Inside Rawkets - onGameStart

Hi, I’m Rob Hawkes and I’m here today to give an inside look at the development of Rawkets, my HTML5 and JavaScript multiplayer space shooter.

If you don’t already know, I work at Mozilla.

My official job title is Technical Evangelist, but I prefer Rawket Scientist, which is what it says on my business card.

Part of my job is to engage with developers like you and me about cool new technologies on the Web.

Created by Phil Banks (@emirpprime)

Aside from that I spend most of my time experimenting with HTML5 and other cool technologies.

I basically have an addiction to visual programming and hacking around with code.

It’s fun!

Throughout this talk I plan to take a journey through some of the issues that plagued early development of the game, and cover the subsequent solutions that helped resolve them.

There’ll be some code, but I’m more interested in highlighting the concepts and theories involved.

So far the talks have all been awesome and have often gone into some great detail about particular technologies and techniques. This talk will be quite light-hearted in comparison and more of an overview of technologies and techniques, so hopefully I’ll be able to postpone that feeling of death-by-awesomeness for a little while longer.

However, if you have any questions with what I talk about then feel free to grab me after the talk either in person or online. I’ll be putting my contact details up at the end and the slides will be put online at my blog, which I’ll also mention at the end.

Experimentation

Rawkets is a graduate from my lab

Rawkets is a project that originally came out of this experimentation, from a desire to learn more about WebSockets in regards to multiplayer gaming.

Now, the game is much more mature and I’d consider it as a separate entity aside from the experiments. It’s something that I plan to develop and support way beyond the original scope of learning WebSockets.

What is Rawkets?

Rawkes, WebSockets, and Rockets

Rawkets stands for Rawkes (my blog), WebSockets, and Rockets.

Rawkets is a multiplayer space game that allows you to shoot your friends in the face with HTML5 technologies.

It’s still not really at a beta release level yet, hence the bugs you might notice in this video.

http://rawkets.com

By now you’ve probably realised that the graphics at the beginning of this talk and on the posters aren’t the real game graphics.

They are actually an awesome “artists impression” illustration that I commissioned a guy called Reid Southen to create.

Perhaps when WebGL gets better it will become a reality. Who knows.

It looks pretty awesome as a 6ft banner. So awesome in fact that my girlfriend actually asked me if I was going to put it up in our flat our not. She seemed pretty down about me saying no (it smells pretty horrible).

This is a photo of me in front of the banner at my university end-of-year show. If you think it looks small then let me put it into perspective by telling you that it’s about 8ft away.

Technology

Some of the tech involved in Web gaming

There are a few key technologies that are involved in the development of the game.

All of them bar one are tightly related to HTML or JavaScript.

Canvas

2D graphics platform

Canvas for 2D graphics.

Flash audio

Sound effects and background music

Flash audio for game sound effects and background music.

I’ll explain why I use Flash over HTML5 Audio further on in the talk.

WebSockets

Multiplayer communication

WebSockets is used for the communication between each player and the game server.

For anyone not up to speed with WebSockets, you effectively use HTTP to upgrade a connection and access fully bi-directional streaming communication.

Node.js

Game logic and network communication

Node is used as the game server, controlling the logic and handling the WebSockets connections to the players.

It will eventually be used for player authentication and the storage of data so gameplay can persist over multiple game sessions.

This is all made relatively easy with great third-party modules, like Socket.IO for WebSockets, and others that handle Redis and MongoDB for storage, for example.

http://nodejs.org

Issues

Making games can be a challenge

It’s not all plain sailing when making a gaming using HTML5 and JavaScript.

I’m going to cover a few of the main issues that tripped me up during the development of Rawkets.

Networking

Not as easy as I thought

Issues with networking have plagued development of the game right from the beginning.

This probably stems from my lack of prior experience with socket connection and multiplayer gaming in general.

In the original prototype of the game the network communication was woefully simple and everything was transmitted in a verbose format with no further thought.

In hindsight it’s obvious why I was experiencing massive performance issues with the network communication. I was basically sending way too much data back and forth.

Message protocol

Structured and short communication

One of the ways that I solved these problems was by implementing a structured protocol for the messages that are being sent and received.

