PongR part 2: server side


Heads up!

The blog has moved!
If you are interested in reading new posts, the new URL to bookmark is http://blog.valeriogheri.com/


This is part 2 of “PongR – my first experience with HTML5 multiplayer gaming”  (https://thewayofcode.wordpress.com/2012/11/05/pongr-my-first-experience-with-html5-multiplayer-gaming/)

Code and Demo

As always, the code is fully published on github [https://github.com/vgheri/PongR] and a (boring!) demo is available at http://pongr-1.apphb.com . If you don’t have anyone to play with, just open two tabs and enjoy!
Following a screenshot of the game:


Initial countdown before the game starts

Project structure and technologies used

As a quick recap, this project has been realised using Asp.Net MVC 3, HTML5 (Canvas), javascript, C#, SignalR for client/server real-time communication and qUnit for javascript unit testing.

PongR is a multiplayer HTML5 game, and this means that the client itself is doing a good amount of work.
The engine is replicated both on the server and on the client, as we have seen in the first part of this article when I talked about the server authoritative model + client side prediction that force us to add logic onto the client.
In the server this logic is built into a class named Engine and later on we will go into more details.
In the client this logic is built into a javascript module named PongR.

What we’ll see

While writing this second part of the post, I realised that tackling both the server and the client side in one single post will result in a wall of text.
Therefore in part 2 I will start covering the server side code, that is the authoritative source for our game.
Finally in part 3 I will cover the client side code, and this will allow me to dive deeper into HTML5 canvas, qUnit and so on.
So, let’s start!

Connecting to the server

I think that an intuitive way to dive into the code and understand it, is to to navigate the flow of the system: let’s pretend there are no matches currently played on the server, and that we are a client that lands on PongR page.
When the client enters its username and posts the form, the system redirects the client to the page where the actual game will be displayed, PongR.cshtml. This page will load all the necessary javascript libraries and modules, PongR comprised, and will then perform this piece of code on the document onready event:

The Joined() method is inside the PongRHub class, on the server, and will accomplish the following tasks:

  1. Add user to list of connected users
  2. If waiting list is empty add user to waiting list
  3. Else find an opponent (first in the waiting list) and remove him from the waiting list
  4. Create room and assign both users
  5. Create a group for this room
  6. Setup match (playRoom Id, initial ball direction, player on the left and right etc…)
  7. Notify the group the match can start
  8. Add the game to the list of games that the Engine must simulate

Now, the first time I implemented this method I had a problem in between point 5 & 7: sometimes the clients didn’t receive the signal from the server to setup the match.
So I went back to the project’s wiki and I browsed the source code and I understood where the problem lies: when the server invokes Groups.Add(), it sends a signal to the client and notify it about its appartenance to a certain group.
Therefore sometimes I had race conditions where the second signal arrived before the first one, causing the client to drop the message.
I quickly and roughly fixed the issue adding a Thread.Sleep() call after the Groups.Add calls.
Here is the code for Joined():

I could have found a more elegant solution adding one more call after the joined to split the work, but the focus of this project is not SignalR per se, so I guess it’s ok anyway…
Next I signal clients that the match can start along with the game options, like initial direction of the ball, players info etc., and finally enqueue the game on the server so that its state will start to be processed.

// sends a message only to the clients belonging to the group, invoking the function setupMatch() that is part of the PongR module.

// Adds the freshly created game to the list of games to be simulated by the server

This is where our story splits in two parts: we now simultaneously have the clients AND the server playing the game.
How does it work? Again the theory behind what follows is taken from http://buildnewgames.com/real-time-multiplayer/ , which in turn comes from Valve’ Source Engine, I just adjusted it to my needs.
Basically both the server and the clients run two loops, and we call them:

  1. The physics update loop
  2. The update loop

Now these two loops do different things and run at different speeds depending if we are on the server or on the client.
Before moving on talking about the loops, it might be useful to describe what kind of messages the client is sending over to the server.
In PongR a client can only move up or down, and a sequence number is associated to each command to uniquely identify it.
So our payload will contain the following info:

