Add some perf tests to compare Mojo overhead.

ipc/ipc_mojo_perftest.cc

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include <stddef.h>
 #include <memory>
 
 #include "base/memory/ptr_util.h"
 #include "base/process/process_metrics.h"
 #include "base/run_loop.h"
+#include "base/strings/stringprintf.h"
+#include "base/test/perf_time_logger.h"
+#include "base/test/test_io_thread.h"
 #include "base/threading/thread_task_runner_handle.h"
 #include "build/build_config.h"
 #include "ipc/ipc_channel_mojo.h"
-#include "ipc/ipc_perftest_support.h"
+#include "ipc/ipc_test.mojom.h"
+#include "ipc/ipc_test_base.h"
 #include "mojo/edk/embedder/embedder.h"
 #include "mojo/edk/embedder/platform_channel_pair.h"
+#include "mojo/edk/test/mojo_test_base.h"
 #include "mojo/edk/test/multiprocess_test_helper.h"
 #include "mojo/edk/test/scoped_ipc_support.h"
+#include "mojo/public/cpp/bindings/binding.h"
+#include "mojo/public/cpp/system/message_pipe.h"
 
 namespace IPC {
 namespace {
 
-class MojoChannelPerfTest : public test::IPCChannelPerfTestBase {
- public:
-  void TearDown() override {
-    test::IPCChannelPerfTestBase::TearDown();
-  }
-
-  mojo::edk::test::MultiprocessTestHelper helper_;
+// This class simply collects stats about abstract "events" (each of which has a
+// start time and an end time).
+class EventTimeTracker {
+ public:
+  explicit EventTimeTracker(const char* name)
+      : name_(name),
+        count_(0) {
+  }
+
+  void AddEvent(const base::TimeTicks& start, const base::TimeTicks& end) {
+    DCHECK(end >= start);
+    count_++;
+    base::TimeDelta duration = end - start;
+    total_duration_ += duration;
+    max_duration_ = std::max(max_duration_, duration);
+  }
+
+  void ShowResults() const {
+    VLOG(1) << name_ << " count: " << count_;
+    VLOG(1) << name_ << " total duration: "
+            << total_duration_.InMillisecondsF() << " ms";
+    VLOG(1) << name_ << " average duration: "
+            << (total_duration_.InMillisecondsF() / static_cast<double>(count_))
+            << " ms";
+    VLOG(1) << name_ << " maximum duration: "
+            << max_duration_.InMillisecondsF() << " ms";
+  }
+
+  void Reset() {
+    count_ = 0;
+    total_duration_ = base::TimeDelta();
+    max_duration_ = base::TimeDelta();
+  }
+
+ private:
+  const std::string name_;
+
+  uint64_t count_;
+  base::TimeDelta total_duration_;
+  base::TimeDelta max_duration_;
+
+  DISALLOW_COPY_AND_ASSIGN(EventTimeTracker);
+};
+
+class PerformanceChannelListener : public Listener {
+ public:
+  explicit PerformanceChannelListener(const std::string& label)
+      : label_(label),
+        sender_(NULL),
+        msg_count_(0),
+        msg_size_(0),
+        count_down_(0),
+        latency_tracker_("Server messages") {
+    VLOG(1) << "Server listener up";
+  }
+
+  ~PerformanceChannelListener() override {
+    VLOG(1) << "Server listener down";
+  }
+
+  void Init(Sender* sender) {
+    DCHECK(!sender_);
+    sender_ = sender;
+  }
+
+  // Call this before running the message loop.
+  void SetTestParams(int msg_count, size_t msg_size) {
+    DCHECK_EQ(0, count_down_);
+    msg_count_ = msg_count;
+    msg_size_ = msg_size;
+    count_down_ = msg_count_;
+    payload_ = std::string(msg_size_, 'a');
+  }
+
+  bool OnMessageReceived(const Message& message) override {
+    CHECK(sender_);
+
+    base::PickleIterator iter(message);
+    int64_t time_internal;
+    EXPECT_TRUE(iter.ReadInt64(&time_internal));
+    int msgid;
+    EXPECT_TRUE(iter.ReadInt(&msgid));
+    std::string reflected_payload;
+    EXPECT_TRUE(iter.ReadString(&reflected_payload));
+
+    // Include message deserialization in latency.
+    base::TimeTicks now = base::TimeTicks::Now();
+
+    if (reflected_payload == "hello") {
+      // Start timing on hello.
+      latency_tracker_.Reset();
+      DCHECK(!perf_logger_.get());
+      std::string test_name =
+          base::StringPrintf("IPC_%s_Perf_%dx_%u",
+                             label_.c_str(),
+                             msg_count_,
+                             static_cast<unsigned>(msg_size_));
+      perf_logger_.reset(new base::PerfTimeLogger(test_name.c_str()));
+    } else {
+      DCHECK_EQ(payload_.size(), reflected_payload.size());
+
+      latency_tracker_.AddEvent(
+          base::TimeTicks::FromInternalValue(time_internal), now);
+
+      CHECK(count_down_ > 0);
