well_known_type_protoc_inputs) $(benchmarks_protoc_inputs_benchmark_wrapper)
-	oldpwd=`pwd` && ( cd $(srcdir) && $$oldpwd/../src/protoc$(EXEEXT) -I. -I$(top_srcdir)/src --cpp_out=$$oldpwd --java_out=$$oldpwd/tmp/java/src/main/java --python_out=$$oldpwd/tmp $(benchmarks_protoc_inputs) $(benchmarks_protoc_inputs_benchmark_wrapper) )
+	oldpwd=`pwd` && ( cd $(srcdir) && $$oldpwd/../src/protoc$(EXEEXT) -I. -I$(top_srcdir)/src --cpp_out=$$oldpwd/cpp --java_out=$$oldpwd/tmp/java/src/main/java --python_out=$$oldpwd/tmp $(benchmarks_protoc_inputs) $(benchmarks_protoc_inputs_benchmark_wrapper) )
 	touch protoc_middleman
 
 protoc_middleman2:  make_tmp_dir $(top_srcdir)/src/protoc$(EXEEXT) $(benchmarks_protoc_inputs_proto2) $(well_known_type_protoc_inputs)
-	oldpwd=`pwd` && ( cd $(srcdir) && $$oldpwd/../src/protoc$(EXEEXT) -I. -I$(top_srcdir)/src --cpp_out=$$oldpwd --java_out=$$oldpwd/tmp/java/src/main/java --python_out=$$oldpwd/tmp $(benchmarks_protoc_inputs_proto2) )
+	oldpwd=`pwd` && ( cd $(srcdir) && $$oldpwd/../src/protoc$(EXEEXT) -I. -I$(top_srcdir)/src --cpp_out=$$oldpwd/cpp --java_out=$$oldpwd/tmp/java/src/main/java --python_out=$$oldpwd/tmp $(benchmarks_protoc_inputs_proto2) )
 	touch protoc_middleman2
 
 endif
 
-all_data = `find . -type f -name "dataset.*.pb"`
+all_data = $$(find . -type f -name "dataset.*.pb" -not -path "./tmp/*")
 
 ############# CPP RULES ##############
 
 benchmarks_protoc_outputs =                                                \
-	benchmarks.pb.cc                                                         \
-	datasets/google_message1/proto3/benchmark_message1_proto3.pb.cc
-  
+	cpp/benchmarks.pb.cc                                                     \
+	cpp/datasets/google_message1/proto3/benchmark_message1_proto3.pb.cc
+
 benchmarks_protoc_outputs_header =                                         \
-	benchmarks.pb.h                                                          \
-	datasets/google_message1/proto3/benchmark_message1_proto3.pb.h
+	cpp/benchmarks.pb.h                                                      \
+	cpp/datasets/google_message1/proto3/benchmark_message1_proto3.pb.h
 
 benchmarks_protoc_outputs_proto2_header =                                  \
-	datasets/google_message1/proto2/benchmark_message1_proto2.pb.h           \
-	datasets/google_message2/benchmark_message2.pb.h                         \
-	datasets/google_message3/benchmark_message3.pb.h                         \
-	datasets/google_message3/benchmark_message3_1.pb.h                       \
-	datasets/google_message3/benchmark_message3_2.pb.h                       \
-	datasets/google_message3/benchmark_message3_3.pb.h                       \
-	datasets/google_message3/benchmark_message3_4.pb.h                       \
-	datasets/google_message3/benchmark_message3_5.pb.h                       \
-	datasets/google_message3/benchmark_message3_6.pb.h                       \
-	datasets/google_message3/benchmark_message3_7.pb.h                       \
-	datasets/google_message3/benchmark_message3_8.pb.h                       \
-	datasets/google_message4/benchmark_message4.pb.h                         \
-	datasets/google_message4/benchmark_message4_1.pb.h                       \
-	datasets/google_message4/benchmark_message4_2.pb.h                       \
-	datasets/google_message4/benchmark_message4_3.pb.h
+	cpp/datasets/google_message1/proto2/benchmark_message1_proto2.pb.h       \
+	cpp/datasets/google_message2/benchmark_message2.pb.h                     \
+	cpp/datasets/google_message3/benchmark_message3.pb.h                     \
+	cpp/datasets/google_message3/benchmark_message3_1.pb.h                   \
+	cpp/datasets/google_message3/benchmark_message3_2.pb.h                   \
+	cpp/datasets/google_message3/benchmark_message3_3.pb.h                   \
+	cpp/datasets/google_message3/benchmark_message3_4.pb.h                   \
+	cpp/datasets/google_message3/benchmark_message3_5.pb.h                   \
+	cpp/datasets/google_message3/benchmark_message3_6.pb.h                   \
+	cpp/datasets/google_message3/benchmark_message3_7.pb.h                   \
+	cpp/datasets/google_message3/benchmark_message3_8.pb.h                   \
+	cpp/datasets/google_message4/benchmark_message4.pb.h                     \
+	cpp/datasets/google_message4/benchmark_message4_1.pb.h                   \
+	cpp/datasets/google_message4/benchmark_message4_2.pb.h                   \
+	cpp/datasets/google_message4/benchmark_message4_3.pb.h
 
 benchmarks_protoc_outputs_proto2 =                                         \
-	datasets/google_message1/proto2/benchmark_message1_proto2.pb.cc          \
-	datasets/google_message2/benchmark_message2.pb.cc                        \
-	datasets/google_message3/benchmark_message3.pb.cc                        \
-	datasets/google_message3/benchmark_message3_1.pb.cc                      \
-	datasets/google_message3/benchmark_message3_2.pb.cc                      \
-	datasets/google_message3/benchmark_message3_3.pb.cc                      \
-	datasets/google_message3/benchmark_message3_4.pb.cc                      \
-	datasets/google_message3/benchmark_message3_5.pb.cc                      \
-	datasets/google_message3/benchmark_message3_6.pb.cc                      \
-	datasets/google_message3/benchmark_message3_7.pb.cc                      \
-	datasets/google_message3/benchmark_message3_8.pb.cc                      \
-	datasets/google_message4/benchmark_message4.pb.cc                        \
-	datasets/google_message4/benchmark_message4_1.pb.cc                      \
-	datasets/google_message4/benchmark_message4_2.pb.cc                      \
-	datasets/google_message4/benchmark_message4_3.pb.cc
+	cpp/datasets/google_message1/proto2/benchmark_message1_proto2.pb.cc      \
+	cpp/datasets/google_message2/benchmark_message2.pb.cc                    \
+	cpp/datasets/google_message3/benchmark_message3.pb.cc                    \
+	cpp/datasets/google_message3/benchmark_message3_1.pb.cc                  \
+	cpp/datasets/google_message3/benchmark_message3_2.pb.cc                  \
+	cpp/datasets/google_message3/benchmark_message3_3.pb.cc                  \
+	cpp/datasets/google_message3/benchmark_message3_4.pb.cc                  \
+	cpp/datasets/google_message3/benchmark_message3_5.pb.cc                  \
+	cpp/datasets/google_message3/benchmark_message3_6.pb.cc                  \
+	cpp/datasets/google_message3/benchmark_message3_7.pb.cc                  \
+	cpp/datasets/google_message3/benchmark_message3_8.pb.cc                  \
+	cpp/datasets/google_message4/benchmark_message4.pb.cc                    \
+	cpp/datasets/google_message4/benchmark_message4_1.pb.cc                  \
+	cpp/datasets/google_message4/benchmark_message4_2.pb.cc                  \
+	cpp/datasets/google_message4/benchmark_message4_3.pb.cc
 