This included assigning each message type a number and using enumeration to represent those types in the code.

types = {

PING: 1,

SYNC: 2,

SYNC_COMPLETED: 3,

NEW_PLAYER: 4,

UPDATE_PLAYER: 5,

UPDATE_INPUT: 6,

REMOVE_PLAYER: 7

};

Enumeration

By enumerating the messages types like this I was able to refer to them in a verbose format within the code, but benefit from only sending the one or two digit number when transmitting a message.

This is only possible if both the client and server follow the same protocol in regards to which number refers to which message type.

It’s a simple but effective solution and allowed me to cut a large number of characters from transmitted messages.

{

z: 1,

id: 1234567890,

t: 1316763202872,

s: {

x: 5,

y: 34,

v: 3,

a: 0.46

}

}

Message package

Put together with the message types, a full message package is put together as a JSON representation of a JavaScript object.

All the other pieces of data are attached to the object with a key that is as short as possible.

The z key that you can see above is a reserved key that is used solely for the message type.

The other keys in this example are a timestamp, the player id, and the player state.

Compression

Reducing data as much as possible

Data in WebSockets is transmitted as verbose plain text, so it’s important to cut down and compress it as much as possible.

Some of the ways that I’ve done this include rounding numerical values, reducing the length of words if they’re only used for reference, and generally removing any data that isn’t necessary.

There is also BISON, which is a binary representation of JSON that can cut down the amount of data sent back and forth even further, but I’ve heard conflicting reports about its performance so your mileage may vary.

http://kaijaeger.com/articles/introducing-bison-binary-interchange-standard.html

Rate limitin

g

Cutting down on communication

Aside from the message protocol, one of the biggest issues with networking has been dealing with the sheer number of messages being sent back and forth during the lifetime of a game.

1 1

MESSAGES IN

1MESSAGES OUT

1

Having only one player in the game is easy, you have one message coming in to the server, saying the player has moved, for example, and one message coming back out, updating the player with details from the server.

1 2

2 1

MESSAGES IN

2MESSAGES OUT

4

So say we now have two players, there is still only 1 message in from each player, but now each player receives 2 messages back from the server; one for them, and one for the other player.

This isn’t too bad, but notice how the server is having to send 4 messages – 2 for each player.

1 4

4 1

1

4

4

1

MESSAGES IN

4MESSAGES OUT

16

4 players now, look how the server is having to send 16 messages, yet it only receives 4.

If you haven’t already noticed, the messages out from the server are the square of the number of players.

But 16 messages out is alright, it’s hardly going to tax the server.

MESSAGES IN

30MESSAGES OUT

9001 30

30 1

1

30

30

1

So imagine if we now move into properly multiplayer territory.

30 players in the game would mean 30 messages coming in to the server, and 900 – NINE HUNDRED – messages going out, every update. That’s a silly amount of data for just 30 people.

But let’s go further still…

MESSAGES IN

100MESSAGES OUT

100001 100

100 1

1

100

100

1

Say we go massively multiplayer and have 100 players in the game at one time.

It’s not so bad for each individual player, they send one message in and get 100 back – that’s bearable.

But check out the server, it’s getting 100 messages in and is having to send out 10,000 back, every update. That’s just a mentally stupid number that’s going to cause a lot of grief.

Intelligence

Letting the game prioritis

e messages

Fortunately there is a way around this that cuts down the amount of messages sent; you just need to send data only for players visible to another player, in essence filtering out game data that doesn't affect the current player.

Another trick I used is to only send updates when a player is active and moving. If they haven’t moved since the last frame and nothing else has changed then why bother sending an update and wasting bandwidth?

These are such simple solutions, but ones that I never even considered at first.

Respecting TCP

WebSockets uses TCP. Deal with it

Something else that I discovered is important to be aware of when making a game with WebSockets is that you’re using TCP.

This is a problem as such, but it means that you need to play by a certain set of rules, and to expect a certain set of issues.

By the way, I should point out that that you could argue that the icon that I’ve used could represent WebSockets, but that’s not why I used it. It’s the US plug symbol and I just thought it was funny because it looks like a surprised face. The UK plug symbol is boring in comparison.

Obey the order

You can’t do much about it

One issue with TCP is that packets will come through in order and get queued up if there are any significant connection issues.

This can be a problem with a real-time game as it can cause hang-ups in the transmission and subsequently a hang-up in the graphic display.