+      count_down_--;
+      if (count_down_ == 0) {
+        perf_logger_.reset();  // Stop the perf timer now.
+        latency_tracker_.ShowResults();
+        base::MessageLoop::current()->QuitWhenIdle();
+        return true;
+      }
+    }
+
+    Message* msg = new Message(0, 2, Message::PRIORITY_NORMAL);
+    msg->WriteInt64(base::TimeTicks::Now().ToInternalValue());
+    msg->WriteInt(count_down_);
+    msg->WriteString(payload_);
+    sender_->Send(msg);
+    return true;
+  }
+
+ private:
+  std::string label_;
+  Sender* sender_;
+  int msg_count_;
+  size_t msg_size_;
+
+  int count_down_;
+  std::string payload_;
+  EventTimeTracker latency_tracker_;
+  std::unique_ptr<base::PerfTimeLogger> perf_logger_;
+};
+
+// This channel listener just replies to all messages with the exact same
+// message. It assumes each message has one string parameter. When the string
+// "quit" is sent, it will exit.
+class ChannelReflectorListener : public Listener {
+ public:
+  ChannelReflectorListener()
+      : channel_(NULL),
+        latency_tracker_("Client messages") {
+    VLOG(1) << "Client listener up";
+  }
+
+  ~ChannelReflectorListener() override {
+    VLOG(1) << "Client listener down";
+    latency_tracker_.ShowResults();
+  }
+
+  void Init(Channel* channel) {
+    DCHECK(!channel_);
+    channel_ = channel;
+  }
+
+  bool OnMessageReceived(const Message& message) override {
+    CHECK(channel_);
+
+    base::PickleIterator iter(message);
+    int64_t time_internal;
+    EXPECT_TRUE(iter.ReadInt64(&time_internal));
+    int msgid;
+    EXPECT_TRUE(iter.ReadInt(&msgid));
+    base::StringPiece payload;
+    EXPECT_TRUE(iter.ReadStringPiece(&payload));
+
+    // Include message deserialization in latency.
+    base::TimeTicks now = base::TimeTicks::Now();
+
+    if (payload == "hello") {
+      latency_tracker_.Reset();
+    } else if (payload == "quit") {
+      latency_tracker_.ShowResults();
+      base::MessageLoop::current()->QuitWhenIdle();
+      return true;
+    } else {
+      // Don't track hello and quit messages.
+      latency_tracker_.AddEvent(
+          base::TimeTicks::FromInternalValue(time_internal), now);
+    }
+
+    Message* msg = new Message(0, 2, Message::PRIORITY_NORMAL);
+    msg->WriteInt64(base::TimeTicks::Now().ToInternalValue());
+    msg->WriteInt(msgid);
+    msg->WriteString(payload);
+    channel_->Send(msg);
+    return true;
+  }
+
+ private:
+  Channel* channel_;
+  EventTimeTracker latency_tracker_;
+};
+
+// This class locks the current thread to a particular CPU core. This is
+// important because otherwise the different threads and processes of these
+// tests end up on different CPU cores which means that all of the cores are
+// lightly loaded so the OS (Windows and Linux) fails to ramp up the CPU
+// frequency, leading to unpredictable and often poor performance.
+class LockThreadAffinity {
+ public:
+  explicit LockThreadAffinity(int cpu_number) : affinity_set_ok_(false) {
+#if defined(OS_WIN)
+    const DWORD_PTR thread_mask = static_cast<DWORD_PTR>(1) << cpu_number;
+    old_affinity_ = SetThreadAffinityMask(GetCurrentThread(), thread_mask);
+    affinity_set_ok_ = old_affinity_ != 0;
+#elif defined(OS_LINUX)
+    cpu_set_t cpuset;
+    CPU_ZERO(&cpuset);
+    CPU_SET(cpu_number, &cpuset);
+    auto get_result = sched_getaffinity(0, sizeof(old_cpuset_), &old_cpuset_);
+    DCHECK_EQ(0, get_result);
+    auto set_result = sched_setaffinity(0, sizeof(cpuset), &cpuset);
+    // Check for get_result failure, even though it should always succeed.
+    affinity_set_ok_ = (set_result == 0) && (get_result == 0);
+#endif
+    if (!affinity_set_ok_)
+      LOG(WARNING) << "Failed to set thread affinity to CPU " << cpu_number;
+  }
+
+  ~LockThreadAffinity() {
+    if (!affinity_set_ok_)
+      return;
+#if defined(OS_WIN)
+    auto set_result = SetThreadAffinityMask(GetCurrentThread(), old_affinity_);
+    DCHECK_NE(0u, set_result);
+#elif defined(OS_LINUX)
+    auto set_result = sched_setaffinity(0, sizeof(old_cpuset_), &old_cpuset_);
+    DCHECK_EQ(0, set_result);
+#endif
+  }
+
+ private:
+  bool affinity_set_ok_;
+#if defined(OS_WIN)
+  DWORD_PTR old_affinity_;
+#elif defined(OS_LINUX)
+  cpu_set_t old_cpuset_;
+#endif
+
+  DISALLOW_COPY_AND_ASSIGN(LockThreadAffinity);
+};
+
+class PingPongTestParams {
+ public:
+  PingPongTestParams(size_t size, int count)
+      : message_size_(size), message_count_(count) {
+  }
+
+  size_t message_size() const { return message_size_; }
+  int message_count() const { return message_count_; }
+
+ private:
+  size_t message_size_;
+  int message_count_;
+};
+
+std::vector<PingPongTestParams> GetDefaultTestParams() {
+  // Test several sizes. We use 12^N for message size, and limit the message
+  // count to keep the test duration reasonable.
+#ifdef NDEBUG
+  const int kMultiplier = 100;
+#else
+  // Debug builds on Windows run these tests orders of magnitude more slowly.
+  const int kMultiplier = 1;
+#endif
+  std::vector<PingPongTestParams> list;
+  list.push_back(PingPongTestParams(12, 500 * kMultiplier));
+  list.push_back(PingPongTestParams(144, 500 * kMultiplier));
+  list.push_back(PingPongTestParams(1728, 500 * kMultiplier));
+  list.push_back(PingPongTestParams(20736, 120 * kMultiplier));
+  list.push_back(PingPongTestParams(248832, 10 * kMultiplier));
+  return list;
+}
+
+// Avoid core 0 due to conflicts with Intel's Power Gadget.
+// Setting thread affinity will fail harmlessly on single/dual core machines.
+const int kSharedCore = 2;
+
+class MojoChannelPerfTest : public IPCChannelMojoTestBase {
+ public:
+  MojoChannelPerfTest() = default;
+  ~MojoChannelPerfTest() override = default;
+
+  void RunTestChannelPingPong() {
+    Init("MojoPerfTestClient");
+
+    // Set up IPC channel and start client.
+    PerformanceChannelListener listener("Channel");
+    CreateChannel(&listener);
+    listener.Init(channel());
+    ASSERT_TRUE(ConnectChannel());
+
+    LockThreadAffinity thread_locker(kSharedCore);
+    std::vector<PingPongTestParams> params = GetDefaultTestParams();
+    for (size_t i = 0; i < params.size(); i++) {
+      listener.SetTestParams(params[i].message_count(),
+                             params[i].message_size());
+
+      // This initial message will kick-start the ping-pong of messages.
+      Message* message =
+          new Message(0, 2, Message::PRIORITY_NORMAL);
+      message->WriteInt64(base::TimeTicks::Now().ToInternalValue());
+      message->WriteInt(-1);
+      message->WriteString("hello");
+      sender()->Send(message);
+
+      // Run message loop.
+      base::RunLoop().Run();
+    }
+
+    // Send quit message.
+    Message* message = new Message(0, 2, Message::PRIORITY_NORMAL);
+    message->WriteInt64(base::TimeTicks::Now().ToInternalValue());
+    message->WriteInt(-1);
+    message->WriteString("quit");
+    sender()->Send(message);
+
+    EXPECT_TRUE(WaitForClientShutdown());
+    DestroyChannel();
+}
+
+  void RunTestChannelProxyPingPong() {
+    io_thread_.reset(new base::TestIOThread(base::TestIOThread::kAutoStart));
+
+    Init("MojoPerfTestClient");
+
+    // Set up IPC channel and start client.
+    PerformanceChannelListener listener("ChannelProxy");
+    auto channel_proxy = IPC::ChannelProxy::Create(
+        TakeHandle().release(), IPC::Channel::MODE_SERVER, &listener,
+        io_thread_->task_runner());
+    listener.Init(channel_proxy.get());
+
+    LockThreadAffinity thread_locker(kSharedCore);
+    std::vector<PingPongTestParams> params = GetDefaultTestParams();
+    for (size_t i = 0; i < params.size(); i++) {
+      listener.SetTestParams(params[i].message_count(),
+                              params[i].message_size());
+
+      // This initial message will kick-start the ping-pong of messages.
+      Message* message = new Message(0, 2, Message::PRIORITY_NORMAL);
+      message->WriteInt64(base::TimeTicks::Now().ToInternalValue());
+      message->WriteInt(-1);
+      message->WriteString("hello");
+      channel_proxy->Send(message);
+
+      // Run message loop.
+      base::RunLoop().Run();
+    }
+
+    // Send quit message.
+    Message* message = new Message(0, 2, Message::PRIORITY_NORMAL);
+    message->WriteInt64(base::TimeTicks::Now().ToInternalValue());
+    message->WriteInt(-1);
+    message->WriteString("quit");
+    channel_proxy->Send(message);
+
+    EXPECT_TRUE(WaitForClientShutdown());
+    channel_proxy.reset();
+
+    io_thread_.reset();
+  }
+
+  scoped_refptr<base::TaskRunner> io_task_runner() {
+    if (io_thread_)
+      return io_thread_->task_runner();
+    return base::ThreadTaskRunnerHandle::Get();
+  }
+
+ private:
+  std::unique_ptr<base::TestIOThread> io_thread_;
 };
 