 
 $(benchmarks_protoc_outputs): protoc_middleman
@@ -118,12 +118,12 @@ AM_CXXFLAGS = $(NO_OPT_CXXFLAGS) $(PROTOBUF_OPT_FLAG) -Wall -Wwrite-strings -Wov
 bin_PROGRAMS = cpp-benchmark
 
 cpp_benchmark_LDADD = $(top_srcdir)/src/libprotobuf.la $(top_srcdir)/third_party/benchmark/src/libbenchmark.a
-cpp_benchmark_SOURCES = cpp_benchmark.cc
-cpp_benchmark_CPPFLAGS = -I$(top_srcdir)/src -I$(srcdir) -I$(top_srcdir)/third_party/benchmark/include
+cpp_benchmark_SOURCES = cpp/cpp_benchmark.cc
+cpp_benchmark_CPPFLAGS = -I$(top_srcdir)/src -I$(srcdir)/cpp -I$(top_srcdir)/third_party/benchmark/include
 # Explicit deps because BUILT_SOURCES are only done before a "make all/check"
 # so a direct "make test_cpp" could fail if parallel enough.
 # See: https://www.gnu.org/software/automake/manual/html_node/Built-Sources-Example.html#Recording-Dependencies-manually
-cpp_benchmark-cpp_benchmark.$(OBJEXT): $(benchmarks_protoc_outputs) $(benchmarks_protoc_outputs_proto2) $(benchmarks_protoc_outputs_header) $(benchmarks_protoc_outputs_proto2_header) $(top_srcdir)/src/libprotobuf.la $(top_srcdir)/third_party/benchmark/src/libbenchmark.a
+cpp/cpp_benchmark-cpp_benchmark.$(OBJEXT): $(benchmarks_protoc_outputs) $(benchmarks_protoc_outputs_proto2) $(benchmarks_protoc_outputs_header) $(benchmarks_protoc_outputs_proto2_header) $(top_srcdir)/src/libprotobuf.la $(top_srcdir)/third_party/benchmark/src/libbenchmark.a
 nodist_cpp_benchmark_SOURCES =                                             \
 	$(benchmarks_protoc_outputs)                                             \
 	$(benchmarks_protoc_outputs_proto2)                                      \
@@ -174,11 +174,11 @@ python_add_init: protoc_middleman protoc_middleman2
 python_cpp_pkg_flags = `pkg-config --cflags --libs python`
 
 lib_LTLIBRARIES = libbenchmark_messages.la
-libbenchmark_messages_la_SOURCES = python_benchmark_messages.cc
+libbenchmark_messages_la_SOURCES = python/python_benchmark_messages.cc
 libbenchmark_messages_la_LIBADD = $(top_srcdir)/src/.libs/libprotobuf.la
 libbenchmark_messages_la_LDFLAGS = -version-info 1:0:0 -export-dynamic
-libbenchmark_messages_la_CPPFLAGS = -I$(top_srcdir)/src -I$(srcdir) $(python_cpp_pkg_flags)
-libbenchmark_messages_la-libbenchmark_messages_la.$(OBJEXT): $(benchmarks_protoc_outputs_header) $(benchmarks_protoc_outputs_proto2_header) $(benchmarks_protoc_outputs) $(benchmarks_protoc_outputs_proto2)
+libbenchmark_messages_la_CPPFLAGS = -I$(top_srcdir)/src -I$(srcdir)/cpp $(python_cpp_pkg_flags)
+libbenchmark_messages_la-python_benchmark_messages.$(OBJEXT): $(benchmarks_protoc_outputs_header) $(benchmarks_protoc_outputs_proto2_header) $(benchmarks_protoc_outputs) $(benchmarks_protoc_outputs_proto2)
 nodist_libbenchmark_messages_la_SOURCES =                         \
 	$(benchmarks_protoc_outputs)                                    \
 	$(benchmarks_protoc_outputs_proto2)                             \
@@ -191,8 +191,8 @@ python-pure-python-benchmark: python_add_init
 	@echo export LD_LIBRARY_PATH=$(top_srcdir)/src/libprotobuf.la >> python-pure-python-benchmark
 	@echo export DYLD_LIBRARY_PATH=$(top_srcdir)/src/libprotobuf.la >> python-pure-python-benchmark
 	@echo export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=\'python\' >> python-pure-python-benchmark
-	@echo cp py_benchmark.py tmp >> python-pure-python-benchmark
-	@echo python tmp/py_benchmark.py false '$$@' >> python-pure-python-benchmark
+	@echo cp python/py_benchmark.py tmp >> python-pure-python-benchmark
+	@echo python tmp/py_benchmark.py '$$@' >> python-pure-python-benchmark
 	@chmod +x python-pure-python-benchmark
 
 python-cpp-reflection-benchmark: python_add_init
@@ -201,8 +201,8 @@ python-cpp-reflection-benchmark: python_add_init
 	@echo export LD_LIBRARY_PATH=$(top_srcdir)/src/libprotobuf.la >> python-cpp-reflection-benchmark
 	@echo export DYLD_LIBRARY_PATH=$(top_srcdir)/src/libprotobuf.la >> python-cpp-reflection-benchmark
 	@echo export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=\'cpp\' >> python-cpp-reflection-benchmark
-	@echo cp py_benchmark.py tmp >> python-cpp-reflection-benchmark
-	@echo python tmp/py_benchmark.py false '$$@' >> python-cpp-reflection-benchmark
+	@echo cp python/py_benchmark.py tmp >> python-cpp-reflection-benchmark
+	@echo python tmp/py_benchmark.py '$$@' >> python-cpp-reflection-benchmark
 	@chmod +x python-cpp-reflection-benchmark
 
 python-cpp-generated-code-benchmark: python_add_init libbenchmark_messages.la
@@ -211,8 +211,8 @@ python-cpp-generated-code-benchmark: python_add_init libbenchmark_messages.la
 	@echo export LD_LIBRARY_PATH=$(top_srcdir)/src/libprotobuf.la >> python-cpp-generated-code-benchmark
 	@echo export DYLD_LIBRARY_PATH=$(top_srcdir)/src/libprotobuf.la >> python-cpp-generated-code-benchmark
 	@echo export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=\'cpp\' >> python-cpp-generated-code-benchmark
-	@echo cp py_benchmark.py tmp >> python-cpp-generated-code-benchmark
-	@echo python tmp/py_benchmark.py true '$$@' >> python-cpp-generated-code-benchmark
+	@echo cp python/py_benchmark.py tmp >> python-cpp-generated-code-benchmark
+	@echo python tmp/py_benchmark.py --cpp_generated '$$@' >> python-cpp-generated-code-benchmark
 	@chmod +x python-cpp-generated-code-benchmark
 