In short, the ordering issue can result in jumpy gameplay. Not fun.

With UDP this wouldn’t be so much of a problem, but we don’t have that luxury yet. Although similar protocols are in the pipeline and may make their way into our lives relatively soon, things like Media Streaming APIs and WebRTC.

Use Socket.IO

It intelligently drops messages

I use Socket.IO for all the WebSockets communication for Rawkets simply because it’s easy to use and it has Flash fallback for browsers that don’t support the technology yet.

The cool thing about the latest version of Socket.IO is that is has introduced the concept of volatile messages.

These volatile messages are completely voluntary, and messages sent with them aren't actually sent if the client is having network problems, saving bandwidth for other more important messages.

It’s not an ideal solution, but it’ll certainly help alleviate some of the issues that come with TCP.

Cheaters

A blessing and a curse

There’s no denying it, your code is going to be visible to anyone who wants to look at the source.

I experienced this early on in the development of the game with players adding in their own features, like invincibility, epic speed, rapid-fire, and even creating completely new weapons like cluster bombs!

Now don’t get me wrong, I actually appreciate the cheaters because they highlighted all the errors of my ways, for free. One of the benefits of the open nature of JavaScript is that it can be looked at and poked very easily by others, which means that I can fix bugs quicker than if I was testing on my own.

Globals and closures

Keeping code wide open is bad

There are two reasons why cheating was so prevalent and so easy to do.

The first is that by keeping all the game code in the global namespace and not using any closures, I was practically inviting people to come in and edit the game code. It was too easy to do!

It was so easy in fact that after a few hours of releasing the first prototype, players were already sharing code snippets that others could paste into their browser console to get new features. Annoying, but actually pretty cool.

(function() {

var rawkets = rawkets || {},

r = rawkets;

rawkets.namespace = function(namespace_str) {

var parts = namespace_str.split("."),

parent = rawkets,

i;

if (parts[0] === "rawkets") {

parts = parts.slice(1);

};

for (i = 0; i < parts.length; i++) {

if (typeof parent[parts[i]] === "undefined") {

parent[parts[i]] = {};

};

parent = parent[parts[i]];

};

return parent;

};

window.rawkets = window.r = rawkets;

})(window);

By adding my own “rawkets” namespace I was able to hide code away, and by deliberately utilising closures and private variables I was able to further frustrate efforts by cheaters to overwrite game functionality.

Plus the new namespace makes writing new code and modules so much neater.

Code manipulation isn’t something that I can prevent entirely, but I can at least make it as difficult as possible.

Client authority

Power isn’t always a good thing

I’m not going to lie, the first version of Rawkets was way too trusting.

I used what is referred to as the authoritative client model, which basically means that the client, the player, made all the decisions regarding its position and then sent those positions to the server.

The server than trusted those positions and transmitted them to all the other players, which is fine until the client edits their position and increments it by 100 pixel per frame, rather than 5. Bad times.

This can be referred to as the “Here I am” approach.

Server authority

Relinquish that power

The solution is to make the server authoritative, which means that you prevent manipulation of the client’s code from doing any damage.

All the movement logic is now performed on the server, meaning that when a client moves it simply lets the server know which direction it wants to move. From there the server calculates the new position and sends it back to the client.

This can be referred to as the “Where am I?” approach, and if done right it can completely remove the ability to cheat.

Input Move Update+0 +40 +80

40ms 40ms

80ms total round-trip

Inherent latency

However, the problem with the authoritative server model is that there is some inherent latency within the system.

What I mean by this is that it obviously takes some time for a movement to be sent from the client to the server, then for the server to move the client, and then for the server to send the new position back again.

In the example here imagine that there is a 40ms latency between the client and server, which means that a message sent to the server will take a total of 80ms to make the round-trip.

The problem here is what happens during that 80ms period that you’re waiting for the updated position? If you do nothing then there’s going to be an 80ms delay between you pressing the up arrow and your rawket moving forward. Not good.

Client prediction

Server authority isn’t enough

To solve the latency issues with the authoritative server you need to implement some element of prediction on the client.

What I mean by prediction is an ability for the client to guess, quite accurately, where it should move the player before the message comes back from the server detailing the new position.

Prediction happens here

Input Move Update+0 +40 +80

40ms 40ms

Instant movement

The prediction happens as soon as the client performs some sort of movement (a key-press, etc), before the server has received the input.