 TEST_F(MojoChannelPerfTest, ChannelPingPong) {
-  RunTestChannelPingPong(GetDefaultTestParams());
+  RunTestChannelPingPong();
 
   base::RunLoop run_loop;
   run_loop.RunUntilIdle();
 }
 
 TEST_F(MojoChannelPerfTest, ChannelProxyPingPong) {
-  RunTestChannelProxyPingPong(GetDefaultTestParams());
+  RunTestChannelProxyPingPong();
 
   base::RunLoop run_loop;
   run_loop.RunUntilIdle();
 }
 
 // Test to see how many channels we can create.
 TEST_F(MojoChannelPerfTest, DISABLED_MaxChannelCount) {
 #if defined(OS_POSIX)
   LOG(INFO) << "base::GetMaxFds " << base::GetMaxFds();
   base::SetFdLimit(20000);
 #endif
 
   std::vector<mojo::edk::PlatformChannelPair*> channels;
   for (size_t i = 0; i < 10000; ++i) {
     LOG(INFO) << "channels size: " << channels.size();
     channels.push_back(new mojo::edk::PlatformChannelPair());
   }
 }
 
-class MojoPerfTestClient : public test::PingPongTestClient {
+class MojoPerfTestClient {
  public:
-  typedef test::PingPongTestClient SuperType;
+  MojoPerfTestClient()
+      : listener_(new ChannelReflectorListener()) {
+    mojo::edk::test::MultiprocessTestHelper::ChildSetup();
+  }
 
-  MojoPerfTestClient();
+  ~MojoPerfTestClient() = default;
 
-  std::unique_ptr<Channel> CreateChannel(Listener* listener) override;
+  int Run(MojoHandle handle) {
+    handle_ = mojo::MakeScopedHandle(mojo::MessagePipeHandle(handle));
+    LockThreadAffinity thread_locker(kSharedCore);
+    std::unique_ptr<Channel> channel = ChannelMojo::Create(
+        std::move(handle_), Channel::MODE_CLIENT, listener_.get());
+    listener_->Init(channel.get());
+    CHECK(channel->Connect());
 
-  int Run(MojoHandle handle);
+    base::RunLoop().Run();
+    return 0;
+  }
 
  private:
-  mojo::edk::test::ScopedIPCSupport ipc_support_;
+  base::MessageLoopForIO main_message_loop_;
+  std::unique_ptr<ChannelReflectorListener> listener_;
+  std::unique_ptr<Channel> channel_;
   mojo::ScopedMessagePipeHandle handle_;
 };
 