  • the id of the game being played
  • the client id that just sent the command
  • the command
    • a sequence number
    • a string identifying the type of movement: “up” or ”down”

The meaning of the sequence identifier is related to the Client-side prediction mechanism and the input buffer that both the server and the client keep. Here are the steps:

  1. The client moves and stores into its buffer the input along with its id
  2. The client sends the input to the server
  3. The server enqueues the input into a buffer
  4. When the server processes the input, it removes it from the buffer
  5. When the server update loop runs (will see it later in more detail), it notifies the client of the latest input that has been processed
  6. The client can remove from its buffer all the input with ID lower than or equal to the one just ack’ed


Update Packet

Update Packet with sequence identifier acknowledgement

The payload size could be slimmed down by a lot, but that is again another story (for some ideas, take a look here: [www.buildnewgames.com/optimizing-websockets-bandwidth ] )
So let’s start with the server, the authoritative source of our game.

The server loops

To run the loops at fixed intervals, I used the System.Timers.Timer object, and I set them up like this in the Global.asax:

In the second gist we also see the handlers that will run when the Elapsed event is raised.

The server physics update loop

The server physics update loop runs around 60 times per second, that is every 15 msecs.
Its job is to update, for every tick of the simulation, players and ball position based on the inputs received by the clients and on the law of the physics of the game.
Let’s see what a loop round looks like:

MovePlayer() and UpdateBallPosition() are relatively simple so I won’t show them here.
Shortly, the former modifies the Y coordinate of the player and its direction (up/down), based on the inputs that it has received, while acknowledging every processed input sequence number, and the latter simple modifies the (X,Y) coordinates of the ball based only on its angle (that implictitly gives us the direction). For every step of the simulation, both the ball and the players (if any input arrived) will move by a fixed amount of pixels.
Browsing the code you will notice that both the player and the ball movement are based on something called DELTA_TIME. This is related to the concept of framerate independence and I invite you to read more about that here http://www.scirra.com/tutorials/67/delta-time-and-framerate-independence/page-1.
Briefly, quoting the link above:
“Framerate independent games are games that run at the same speed, no matter the framerate.”

Moving on to

var goal = CheckGoalConditionAndUpdateStatus(game);

we see that it’s the server  who is checking if we have a goal situation, and if it’s the case, we update the score of each player (important, and later we’ll see why), add the timestamp of this event to a collection of goal timestamps and then we proceed to restart the game after a goal, that is really resetting positions of players and ball to initial state and choosing randomly the new initial ball direction.
I’m storing the goal timestamp because on the client we provide visual feedback of a goal, running a countdown after each so that the user has the time to understand what is going on and get ready for the next ball to play. During the countdown the ball won’t obviously move, therefore the server must not continue to simulate as well during this time.
We can see this check implemented in the outer physics loop, that is the method actually invoked by the Time elapsed event handler:

That’s it for the server physics update loop.

The server update loop

The update loop runs every 45 msecs, around 22 times per second, so at a much slower rate then the physics loop.
It’s all about finding a good trade-off between accuracy of the simulation (faster update rates) and network usage and performance (slower update rates).
The update loop job is to send the updated state of the game to the right clients.
To do that, I used an external class, called Notifier, which has one public method that will use SignalR to invoke a javascript function on the client even if we are outside of the context of our Hub. This is a very cool SignalR feature, as it allows for a lot of flexibility. Let’s see how it’s done:

The following line will send the message to the two clients that are part of the SignalR group



That’s it for part 2 of this article. We have covered the connection process and the server update loops. We have seen all the workarounds used to cope with SignalR Groups and to skip simulation when one of the player scored in a match, so that the clients can display a countdown before restarting the match.
Part 3 will be the last and we’ll take a look at the client side code, even though you can start by yourself cloning the repository on github.

PongR – my first experience with HTML5 multiplayer gaming


Heads up!