 python-pure-python: python-pure-python-benchmark
@@ -287,9 +287,13 @@ endif
 go-benchmark: go_protoc_middleman go_protoc_middleman2 
 	@echo "Writing shortcut script go-benchmark..."
 	@echo '#! /bin/sh' > go-benchmark
-	@echo 'mkdir tmp_cc && mv *.cc tmp_cc' >> go-benchmark
-	@echo 'go test -bench=. -- $$@' >> go-benchmark
-	@echo 'mv tmp_cc/* . && rm -rf tmp_cc' >> go-benchmark
+	@echo 'cd go' >> go-benchmark
+	@echo 'all_data=""' >> go-benchmark
+	@echo 'conf=()' >> go-benchmark
+	@echo 'data_files=()' >> go-benchmark
+	@echo 'for arg in $$@; do if [[ $${arg:0:1} == "-" ]]; then conf+=($$arg); else data_files+=("../$$arg"); fi; done' >> go-benchmark
+	@echo 'go test -bench=. $${conf[*]} -- $${data_files[*]}' >> go-benchmark
+	@echo 'cd ..' >> go-benchmark
 	@chmod +x go-benchmark
 
 go: go_protoc_middleman go_protoc_middleman2 go-benchmark
diff --git a/benchmarks/README.md b/benchmarks/README.md
index 74c042d8..8460a18a 100644
--- a/benchmarks/README.md
+++ b/benchmarks/README.md
@@ -107,49 +107,51 @@ $ make python-cpp-generated-code
 $ make go
 ```
 
-To run a specific dataset:
+To run a specific dataset or run with specific options:
 
 ### Java:
 
 ```
 $ make java-benchmark
-$ ./java-benchmark $(specific generated dataset file name) [-- $(caliper option)]
+$ ./java-benchmark $(specific generated dataset file name) [$(caliper options)]
 ```
 
 ### CPP:
 
 ```
 $ make cpp-benchmark
-$ ./cpp-benchmark $(specific generated dataset file name)
+$ ./cpp-benchmark $(specific generated dataset file name) [$(benchmark options)]
 ```
 
 ### Python:
 
+For Python benchmark we have `--json` for outputing the json result
+
 #### Pure Python:
 
 ```
 $ make python-pure-python-benchmark
-$ ./python-pure-python-benchmark $(specific generated dataset file name)
+$ ./python-pure-python-benchmark [--json] $(specific generated dataset file name)
 ```
 
 #### CPP reflection:
 
 ```
 $ make python-cpp-reflection-benchmark
-$ ./python-cpp-reflection-benchmark $(specific generated dataset file name)
+$ ./python-cpp-reflection-benchmark [--json] $(specific generated dataset file name)
 ```
 
 #### CPP generated code:
 