-MojoPerfTestClient::MojoPerfTestClient()
-    : ipc_support_(base::ThreadTaskRunnerHandle::Get()) {
-  mojo::edk::test::MultiprocessTestHelper::ChildSetup();
-}
-
-std::unique_ptr<Channel> MojoPerfTestClient::CreateChannel(Listener* listener) {
-  return ChannelMojo::Create(std::move(handle_), Channel::MODE_CLIENT,
-                             listener);
-}
-
-int MojoPerfTestClient::Run(MojoHandle handle) {
-  handle_ = mojo::MakeScopedHandle(mojo::MessagePipeHandle(handle));
-  return RunMain();
-}
-
 MULTIPROCESS_TEST_MAIN(MojoPerfTestClientTestChildMain) {
   MojoPerfTestClient client;
   int rv = mojo::edk::test::MultiprocessTestHelper::RunClientMain(
       base::Bind(&MojoPerfTestClient::Run, base::Unretained(&client)));
 
   base::RunLoop run_loop;
   run_loop.RunUntilIdle();
 
   return rv;
 }
 
+class ReflectorImpl : public IPC::mojom::Reflector {
+ public:
+  explicit ReflectorImpl(mojo::ScopedMessagePipeHandle handle)
+      : binding_(this, std::move(handle)) {}
+  ~ReflectorImpl() override {
+    ignore_result(binding_.Unbind().PassMessagePipe().release());
+  }
+
+ private:
+  // IPC::mojom::Reflector:
+  void Ping(const std::string& value, const PingCallback& callback) override {
+    callback.Run(value);
+  }
+
+  void Quit() override {
+    base::MessageLoop::current()->QuitWhenIdle();
+  }
+
+  mojo::Binding<IPC::mojom::Reflector> binding_;
+};
+
+class MojoInterfacePerfTest : public mojo::edk::test::MojoTestBase {
+ public:
+  MojoInterfacePerfTest() : message_count_(0), count_down_(0) {}
+
+ protected:
+  void RunPingPongServer(MojoHandle mp, const std::string& label) {
+    base::MessageLoop main_message_loop;
+    label_ = label;
+
+    mojo::MessagePipeHandle mp_handle(mp);
+    mojo::ScopedMessagePipeHandle scoped_mp(mp_handle);
+    ping_receiver_.Bind(IPC::mojom::ReflectorPtrInfo(
+        std::move(scoped_mp), 0u));
+
+    LockThreadAffinity thread_locker(kSharedCore);
+    std::vector<PingPongTestParams> params = GetDefaultTestParams();
+    for (size_t i = 0; i < params.size(); i++) {
+      ping_receiver_->Ping(
+          "hello",
+          base::Bind(&MojoInterfacePerfTest::OnPong, base::Unretained(this)));
+      message_count_ = count_down_ = params[i].message_count();
+      payload_ = std::string(params[i].message_size(), 'a');
+
+      base::RunLoop().Run();
+    }
+
+    ping_receiver_->Quit();
+
+    ignore_result(ping_receiver_.PassInterface().PassHandle().release());
+  }
+
+  void OnPong(const std::string& value) {
+    if (value == "hello") {
+      DCHECK(!perf_logger_.get());
+      std::string test_name =
+          base::StringPrintf("IPC_%s_Perf_%dx_%zu",
+                             label_.c_str(),
+                             message_count_,
+                             payload_.size());
+      perf_logger_.reset(new base::PerfTimeLogger(test_name.c_str()));
+    } else {
+      DCHECK_EQ(payload_.size(), value.size());
+
+      CHECK(count_down_ > 0);
+      count_down_--;
+      if (count_down_ == 0) {
+        perf_logger_.reset();
+        base::MessageLoop::current()->QuitWhenIdle();
+        return;
+      }
+    }
+
+    ping_receiver_->Ping(
+        payload_,
+        base::Bind(&MojoInterfacePerfTest::OnPong, base::Unretained(this)));
+  }
+
+  static int RunPingPongClient(MojoHandle mp) {
+    mojo::MessagePipeHandle mp_handle(mp);
+    mojo::ScopedMessagePipeHandle scoped_mp(mp_handle);
+
+    // In single process mode, this is running in a task and by default other
+    // tasks (in particular, the binding) won't run. To keep the single process
+    // and multi-process code paths the same, enable nestable tasks.