The blog has moved!
If you are interested in reading new posts, the new URL to bookmark is http://blog.valeriogheri.com/


PongR is a Pong like HTML5 multiplayer game.
I started this project to improve my javascript and HTML5 skills, and being a lover of SignalR I thought that it would have been nice to add a touch of multiplayer, even if I had never done such a thing!
Hint: it turns out that adding multiplayer is a HUGE effort, if you want to do it right or even just not completely wrong. 😀
What I still needed to start was a game with an easy model, that required not much graphics skill and that had multiplayer “built-in”… and that’s how I chose Pong!

It is important to note that this is a project I did to learn, and at least for this very first release, the game is really boring(!) and awfully looking 😦
Furthermore this article just describes my experience and my approach: I will point out all the difficulties I came across, all the things I understood, all the points that are still lacking (an awful lot, unfortunately), better technologies that could be used etc… Therefore it’s not a true tutorial about how a HTML5 multiplayer game should be done (I will post some interesting links, though), it’s more about how I put the pieces of the puzzle together and my considerations about HTML5 game development!

This post will be split in two sub-posts, just to avoid having a single, huge, unreadable, wall of text.

  1. The first half is about about the fundamentals: the (simple) math behind the physics model, the architecture of the system, the technologies used and finally the game client/server model, with an insight of how and why I failed hard at my first attempt.
  2. The second half will be about the implementation details: the game simulation engine (both on server side/client side), javascript code structure and unit tests, usage of canvas and more.

The “mathematical” model

First off I started with the mathematical model that drives the game’s physics: player movements, ball movements, collisions with players and the field, goal event.
Players can only move up/down, as per the original Pong game; about the ball, in order to keep things simple, I adopted a basic model: initially it is moving horizontally, right or left .
During the game, the ball can collide either with a player (N.B. if the player is not moving, the ball will only change it’s horizontal direction) either with the field, and the new direction is computed based on the following schema:

For this first version I deliberately chose not to add more complexity, like a variable speed of the ball and players, that in turn can lead to different bouncing angles.

System architecture and technologies used

Because this is a multiplayer game, there will be an important part of server side code and an equally important frontend part (this point is crucial, I will come back to this).
The backend is implemented using Asp.Net MVC and C#, and the realtime communication is managed by SignalR.
The frontend is plain javascript plus some knockout.js to automatically update some parts of the UI.

Two pages: the landing page where the user enters a name and starts the game, and the second is really the game: the pitch and few info on the screen (players’ names and score).
This leads us to talk about the client connection flow:

  1. A client connects to http://pongr-1.appharbor.com, enters a name and submits the form
    1. If there is no match to join, then the client enters into a waiting state
  2. When there are two clients ready to play, the system
    1. Creates a match
    2. Assigns the two clients to the match
    3. Initialise the state of the game
    4. Adds the game it to the list of games that are currently being played
    5. Signals the clients the match can start, along with the initial state of the game
  3. Clients receive the start signal, setup the match and start their partial simulation of the game and draw the scene

Because we can have multiple games going on at the same time, the server needs to know which client belongs to which game (as per point 2b), so that whenever it has to send update about a game, the server will send the message to only the two clients that are playing that match, and not to everyone.
To do this, the server behind the scenes is storing games and players into an in-memory repository, using SignalR Groups feature to be able to send messages only to the right players.
SignalR Groups feature is a powerful, yet strange beast, as we will see in the second part of this post, posing some constraint on the way the system must be designed in order to leverage it.

Game design

Round 1, fight: a completely wrong approach!

As a first approach, because I didn’t know anything about the HTML5 Canvas object, I tried to bypass it entirely using HTML and CSS to draw each frame of the game: the field was a div, the players were divs, even the ball was a div rounded using CSS (border radius…)!! The UI was heavily automatized using Knockout.js.
Each client was simulating its own game, sending updates to the other client (through the SignalR Hub on the server) about its position on a fixed basis.
This means that each client is authoritative regarding its own state: position, score etc… and that the server is only a mere repeater for status updates.
As I quickly understood by playing the game, the first attempt had several fatal issues:

  1. DOM manipulation is really slow and limited when facing typical game requirements (not surprisingly)
  2. After a few seconds of playing, the players were not playing the same game anymore, they were completely out of synch! This is due to the fact that the network is not a perfect channel with no latency, and even with very few delays, the match is rapidly going out of synch.