All the prediction does is run the same physics as the server, based on the new input.

This is exactly as if were using the authoritative client model, apart from one important difference.

Correction

When prediction goes wrong

Whereas the authoritative client model would be in control, with the authoritative server model and client prediction, the server is in control.

The whole point of using the authoritative server is because the client can’t be trusted. So it makes sense that prediction can’t be trusted either.

To get around this you use periodically check the client position against the server and perform a correction if necessary.

This may sound simple in concept, but it’s one of the hardest aspect of multiplayer gaming to get right. Simply because it’s obvious when you get it wrong.

var correction = function(time, state, input, entity, rk4) {

...

if (Math.abs(state.x - lastMove.state.x) > 2) {

...

var currentTime = time,

currentInput = input;

entity.setState(state); // Rewind entity state

var move, // Current move

frameTime; // Time between correction and stored move

for (m = 0; m < moveCount; m++) {

move = moves[m];

frameTime = (move.time - currentTime) / 1000;

// Update physics based on corrected time, input and state

...

currentTime = move.time;

currentInput = move.input;

move.state = entity.getState();

};

};

};

This is the Gaffer on Games approach by Glen Fiedler.

There are also other solutions, like Valve’s approach which is based on the old QuakeWorld theory.

I’m actually yet to get client-side prediction working faultlessly, and I’d love to see other games developers sharing their JavaScript-based prediction techniques. I know that a lot of developers are working on this.

http://gafferongames.com/game-physics/networked-physics/http://developer.valvesoftware.com/wiki/Latency_Compensating_Methods_in_Client/Server_In-game_Protocol_Design_and_Optimization

Stability

Keeping the game running

Keeping the game running is massively important, especially while it’s in rapid development and is prone to crashing (through errors of my own I must add).

I needed a way to automatically restart the game server if it crashed or something went horribly wrong.

mmonit.com/monit

To do that I originally went with Monit scripts that monitor the WebSockets port for activity.

If the port is inactive for 3 separate checks in a row then the game is automatically restarted in Node.

Ideally the game shouldn't crash, but this has proven to be a great solution that allows me to continue rapidly developing the game without worrying about it going down permanently if I'm not around.

http://mmonit.com/monit/

check host rawkets.com with address 127.0.0.1 start program = "/etc/init.d/rawkets start"

stop program = "/etc/init.d/rawkets stop" if failed port 8000 type tcp for 2 times within 3 cycles then restart

Monit script

Just as a quick example, here is the entire monit script for keeping the game up and running.

I’d say four lines of code was pretty impressive!

This effectively checks port 8000 on rawkets.com for an active tcp connection every cycle (60 seconds), and will only restart the game if there is a problem for at least 2 cycles.

The time between cycles can be changed in the monit configuration files.

#!/bin/sh

case "$1" in

start)

cd /rawkets

/usr/local/bin/node rawkets.js 2>&1 >> /var/log/node.log &

exit 1

;;

stop)

/usr/bin/pkill -f 'node rawkets.js'

exit 1

;;

esac

exit 1

Init.d script

This is the init.d script that I use to start and stop the game server from the command line.

All it does is binds a particular set of shell commands to “start” and “stop” keywords, which I use to start and stop the Node process for the game.

I’m not massively experienced with init.d scripts so I’m sure that it can probably be optimised, but it works perfectly for my needs and is pretty damn cool.

github.com/indexzero/forever

Now I use a little Node module called Forever. It’s amazing!

https://github.com/indexzero/forever

forever start rawkets.js

Forever

All I have to do now is make sure the game process quits on a catastrophic error and Forever will automatically restart it for me.

Using Forever is as simple as installing the module with NPM and then starting your Node script using the Forever demon. The rest is taken care of for you.

github.com/hookio/hook.io

Some of you may also be interested in hook.io, which can help create more stable Node applications.

The concept is to decouple your application logic by breaking it into individual processes so that if one process goes down the rest can continue to run and your entire game doesn’t crash.

You use hook.io through its event system that lets you communicate between these separate processes, regardless of whether they’re on the same server or not. It’s a pretty cool concept.

https://github.com/hookio/hook.io

Why Flash audio?