+    base::MessageLoop::ScopedNestableTaskAllower nest_loop(
+        base::MessageLoop::current());
+
+    LockThreadAffinity thread_locker(kSharedCore);
+    ReflectorImpl impl(std::move(scoped_mp));
+    base::RunLoop().Run();
+    return 0;
+  }
+
+ private:
+  int message_count_;
+  int count_down_;
+  std::string label_;
+  std::string payload_;
+  IPC::mojom::ReflectorPtr ping_receiver_;
+  std::unique_ptr<base::PerfTimeLogger> perf_logger_;
+
+  DISALLOW_COPY_AND_ASSIGN(MojoInterfacePerfTest);
+};
+
+DEFINE_TEST_CLIENT_WITH_PIPE(PingPongClient, MojoInterfacePerfTest, h) {
+  base::MessageLoop main_message_loop;
+  return RunPingPongClient(h);
+}
+
+// Similar to MojoChannelPerfTest above, but uses a Mojo interface instead of
+// raw IPC::Messages.
+TEST_F(MojoInterfacePerfTest, MultiprocessPingPong) {
+  RUN_CHILD_ON_PIPE(PingPongClient, h)
+    RunPingPongServer(h, "MultiProcess");
+  END_CHILD()
+}
+
+// A single process version of the above test.
+TEST_F(MojoInterfacePerfTest, SingleProcessPingPong) {
+  MojoHandle server_handle, client_handle;
+  CreateMessagePipe(&server_handle, &client_handle);
+
+  base::Thread client_thread("PingPongClient");
+  client_thread.Start();
+  client_thread.task_runner()->PostTask(
+      FROM_HERE,
+      base::Bind(base::IgnoreResult(&RunPingPongClient), client_handle));
+
+  RunPingPongServer(server_handle, "SingleProcess");
+}
+
+class CallbackPerfTest : public testing::Test {
+ public:
+  CallbackPerfTest()
+      : client_thread_("PingPongClient"), message_count_(0), count_down_(0) {}
+
+ protected:
+  void RunPingPongServer() {
+    client_thread_.Start();
+
+    LockThreadAffinity thread_locker(kSharedCore);
+    std::vector<PingPongTestParams> params = GetDefaultTestParams();
+    for (size_t i = 0; i < params.size(); i++) {
+      std::string hello("hello");
+      client_thread_.task_runner()->PostTask(
+          FROM_HERE,
+          base::Bind(&CallbackPerfTest::Ping, base::Unretained(this), hello));
+      message_count_ = count_down_ = params[i].message_count();
+      payload_ = std::string(params[i].message_size(), 'a');
+
+      base::RunLoop().Run();
+    }
+  }
+
+  void Ping(const std::string& value) {
+    main_message_loop.task_runner()->PostTask(
+        FROM_HERE,
+        base::Bind(&CallbackPerfTest::OnPong, base::Unretained(this),
+        value));
+  }
+
+  void OnPong(const std::string& value) {
+    if (value == "hello") {
+      DCHECK(!perf_logger_.get());
+      std::string test_name =
+          base::StringPrintf("Callback_Perf_%dx_%zu",
+                             message_count_,
+                             payload_.size());
+      perf_logger_.reset(new base::PerfTimeLogger(test_name.c_str()));
+    } else {
+      DCHECK_EQ(payload_.size(), value.size());
+
+      CHECK(count_down_ > 0);
+      count_down_--;
+      if (count_down_ == 0) {
+        perf_logger_.reset();
+        base::MessageLoop::current()->QuitWhenIdle();
+        return;
+      }
+    }
+
+    client_thread_.task_runner()->PostTask(
+        FROM_HERE,
+        base::Bind(&CallbackPerfTest::Ping, base::Unretained(this), payload_));
+  }
+
+ private:
+  base::Thread client_thread_;
+  base::MessageLoop main_message_loop;
+  int message_count_;
+  int count_down_;
+  std::string payload_;
+  std::unique_ptr<base::PerfTimeLogger> perf_logger_;
+
+  DISALLOW_COPY_AND_ASSIGN(CallbackPerfTest);
+};
+
+// Sends the same data as above using PostTask instead of IPCs for comparison.
+TEST_F(CallbackPerfTest, PingPong) {
+  RunPingPongServer();
+}
+
 }  // namespace
 }  // namespace IPC