 ```
 $ make python-cpp-generated-code-benchmark
-$ ./python-cpp-generated-code-benchmark $(specific generated dataset file name)
+$ ./python-cpp-generated-code-benchmark [--json] $(specific generated dataset file name)
 ```
 
 ### Go:
 ```
 $ make go-benchmark
-$ ./go-benchmark $(specific generated dataset file name)
+$ ./go-benchmark $(specific generated dataset file name) [go testing options]
 ```
 
 
diff --git a/benchmarks/cpp_benchmark.cc b/benchmarks/cpp/cpp_benchmark.cc
index f8b55291..f8b55291 100644
--- a/benchmarks/cpp_benchmark.cc
+++ b/benchmarks/cpp/cpp_benchmark.cc
diff --git a/benchmarks/go_benchmark_test.go b/benchmarks/go/go_benchmark_test.go
index e747465e..8c741b71 100644
--- a/benchmarks/go_benchmark_test.go
+++ b/benchmarks/go/go_benchmark_test.go
@@ -1,12 +1,12 @@
 package main
 
 import (
-	benchmarkWrapper "./tmp"
-	googleMessage1Proto2 "./tmp/datasets/google_message1/proto2"
-	googleMessage1Proto3 "./tmp/datasets/google_message1/proto3"
-	googleMessage2 "./tmp/datasets/google_message2"
-	googleMessage3 "./tmp/datasets/google_message3"
-	googleMessage4 "./tmp/datasets/google_message4"
+	benchmarkWrapper "../tmp"
+	googleMessage1Proto2 "../tmp/datasets/google_message1/proto2"
+	googleMessage1Proto3 "../tmp/datasets/google_message1/proto3"
+	googleMessage2 "../tmp/datasets/google_message2"
+	googleMessage3 "../tmp/datasets/google_message3"
+	googleMessage4 "../tmp/datasets/google_message4"
 	"flag"
 	"github.com/golang/protobuf/proto"
 	"io/ioutil"
diff --git a/benchmarks/java/src/main/java/com/google/protobuf/ProtoBench.java b/benchmarks/java/src/main/java/com/google/protobuf/ProtoBench.java
index f35b180a..02503905 100755
--- a/benchmarks/java/src/main/java/com/google/protobuf/ProtoBench.java
+++ b/benchmarks/java/src/main/java/com/google/protobuf/ProtoBench.java
@@ -69,40 +69,18 @@ public class ProtoBench {
       System.err.println("input data is in the format of \"benchmarks.proto\"");
       System.exit(1);
     }
-    boolean success = true;
-    for (int i = 0; i < args.length; i++) {
-      success &= runTest(args[i]);
-    }
-    System.exit(success ? 0 : 1);
+    
+    System.exit(runTest(args) ? 0 : 1);
   }
   
- 
-  /**
-   * Runs a single test with specific test data. Error messages are displayed to stderr, 
-   * and the return value indicates general success/failure.
-   */
-  public static boolean runTest(String file) {
-    byte[] inputData;
-    BenchmarkDataset benchmarkDataset;
-    try {
-      inputData = readAllBytes(file);
-      benchmarkDataset = BenchmarkDataset.parseFrom(inputData);
-    } catch (IOException e) {
-      System.err.println("Unable to get input data");
-      return false;
-    }
-    List<String> argsList = getCaliperOption(benchmarkDataset);
-    if (argsList == null) {
-      System.err.println("Unable to get default message " + benchmarkDataset.getMessageName());
-      return false;
-    }
-    argsList.add("-DdataFile=" + file);
+  public static boolean runTest(String args[]) {
+    List<String> argsList = getCaliperOption(args);
     argsList.add("com.google.protobuf.ProtoCaliperBenchmark");
     
     try {
-      String args[] = new String[argsList.size()];
-      argsList.toArray(args);
-      CaliperMain.exitlessMain(args,
+      String newArgs[] = new String[argsList.size()];
+      argsList.toArray(newArgs);
+      CaliperMain.exitlessMain(newArgs,
           new PrintWriter(System.out, true), new PrintWriter(System.err, true));
     } catch (Exception e) {
       System.err.println("Error: " + e.getMessage());
@@ -110,54 +88,22 @@ public class ProtoBench {
       e.printStackTrace(System.err);
       return false;
     }
-    try {
-      double mininumScale = 0;
-      // If the file not exist, this will throw IOException, which won't print the warning 
-      // information below.
-      Scanner scanner = new Scanner(new String(readAllBytes("JavaBenchmarkWarning.txt")));
-      while (scanner.hasNext()) {
-        mininumScale = Math.max(mininumScale, scanner.nextDouble());
-      }
-      scanner.close();
-      
-      System.out.println(
-         "WARNING: This benchmark's whole iterations are not enough, consider to config caliper to "
-        + "run for more time to make the result more convincing. You may change the configure "
-        + "code in com.google.protobuf.ProtoBench.getCaliperOption() of benchmark " 
-        + benchmarkDataset.getMessageName()
-        + " to run for more time. e.g. Change the value of "
-        + "instrument.runtime.options.timingInterval or value of "
-        + "instrument.runtime.options.measurements to be at least "
-        + Math.round(mininumScale * 10 + 1) / 10.0 
-        + " times of before, then build and run the benchmark again\n");
-      Files.deleteIfExists(Paths.get("JavaBenchmarkWarning.txt"));
-    } catch (IOException e) {
-      // The IOException here should be file not found, which means there's no warning generated by
-      // The benchmark, so this IOException should be discarded.
-    }
     return true;
   }
   
-  
-  private static List<String> getCaliperOption(final BenchmarkDataset benchmarkDataset) {
+  private static List<String> getCaliperOption(String args[]) {
     List<String> temp = new ArrayList<String>();
-    if (benchmarkDataset.getMessageName().equals("benchmarks.proto3.GoogleMessage1")) {
-    } else if (benchmarkDataset.getMessageName().equals("benchmarks.proto2.GoogleMessage1")) {
-    } else if (benchmarkDataset.getMessageName().equals("benchmarks.proto2.GoogleMessage2")) {
-    } else if (benchmarkDataset.getMessageName().
-        equals("benchmarks.google_message3.GoogleMessage3")) {
-      temp.add("-Cinstrument.runtime.options.timingInterval=3000ms");
-      temp.add("-Cinstrument.runtime.options.measurements=20");
-    } else if (benchmarkDataset.getMessageName().
-        equals("benchmarks.google_message4.GoogleMessage4")) {
-      temp.add("-Cinstrument.runtime.options.timingInterval=1500ms");
-      temp.add("-Cinstrument.runtime.options.measurements=20");
-    } else {
-      return null;
-    }
-    
     temp.add("-i");
     temp.add("runtime");
+    String files = "";
+    for (int i = 0; i < args.length; i++) {
+      if (args[i].charAt(0) == '-') {
+        temp.add(args[i]);
+      } else {
+        files += (files.equals("") ? "" : ",") + args[i];
+      }
+    }
+    temp.add("-DdataFile=" + files);
     temp.add("-b");
     String benchmarkNames = "serializeToByteString,serializeToByteArray,serializeToMemoryStream"
        + ",deserializeFromByteString,deserializeFromByteArray,deserializeFromMemoryStream";
@@ -165,14 +111,4 @@ public class ProtoBench {
     
     return temp;
   }
-  
-  public static byte[] readAllBytes(String filename) throws IOException {
-    if (filename.equals("")) {
-      return new byte[0];
-    }
-    RandomAccessFile file = new RandomAccessFile(new File(filename), "r");
-    byte[] content = new byte[(int) file.length()];
-    file.readFully(content);
-    return content;
-  }
 }
diff --git a/benchmarks/java/src/main/java/com/google/protobuf/ProtoCaliperBenchmark.java b/benchmarks/java/src/main/java/com/google/protobuf/ProtoCaliperBenchmark.java
index b50c6585..94568aea 100755
--- a/benchmarks/java/src/main/java/com/google/protobuf/ProtoCaliperBenchmark.java
+++ b/benchmarks/java/src/main/java/com/google/protobuf/ProtoCaliperBenchmark.java
@@ -99,7 +99,6 @@ public class ProtoCaliperBenchmark {
   private List<ByteArrayInputStream> inputStreamList;
   private List<ByteString> inputStringList;
   private List<Message> sampleMessageList;
-  private long counter;
 
   private BenchmarkMessageType getMessageType() throws IOException {