HTML5 Audio isn’t quite ready yet

So why in a proud HTML5 and JavaScript game have I resorted to Flash?

To put it bluntly, HTML5 Audio support is severely lacking in areas for the needs of game development.

One particular example is support for looping, which is severely lacking and inconsistent across most browsers.

Opera has the best implementation so far, but the rest all have some kind of pause between loops, which makes short and constantly looping audio very obvious and annoying.

There are some solutions, like using JavaScript to try and loop the audio just before it is finished, or using the audio data APIs in Firefox and Chrome, but none are ideal so I went with Flash audio for now until it gets fixed.

Loads more

HTML5 games teach you a lot

There’s loads of stuff that I couldn’t fit in that I’d love to have talked about.

Things like:- Canvas optimisation.- Using internal events to decouple game logic.- Using the same code for client and server.- setTimeout vs. requestAnimationFrame.- How to handle storage.- Control systems for different platforms, like Seb Lee-Delisle’s approach.

The future

What I’d like to see in JavaScript gaming

There are a few things that I’d like to see in the near future to help with game development.

Need a way to benchmark browsers, connections and operating systems.- Like Google Analytics, but for game performance and feature detection.- Measuring FPS.- Measuring network performance.

Better HTML5 audio.

Hardware accelerated canvas on Mac and mobile devices.

Better documentation for game development in JavaScript. This is something I’m working on with Mozilla, but more documentation is needed in general. I’d love to see more game developers sharing their solutions to problems that they encountered during their games.

Something fun

I’ve been working on a little tool

So before I finish I’d like to show you one more thing, something that I’ve been working on recently to help with my game development.

Game profiler

It’s a JavaScript game performance profiler that I’m working on for Rawkets.

The purpose is to visualise performance of key areas of game and network logic in a way that makes it easy to see where things can be improved.

It’s early days, the idea and resulting code are only a week old, but the hope is to eventually release it as a browser addon or module for other games developers and animators to use.

Right now I’m focussing on getting it working well for my own requirements and making sure the resulting visualisation is easy to understand (which it isn’t yet, as you can see).

Webkit timeline-like

The idea is to create something very much like the Webkit timeline panel, just customised for apps and gaming to give a much more fine-grained view of performance.

I’m by no means a competent tools developer, but I’m hoping something like this will at least give me a better idea about exactly where I can increase the performance of Rawkets.

Rob Hawkes

Rawkets.comHTML5 & WebSockets game

Twitter sentiment analysisDelving into your soul

RECENT PROJECTS

Rawkes.comPersonal website and blog

MORE COOL STUFF

ExplicitWeb.co.ukWeb development podcast

Rawket ScientistTechnical Evangelist at Mozilla

@robhawkes

Twitter - @robhawkesRawkes - http://rawkes.com

Become a canvas master

RAWKES.COM/FOUNDATIONCANVAS

Out now

Paperback and digital formats

Learn how to animate

Make two cool space games

Foundation HTML5 Canvas

Foundation HTML5 Canvas is out now on Amazon and other reputable book stores.

November is HTML5 canvas

Every month

Next month is CSS Media Queries

Win prizes (like an Android)

DEVELOPER.MOZILLA.ORG/EN-US/DEMOS/DEVDERBY

Mozilla Dev Derby

Also, you should definitely take part in the Dev Derby, which is a monthly competition run by the Mozilla Developer Network to see what can be done with the latest Web technologies.

This month the focus is on Geolocation, next month is on CSS Media Queries, and November is on HTML5 canvas.

The winners get cool prizes, like an Android phone. It’s a great excuse to play around with these technologies.

https://developer.mozilla.org/en-US/demos/devderby

Hand-picked experts

@ASKMDN & #ASKMDN ON TWITTER

One hour every fortnight

Web development topics

Great discussions

Ask MDN

ASKMDN

onGameStart special (hopefully)

And lastly, I’d like to quickly mention Ask MDN which is a project that I’m working on at Mozilla.

The concept is simple. One hour every fortnight we gather a bunch of experts on Twitter to answer your questions about a particular topic.

We’ve had 4 sessions to date and it’s going down really well. If you follow @AskMDN on Twitter you’ll be sure not to miss the next session.

http://twitter.com/AskMDN

Thank you.

If you have any questions feel free to grab me on Twitter (@robhawkes), or email [email protected]