By googling, I quickly realized that while addressing the first part would turn out to be easy, addressing the second point would be a totally different task.

Round 2 – A new approach: the server authoritative model and its consequences

As a total newbie to game development, I should have probably started following a tutorial from the start, but… it’s good to fail as well, you learn by understanding first hand what works and what does not.
At the end I found a very nice article (which links to other very interesting articles, that I strongly suggest you read, at least for background culture) about multiplayer HTML5 games: http://buildnewgames.com/real-time-multiplayer/
Because the networking architecture of my project is mostly taken from there, and is explained quite well (it’s better to look at the source code though, it will clarify some concepts), I won’t go too much into detail of each step. Once again the goal here is to provide a general idea and to explain my experience.

So what does it mean that the server is authoritative versus the clients in terms of the simulation of the game?
It means that the one and only authority regarding everything that happens in the world is the server: the server is running a simulation of the game, and it updates it based on the inputs that arrive from the clients (key presses, commands etc…).
On a regular basis, the server also notifies the clients of the new game state: all the connected clients  will receive the same update and they will have to update their representation of the game based on what the server computed.
This simple idea will assure that all the connected clients will share almost (we will see later why “almost”) exactly the same representation of the game at any time.
This model is also used to avoid client from cheating: it’s always the server who decides what happens and what doesn’t!

This new approach leads to a new scenario: if before each client was running its own game, now it is the server that is running the game and the clients are left with nothing but to redraw the scene based on current game state.
You can try this approach quite easily but what you get is very poor performance in terms of reactivity: the client presses left, then the client sends a message to the server to notify it about the move, then the server updates the game state and sends the clients a new update, which in turn will redraw the scene, finally moving the player.
This means that after a loooooooooooooooooooooooooooooong wait, the client who moved will see its position change. This is what causes LAG. Not nice.
To avoid these kind of problems, one of the most common techniques is called “Client-side prediction”, and to describe it I can use the words from this very nice article http://www.gabrielgambetta.com/?p=22

“If the game world is deterministic enough (that is, given a game state and a set of inputs, the result is completely predictable),[…], we can assume the inputs we send to the server will be executed successfully. Under this assumption, the client can predict the state of the game world after the inputs are processed, and most of the time this will be correct.
Instead of sending the inputs and waiting for the new game state to start rendering it, we can send the input and start rendering the outcome of that inputs as if they had succeded, while we wait for the server to send the “true” game state – which more often than not, will match the state calculated locally”

This means that now the client is not a dumb spectator anymore, but actually runs some part of the game simulation too!
In the case of PongR, each client is predicting its own position, but because we also have a ball moving and potentially colliding with other objects, then the client needs also to predict every collision with the ball (and this is especially true if you consider sound effects)!
Unfortunately this implies having to duplicate some part of your code, and this can be painful if the languages used in the frontend and in the backend are different, like in this case!
Luckily enough we now have Node.js, which would allow us to have a single code base in a single language, javascript, but this is another story.

There are also other techniques involved, like delta-time to make objects movement independent from the local framerate and client inteporalation to smooth movements of the opponent player. You can find them all detailed in the links I already pointed out.
One final, important, comment: no game state change event is simulated on the client.
Imagine a FPS game, the clients shoots another player and in its own representation of the game, the opponent dies: you see the animation and you get the points… All is good except that as soon as you receive the update from the server, you find out that it’s wrong, you didn’t really kill the opponent as he was 1 meter more on the left that you had known. So… how can you rollback such a situation? Simply you cannot, and that’s why game state change events are only simulated on the server.
In the case of PongR, the only game state change event is when one of the two players score.
We will see how I managed this.

That’s it, we are done for part 1. Next we will see what’s behind the scene.