     if (benchmarkDataset.getMessageName().equals("benchmarks.proto3.GoogleMessage1")) {
@@ -149,8 +148,6 @@ public class ProtoCaliperBenchmark {
       sampleMessageList.add(
           defaultMessage.newBuilderForType().mergeFrom(singleInputData, extensions).build());
     }
-    
-    counter = 0;
   }
   
   
@@ -160,8 +157,9 @@ public class ProtoCaliperBenchmark {
       return;
     }
     for (int i = 0; i < reps; i++) {
-      sampleMessageList.get((int) (counter % sampleMessageList.size())).toByteString();
-      counter++;
+      for (int j = 0; j < sampleMessageList.size(); j++) {
+        sampleMessageList.get(j).toByteString();        
+      }
     }
   }
   
@@ -171,8 +169,9 @@ public class ProtoCaliperBenchmark {
       return;
     }
     for (int i = 0; i < reps; i++) {
-      sampleMessageList.get((int) (counter % sampleMessageList.size())).toByteArray();
-      counter++;
+      for (int j = 0; j < sampleMessageList.size(); j++) {
+        sampleMessageList.get(j).toByteArray();  
+      }
     }
   }
   
@@ -182,9 +181,10 @@ public class ProtoCaliperBenchmark {
       return;
     }
     for (int i = 0; i < reps; i++) {
-      ByteArrayOutputStream output = new ByteArrayOutputStream();
-      sampleMessageList.get((int) (counter % sampleMessageList.size())).writeTo(output);
-      counter++;
+      for (int j = 0; j < sampleMessageList.size(); j++) {
+        ByteArrayOutputStream output = new ByteArrayOutputStream();
+        sampleMessageList.get(j).writeTo(output); 
+      }
     }
   }
   
@@ -194,9 +194,10 @@ public class ProtoCaliperBenchmark {
       return;
     }
     for (int i = 0; i < reps; i++) {
-      benchmarkMessageType.getDefaultInstance().getParserForType().parseFrom(
-          inputStringList.get((int) (counter % inputStringList.size())), extensions);
-      counter++;
+      for (int j = 0; j < inputStringList.size(); j++) {
+        benchmarkMessageType.getDefaultInstance().getParserForType().parseFrom(
+          inputStringList.get(j), extensions);
+      }
     }
   }
   
@@ -206,9 +207,10 @@ public class ProtoCaliperBenchmark {
       return;
     }
     for (int i = 0; i < reps; i++) {
-      benchmarkMessageType.getDefaultInstance().getParserForType().parseFrom(
-          inputDataList.get((int) (counter % inputDataList.size())), extensions);
-      counter++;
+      for (int j = 0; j < inputDataList.size(); j++) {
+        benchmarkMessageType.getDefaultInstance().getParserForType().parseFrom(
+          inputDataList.get(j), extensions);
+      }
     }
   }
   
@@ -218,27 +220,11 @@ public class ProtoCaliperBenchmark {
       return;
     }
     for (int i = 0; i < reps; i++) {
-      benchmarkMessageType.getDefaultInstance().getParserForType().parseFrom(
-          inputStreamList.get((int) (counter % inputStreamList.size())), extensions);
-      inputStreamList.get((int) (counter % inputStreamList.size())).reset();
-      counter++;
-    }
-  }
-  
-  @AfterExperiment
-  void checkCounter() throws IOException {
-    if (counter == 1) {
-      // Dry run
-      return;
-    }
-    if (benchmarkDataset.getPayloadCount() != 1 
-        && counter < benchmarkDataset.getPayloadCount() * 10L) {
-      BufferedWriter writer = new BufferedWriter(new FileWriter("JavaBenchmarkWarning.txt", true));
-      // If the total number of non-warmup reps is smaller than 100 times of the total number of 
-      // datasets, then output the scale that need to multiply to the configuration (either extend
-      // the running time for one timingInterval or run for more measurements). 
-      writer.append(1.0 * benchmarkDataset.getPayloadCount() * 10L / counter + " ");
-      writer.close();
+      for (int j = 0; j < inputStreamList.size(); j++) {
+        benchmarkMessageType.getDefaultInstance().getParserForType().parseFrom(
+            inputStreamList.get(j), extensions);
+        inputStreamList.get(j).reset();
+      }
     }
   }
 }
diff --git a/benchmarks/py_benchmark.py b/benchmarks/python/py_benchmark.py
index ba7a3470..a551ba6d 100755
--- a/benchmarks/py_benchmark.py
+++ b/benchmarks/python/py_benchmark.py
@@ -2,21 +2,35 @@ import sys
 import os
 import timeit
 import math
+import argparse
 import fnmatch
-
+import json
+
+parser = argparse.ArgumentParser(description="Python protobuf benchmark")
+parser.add_argument("data_files", metavar="dataFile", nargs="+", 
+                    help="testing data files.")
+parser.add_argument("--json", action="store_const", dest="json",
+                    const="yes", default="no",
+                    help="Whether to output json results")
+parser.add_argument("--behavior_prefix", dest="behavior_prefix",
+                    help="The output json format's behavior's name's prefix",
+                    default="")
+# BEGIN CPP GENERATED MESSAGE
+parser.add_argument("--cpp_generated", action="store_const",
+                    dest="cpp_generated", const="yes", default="no",
+                    help="Whether to link generated code library")
+# END CPP GENERATED MESSAGE
+args = parser.parse_args()
 # BEGIN CPP GENERATED MESSAGE
 # CPP generated code must be linked before importing the generated Python code
 # for the descriptor can be found in the pool
-if len(sys.argv) < 2:
-  raise IOError("Need string argument \"true\" or \"false\" for whether to use cpp generated code")
-if sys.argv[1] == "true":
+if args.cpp_generated != "no":
   sys.path.append( os.path.dirname( os.path.dirname( os.path.abspath(__file__) ) ) + "/.libs" )
   import libbenchmark_messages
   sys.path.append( os.path.dirname( os.path.dirname( os.path.abspath(__file__) ) ) + "/tmp" )
-elif sys.argv[1] != "false":
-  raise IOError("Need string argument \"true\" or \"false\" for whether to use cpp generated code")
 # END CPP GENERATED MESSAGE
 
+
 import datasets.google_message1.proto2.benchmark_message1_proto2_pb2 as benchmark_message1_proto2_pb2
 import datasets.google_message1.proto3.benchmark_message1_proto3_pb2 as benchmark_message1_proto3_pb2
 import datasets.google_message2.benchmark_message2_pb2 as benchmark_message2_pb2
@@ -26,19 +40,24 @@ import benchmarks_pb2 as benchmarks_pb2
 
 
 def run_one_test(filename):
-  data = open(os.path.dirname(sys.argv[0]) + "/../" + filename).read()
+  data = open(filename).read()
   benchmark_dataset = benchmarks_pb2.BenchmarkDataset()
   benchmark_dataset.ParseFromString(data)
   benchmark_util = Benchmark(full_iteration=len(benchmark_dataset.payload),
                              module="py_benchmark",
                              setup_method="init")
-  print "Message %s of dataset file %s" % \
-    (benchmark_dataset.message_name, filename)
+  result={}
+  result["filename"] =  filename
+  result["message_name"] =  benchmark_dataset.message_name
+  result["benchmarks"] = {}
   benchmark_util.set_test_method("parse_from_benchmark")
-  print benchmark_util.run_benchmark(setup_method_args='"%s"' % (filename))
+  result["benchmarks"][args.behavior_prefix + "_parse_from_benchmark"] = \
+    benchmark_util.run_benchmark(setup_method_args='"%s"' % (filename))
   benchmark_util.set_test_method("serialize_to_benchmark")
-  print benchmark_util.run_benchmark(setup_method_args='"%s"' % (filename))
-  print ""
+  result["benchmarks"][args.behavior_prefix + "_serialize_to_benchmark"] = \
+    benchmark_util.run_benchmark(setup_method_args='"%s"' % (filename))
+  return result
+
 
 def init(filename):
   global benchmark_dataset, message_class, message_list, counter
@@ -66,11 +85,13 @@ def init(filename):
     temp.ParseFromString(one_payload)
     message_list.append(temp)
 
+
 def parse_from_benchmark():
   global counter, message_class, benchmark_dataset
