void sendMessages() throws Exception {
    if (num_threads > 1 && num_msgs % num_threads != 0) {
      System.err.println(
          "num_msgs (" + num_msgs + " ) has to be divisible by num_threads (" + num_threads + ")");
      return;
    }

    System.out.println(
        "sending "
            + num_msgs
            + " messages ("
            + Util.printBytes(msg_size)
            + ") to "
            + remote
            + ": oob="
            + oob
            + ", "
            + num_threads
            + " sender thread(s)");
    ByteBuffer buf =
        ByteBuffer.allocate(Global.BYTE_SIZE + Global.LONG_SIZE).put(START).putLong(num_msgs);
    Message msg = new Message(null, buf.array());
    // msg.writeTo(output);

    ExposedByteArrayOutputStream out_stream = new ExposedByteArrayOutputStream((int) (msg.size()));
    ExposedDataOutputStream dos = new ExposedDataOutputStream(out_stream);
    byte flags = 0;
    dos.writeShort(Version.version); // write the version
    if (msg.getDest() == null) flags += (byte) 2;
    dos.writeByte(flags);
    msg.writeTo(dos);
    Buffer buffer = new Buffer(out_stream.getRawBuffer(), 0, out_stream.size());

    output_lock.lock(); // need to sync if we have more than 1 sender
    try {
      // msg.writeTo(output);
      output.writeInt(buffer.getLength());
      output.write(buffer.getBuf(), buffer.getOffset(), buffer.getLength());
    } finally {
      output_lock.unlock();
    }

    int msgs_per_sender = num_msgs / num_threads;
    Sender[] senders = new Sender[num_threads];
    for (int i = 0; i < senders.length; i++)
      senders[i] = new Sender(msgs_per_sender, msg_size, num_msgs / 10);
    for (Sender sender : senders) sender.start();
    for (Sender sender : senders) sender.join();
    output.flush();
    System.out.println("done sending " + num_msgs + " to " + remote);
  }