   m = message_class().ParseFromString(benchmark_dataset.payload[counter % len(benchmark_dataset.payload)])
   counter = counter + 1
 
+
 def serialize_to_benchmark():
   global counter, message_list, message_class
   s = message_list[counter % len(benchmark_dataset.payload)].SerializeToString()
@@ -108,11 +129,22 @@ class Benchmark:
     t = timeit.timeit(stmt="%s(%s)" % (self.test_method, test_method_args),
                       setup=self.full_setup_code(setup_method_args),
                       number=reps);
-    return "Average time for %s: %.2f ns" % \
-      (self.test_method, 1.0 * t / reps * (10 ** 9))
-
+    return 1.0 * t / reps * (10 ** 9)
+  
 
 if __name__ == "__main__":
-  for i in range(2, len(sys.argv)):
-    run_one_test(sys.argv[i])
-
+  results = []
+  for file in args.data_files:
+    results.append(run_one_test(file))
+  
+  if args.json != "no":
+    print json.dumps(results)
+  else:
+    for result in results:
+      print "Message %s of dataset file %s" % \
+          (result["message_name"], result["filename"])
+      print "Average time for parse_from_benchmark: %.2f ns" % \
+          (result["benchmarks"]["parse_from_benchmark"])
+      print "Average time for serialize_to_benchmark: %.2f ns" % \
+          (result["benchmarks"]["serialize_to_benchmark"])
+      print ""
diff --git a/benchmarks/python_benchmark_messages.cc b/benchmarks/python/python_benchmark_messages.cc
index 55242a2a..ded16fe9 100644
--- a/benchmarks/python_benchmark_messages.cc
+++ b/benchmarks/python/python_benchmark_messages.cc
@@ -1,8 +1,8 @@
 #include <Python.h>
 
 #include "benchmarks.pb.h"
-#include "datasets/google_message1/benchmark_message1_proto2.pb.h"
-#include "datasets/google_message1/benchmark_message1_proto3.pb.h"
+#include "datasets/google_message1/proto2/benchmark_message1_proto2.pb.h"
+#include "datasets/google_message1/proto3/benchmark_message1_proto3.pb.h"
 #include "datasets/google_message2/benchmark_message2.pb.h"
 #include "datasets/google_message3/benchmark_message3.pb.h"
 #include "datasets/google_message4/benchmark_message4.pb.h"
diff --git a/benchmarks/util/__init__.py b/benchmarks/util/__init__.py
new file mode 100644
index 00000000..e69de29b
--- /dev/null
+++ b/benchmarks/util/__init__.py
diff --git a/benchmarks/util/big_query_utils.py b/benchmarks/util/big_query_utils.py
new file mode 100755
index 00000000..14105aa6
--- /dev/null
+++ b/benchmarks/util/big_query_utils.py
@@ -0,0 +1,188 @@
+#!/usr/bin/env python2.7
+
+import argparse
+import json
+import uuid
+import httplib2
+
+from apiclient import discovery
+from apiclient.errors import HttpError
+from oauth2client.client import GoogleCredentials
+
+# 30 days in milliseconds
+_EXPIRATION_MS = 30 * 24 * 60 * 60 * 1000
+NUM_RETRIES = 3
+
+
+def create_big_query():
+    """Authenticates with cloud platform and gets a BiqQuery service object
+  """
+    creds = GoogleCredentials.get_application_default()
+    return discovery.build(
+        'bigquery', 'v2', credentials=creds, cache_discovery=False)
+
+
+def create_dataset(biq_query, project_id, dataset_id):
+    is_success = True
+    body = {
+        'datasetReference': {
+            'projectId': project_id,
+            'datasetId': dataset_id
+        }
+    }
+
+    try:
+        dataset_req = biq_query.datasets().insert(
+            projectId=project_id, body=body)
+        dataset_req.execute(num_retries=NUM_RETRIES)
+    except HttpError as http_error:
+        if http_error.resp.status == 409:
+            print 'Warning: The dataset %s already exists' % dataset_id
+        else:
+            # Note: For more debugging info, print "http_error.content"
+            print 'Error in creating dataset: %s. Err: %s' % (dataset_id,
+                                                              http_error)
+            is_success = False
+    return is_success
+
+
+def create_table(big_query, project_id, dataset_id, table_id, table_schema,
+                 description):
+    fields = [{
+        'name': field_name,
+        'type': field_type,
+        'description': field_description
+    } for (field_name, field_type, field_description) in table_schema]
+    return create_table2(big_query, project_id, dataset_id, table_id, fields,
+                         description)
+
+
+def create_partitioned_table(big_query,
+                             project_id,
+                             dataset_id,
+                             table_id,
+                             table_schema,
+                             description,
+                             partition_type='DAY',
+                             expiration_ms=_EXPIRATION_MS):
+    """Creates a partitioned table. By default, a date-paritioned table is created with
+  each partition lasting 30 days after it was last modified.
+  """
+    fields = [{
+        'name': field_name,
+        'type': field_type,
+        'description': field_description
+    } for (field_name, field_type, field_description) in table_schema]
+    return create_table2(big_query, project_id, dataset_id, table_id, fields,
+                         description, partition_type, expiration_ms)
+
+
+def create_table2(big_query,
+                  project_id,
+                  dataset_id,
+                  table_id,
+                  fields_schema,
+                  description,
+                  partition_type=None,
+                  expiration_ms=None):
+    is_success = True
+
+    body = {
+        'description': description,
+        'schema': {
+            'fields': fields_schema
+        },
+        'tableReference': {
+            'datasetId': dataset_id,
+            'projectId': project_id,
+            'tableId': table_id
+        }
+    }
+
+    if partition_type and expiration_ms:
+        body["timePartitioning"] = {
+            "type": partition_type,
+            "expirationMs": expiration_ms
+        }
+
+    try:
+        table_req = big_query.tables().insert(
+            projectId=project_id, datasetId=dataset_id, body=body)
+        res = table_req.execute(num_retries=NUM_RETRIES)
+        print 'Successfully created %s "%s"' % (res['kind'], res['id'])
+    except HttpError as http_error:
+        if http_error.resp.status == 409:
+            print 'Warning: Table %s already exists' % table_id
+        else:
+            print 'Error in creating table: %s. Err: %s' % (table_id,
+                                                            http_error)
+            is_success = False
+    return is_success
+
+
+def patch_table(big_query, project_id, dataset_id, table_id, fields_schema):
+    is_success = True
+
+    body = {
+        'schema': {
+            'fields': fields_schema
+        },
+        'tableReference': {
+            'datasetId': dataset_id,
+            'projectId': project_id,
+            'tableId': table_id
+        }
+    }
+
+    try:
+        table_req = big_query.tables().patch(
+            projectId=project_id,
+            datasetId=dataset_id,
+            tableId=table_id,
+            body=body)
+        res = table_req.execute(num_retries=NUM_RETRIES)
+        print 'Successfully patched %s "%s"' % (res['kind'], res['id'])
+    except HttpError as http_error:
+        print 'Error in creating table: %s. Err: %s' % (table_id, http_error)
+        is_success = False
+    return is_success
+
+
+def insert_rows(big_query, project_id, dataset_id, table_id, rows_list):
+    is_success = True
+    body = {'rows': rows_list}
+    try:
+        insert_req = big_query.tabledata().insertAll(
+            projectId=project_id,
+            datasetId=dataset_id,
+            tableId=table_id,
+            body=body)
+        res = insert_req.execute(num_retries=NUM_RETRIES)
+        if res.get('insertErrors', None):
+            print 'Error inserting rows! Response: %s' % res
+            is_success = False
+    except HttpError as http_error:
+        print 'Error inserting rows to the table %s' % table_id
+        is_success = False
+
+    return is_success
+
+
+def sync_query_job(big_query, project_id, query, timeout=5000):
+    query_data = {'query': query, 'timeoutMs': timeout}
+    query_job = None
+    try:
+        query_job = big_query.jobs().query(
+            projectId=project_id,
+            body=query_data).execute(num_retries=NUM_RETRIES)
+    except HttpError as http_error:
+        print 'Query execute job failed with error: %s' % http_error
+        print http_error.content
+    return query_job
+
+
+    # List of (column name, column type, description) tuples
+def make_row(unique_row_id, row_values_dict):
+    """row_values_dict is a dictionary of column name and column value.
+  """
+    return {'insertId': unique_row_id, 'json': row_values_dict}
diff --git a/benchmarks/util/run_and_upload.py b/benchmarks/util/run_and_upload.py
new file mode 100755
index 00000000..ae22a668
--- /dev/null
+++ b/benchmarks/util/run_and_upload.py
@@ -0,0 +1,290 @@
+import argparse
+import os
+import re
+import copy
+import uuid
+import calendar
+import time
+import big_query_utils
+import datetime
+import json
+# This import depends on the automake rule protoc_middleman, please make sure
+# protoc_middleman has been built before run this file.
+import os.path, sys
+sys.path.append(os.path.join(os.path.dirname(os.path.realpath(__file__)), os.pardir))
+import tmp.benchmarks_pb2 as benchmarks_pb2
+from click.types import STRING
+
+_PROJECT_ID = 'grpc-testing'
+_DATASET = 'protobuf_benchmark_result'
+_TABLE = 'opensource_result_v1'
+_NOW = "%d%02d%02d" % (datetime.datetime.now().year,
+                       datetime.datetime.now().month,
+                       datetime.datetime.now().day)
+
+file_size_map = {}
+
+def get_data_size(file_name):
+  if file_name in file_size_map:
+    return file_size_map[file_name]
+  benchmark_dataset = benchmarks_pb2.BenchmarkDataset()
+  benchmark_dataset.ParseFromString(
+      open(os.path.dirname(os.path.abspath(__file__)) + "/../" + file_name).read())
+  size = 0
+  count = 0
+  for payload in benchmark_dataset.payload:
+    size += len(payload)
+    count += 1
+  file_size_map[file_name] = (size, 1.0 * size / count)
+  return size, 1.0 * size / count
+
+
+def extract_file_name(file_name):
+  name_list = re.split("[/\.]", file_name)
+  short_file_name = ""
+  for name in name_list:
+    if name[:14] == "google_message":
+      short_file_name = name
+  return short_file_name
+
+
+cpp_result = []
+python_result = []
+java_result = []
+go_result = []
+
+
+# CPP results example:
+# [ 
+#   "benchmarks": [ 
+#     {
+#       "bytes_per_second": int,
+#       "cpu_time": int,
+#       "name: string,
+#       "time_unit: string,
+#       ...
+#     },
+#     ... 
+#   ],
+#   ... 
+# ]
+def parse_cpp_result(filename):
+  global cpp_result
+  if filename == "":
+    return
+  if filename[0] != '/':
+    filename = os.path.dirname(os.path.abspath(__file__)) + '/' + filename
+  with open(filename) as f:
+    results = json.loads(f.read())
+    for benchmark in results["benchmarks"]:
+      data_filename = "".join(
+          re.split("(_parse_|_serialize)", benchmark["name"])[0])
+      behavior = benchmark["name"][len(data_filename) + 1:]
+      cpp_result.append({
+        "language": "cpp",
+        "dataFileName": data_filename,
+        "behavior": behavior,
+        "throughput": benchmark["bytes_per_second"] / 2.0 ** 20
+      })
+
+
+# Python results example:
+# [ 
+#   [ 
+#     {
+#       "filename": string,
+#       "benchmarks": {
+#         behavior: results, 
+#         ...
+#       },
+#       "message_name": STRING
+#     },
+#     ... 
+#   ], #pure-python
+#   ... 
+# ]
+def parse_python_result(filename):
+  global python_result
+  if filename == "":
+    return
+  if filename[0] != '/':
+    filename = os.path.dirname(os.path.abspath(__file__)) + '/' + filename
+  with open(filename) as f:
+    results_list = json.loads(f.read())
+    for results in results_list:
+      for result in results:
+        _, avg_size = get_data_size(result["filename"])
+        for behavior in result["benchmarks"]:
+          python_result.append({
+            "language": "python",
+            "dataFileName": extract_file_name(result["filename"]),
+            "behavior": behavior,
+            "throughput": avg_size /
+                          result["benchmarks"][behavior] * 1e9 / 2 ** 20
+          })
+
+
+# Java results example:
+# [ 
+#   {
+#     "id": string,
+#     "instrumentSpec": {...},
+#     "measurements": [
+#       {
+#         "weight": float,
+#         "value": {
+#           "magnitude": float,
+#           "unit": string
+#         },
+#         ...
+#       },
+#       ...
+#     ],
+#     "run": {...},
+#     "scenario": {
+#       "benchmarkSpec": {
+#         "methodName": string,
+#         "parameters": {
+#            defined parameters in the benchmark: parameters value
+#         },
+#         ...
+#       },
+#       ...
+#     }
+#     
+#   }, 
+#   ... 
+# ]
+def parse_java_result(filename):
+  global average_bytes_per_message, java_result
+  if filename == "":
+    return
+  if filename[0] != '/':
+    filename = os.path.dirname(os.path.abspath(__file__)) + '/' + filename
+  with open(filename) as f:
+    results = json.loads(f.read())
+    for result in results:
+      total_weight = 0
+      total_value = 0
+      for measurement in result["measurements"]:
+        total_weight += measurement["weight"]
+        total_value += measurement["value"]["magnitude"]
+      avg_time = total_value * 1.0 / total_weight
+      total_size, _ = get_data_size(
+          result["scenario"]["benchmarkSpec"]["parameters"]["dataFile"])
+      java_result.append({
+        "language": "java",
+        "throughput": total_size / avg_time * 1e9 / 2 ** 20,
+        "behavior": result["scenario"]["benchmarkSpec"]["methodName"],
+        "dataFileName": extract_file_name(
+            result["scenario"]["benchmarkSpec"]["parameters"]["dataFile"])
+      })
+
+
+# Go benchmark results:
+#
+# goos: linux
+# goarch: amd64
+# Benchmark/.././datasets/google_message2/dataset.google_message2.pb/Unmarshal-12               3000      705784 ns/op
+# Benchmark/.././datasets/google_message2/dataset.google_message2.pb/Marshal-12                 2000      634648 ns/op
+# Benchmark/.././datasets/google_message2/dataset.google_message2.pb/Size-12                    5000      244174 ns/op
+# Benchmark/.././datasets/google_message2/dataset.google_message2.pb/Clone-12                    300     4120954 ns/op
+# Benchmark/.././datasets/google_message2/dataset.google_message2.pb/Merge-12                    300     4108632 ns/op
+# PASS
+# ok    _/usr/local/google/home/yilunchong/mygit/protobuf/benchmarks  124.173s
+def parse_go_result(filename):
+  global go_result
+  if filename == "":
+    return
+  if filename[0] != '/':
+    filename = os.path.dirname(os.path.abspath(__file__)) + '/' + filename
+  with open(filename) as f:
+    for line in f:
+      result_list = re.split("[\ \t]+", line)
+      if result_list[0][:9] != "Benchmark":
+        continue
+      first_slash_index = result_list[0].find('/')
+      last_slash_index = result_list[0].rfind('/')
+      full_filename = result_list[0][first_slash_index+4:last_slash_index] # delete ../ prefix
+      total_bytes, _ = get_data_size(full_filename)
+      behavior_with_suffix = result_list[0][last_slash_index+1:]
+      last_dash = behavior_with_suffix.rfind("-")
+      if last_dash == -1:
+        behavior = behavior_with_suffix
+      else:
+        behavior = behavior_with_suffix[:last_dash]
+      go_result.append({
+        "dataFilename": extract_file_name(full_filename),
+        "throughput": total_bytes / float(result_list[2]) * 1e9 / 2 ** 20,
+        "behavior": behavior,
+        "language": "go"
+      })
+
+
+def get_metadata():
+  build_number = os.getenv('BUILD_NUMBER')
+  build_url = os.getenv('BUILD_URL')
+  job_name = os.getenv('JOB_NAME')
+  git_commit = os.getenv('GIT_COMMIT')
+  # actual commit is the actual head of PR that is getting tested
+  git_actual_commit = os.getenv('ghprbActualCommit')
+
+  utc_timestamp = str(calendar.timegm(time.gmtime()))
+  metadata = {'created': utc_timestamp}
+
+  if build_number:
+    metadata['buildNumber'] = build_number
+  if build_url:
+    metadata['buildUrl'] = build_url
+  if job_name:
+    metadata['jobName'] = job_name
+  if git_commit:
+    metadata['gitCommit'] = git_commit
+  if git_actual_commit:
+    metadata['gitActualCommit'] = git_actual_commit
+
+  return metadata
+
+
+def upload_result(result_list, metadata):
+  for result in result_list:
+    new_result = copy.deepcopy(result)
+    new_result['metadata'] = metadata
+    bq = big_query_utils.create_big_query()
+    row = big_query_utils.make_row(str(uuid.uuid4()), new_result)
+    if not big_query_utils.insert_rows(bq, _PROJECT_ID, _DATASET,
+                                       _TABLE + "$" + _NOW,
+                                       [row]):
+      print 'Error when uploading result', new_result
+
+
+if __name__ == "__main__":
+  parser = argparse.ArgumentParser()
+  parser.add_argument("-cpp", "--cpp_input_file",
+                      help="The CPP benchmark result file's name",
+                      default="")
+  parser.add_argument("-java", "--java_input_file",
+                      help="The Java benchmark result file's name",
+                      default="")
+  parser.add_argument("-python", "--python_input_file",
+                      help="The Python benchmark result file's name",
+                      default="")
+  parser.add_argument("-go", "--go_input_file",
+                      help="The golang benchmark result file's name",
+                      default="")
+  args = parser.parse_args()
+
+  parse_cpp_result(args.cpp_input_file)
+  parse_python_result(args.python_input_file)
+  parse_java_result(args.java_input_file)
+  parse_go_result(args.go_input_file)
+
+  metadata = get_metadata()
+  print "uploading cpp results..."
+  upload_result(cpp_result, metadata)
+  print "uploading java results..."
+  upload_result(java_result, metadata)
+  print "uploading python results..."
+  upload_result(python_result, metadata)
+  print "uploading go results..."
+  upload_result(go_result, metadata)
diff --git a/examples/add_person_test.go b/examples/add_person_test.go
index 0507db6f..f8ba9338 100644
--- a/examples/add_person_test.go
+++ b/examples/add_person_test.go
@@ -30,10 +30,10 @@ unknown
 		t.Errorf("promptForAddress(%q) got %d, want ID %d", in, got.Id, 12345)
 	}
 	if got.Name != "Example Name" {
-		t.Errorf("promptForAddress(%q) => want name %q, got %q", "Example Name", got.Name)
+		t.Errorf("promptForAddress(%q) => want name %q, got %q", in, "Example Name", got.Name)
 	}
 	if got.Email != "name@example.com" {
-		t.Errorf("promptForAddress(%q) => want email %q, got %q", "name@example.com", got.Email)
+		t.Errorf("promptForAddress(%q) => want email %q, got %q", in, "name@example.com", got.Email)
 	}
 
 	want := []*pb.Person_PhoneNumber{
diff --git a/kokoro/linux/benchmark/build.sh b/kokoro/linux/benchmark/build.sh
new file mode 100755
index 00000000..b289290a
--- /dev/null
+++ b/kokoro/linux/benchmark/build.sh
@@ -0,0 +1,86 @@
+#!/bin/bash
+#
+# Change to repo root
+cd $(dirname $0)/../../..
+
+export OUTPUT_DIR=testoutput
+oldpwd=`pwd`
+
+# tcmalloc
+if [ ! -f gperftools/.libs/libtcmalloc.so ]; then
+  git clone https://github.com/gperftools/gperftools.git
+  cd gperftools
+  ./autogen.sh
+  ./configure
+  make -j8
+  cd ..
+fi
+
+# download datasets for benchmark
+cd benchmarks
+./download_data.sh
+datasets=`find . -type f -name "dataset.*.pb"`
+cd $oldpwd
+
+# build Python protobuf
+./autogen.sh
+./configure CXXFLAGS="-fPIC -O2 -fno-semantic-interposition"
+make -j8
+cd python
+python setup.py build --cpp_implementation
+pip install .
+
+# build and run Python benchmark
+cd ../benchmarks
+make python-pure-python-benchmark
+make python-cpp-reflection-benchmark
+make -j8 python-cpp-generated-code-benchmark
+echo "[" > tmp/python_result.json
+echo "benchmarking pure python..."
+./python-pure-python-benchmark --json --behavior_prefix="pure-python-benchmark" $datasets  >> tmp/python_result.json
+echo "," >> "tmp/python_result.json"
+echo "benchmarking python cpp reflection..."
+env LD_PRELOAD="$oldpwd/gperftools/.libs/libtcmalloc.so" ./python-cpp-reflection-benchmark --json --behavior_prefix="cpp-reflection-benchmark" $datasets  >> tmp/python_result.json
+echo "," >> "tmp/python_result.json"
+echo "benchmarking python cpp generated code..."
+env LD_PRELOAD="$oldpwd/gperftools/.libs/libtcmalloc.so" ./python-cpp-generated-code-benchmark --json --behavior_prefix="cpp-generated-code-benchmark" $datasets >> tmp/python_result.json
+echo "]" >> "tmp/python_result.json"
+cd $oldpwd
+
+# build CPP protobuf
+./configure
+make clean && make -j8
+
+# build CPP benchmark
+cd benchmarks
+mv tmp/python_result.json . && make clean && make -j8 cpp-benchmark && mv python_result.json tmp
+echo "benchmarking cpp..."
+env LD_PRELOAD="$oldpwd/gperftools/.libs/libtcmalloc.so" ./cpp-benchmark --benchmark_min_time=5.0 --benchmark_out_format=json --benchmark_out="tmp/cpp_result.json" $datasets
+cd $oldpwd
+
+# build go protobuf 
+export PATH="`pwd`/src:$PATH"
+export GOPATH="$HOME/gocode"
+mkdir -p "$GOPATH/src/github.com/google"
+rm -f "$GOPATH/src/github.com/google/protobuf"
+ln -s "`pwd`" "$GOPATH/src/github.com/google/protobuf"
+export PATH="$GOPATH/bin:$PATH"
+go get github.com/golang/protobuf/protoc-gen-go
+
+# build go benchmark
+cd benchmarks
+make go-benchmark
+echo "benchmarking go..."
+./go-benchmark $datasets > tmp/go_result.txt
+
+# build java benchmark
+make java-benchmark
+echo "benchmarking java..."
+./java-benchmark -Cresults.file.options.file="tmp/java_result.json" $datasets
+
+# upload result to bq
+make python_add_init
+python util/run_and_upload.py -cpp="../tmp/cpp_result.json" -java="../tmp/java_result.json" \
+    -python="../tmp/python_result.json" -go="../tmp/go_result.txt"
+
+cd $oldpwd
diff --git a/kokoro/linux/benchmark/continuous.cfg b/kokoro/linux/benchmark/continuous.cfg
new file mode 100755
index 00000000..a3558c65
--- /dev/null
+++ b/kokoro/linux/benchmark/continuous.cfg
@@ -0,0 +1,11 @@
+# Config file for running tests in Kokoro
+
+# Location of the build script in repository
+build_file: "protobuf/kokoro/linux/benchmark/build.sh"
+timeout_mins: 240 
+
+action {
+  define_artifacts {
+    regex: "**/sponge_log.xml"
+  }
+}
diff --git a/tests.sh b/tests.sh
index fbbfd448..feb90c59 100755
--- a/tests.sh
+++ b/tests.sh
@@ -27,6 +27,9 @@ internal_build_cpp() {
     export CXX="g++-4.8" CC="gcc-4.8"
   fi
 
+  # Initialize any submodules.
+  git submodule update --init --recursive
+
   ./autogen.sh
   ./configure CXXFLAGS="-fPIC"  # -fPIC is needed for python cpp test.
                                 # See python/setup.py for more details
@@ -53,6 +56,8 @@ build_cpp() {
 }
 
 build_cpp_distcheck() {
+  # Initialize any submodules.
+  git submodule update --init --recursive
   ./autogen.sh
   ./configure
   make dist