This is the first part of a six-part series adapted from my master’s thesis, “Creating a First-Person Action Game in Unreal Engine”. The series looks at what it takes to build the next generation of networked multiplayer games, covering everything from networked physics to destructible environments. The original write-up, source project and dev videos live on the NextGenNetGames GitHub wiki.

Introduction

In this topic we will examine networked games as a medium. To specify our needs for the next generation of networked games, we will examine the current generation.

A video is available on this topic.

Networked games as a medium

Chris Crawford, the founder of GDC, often called the Socrates of Games, in his “Dragon’s” speech tells the story of the basic idea behind games [14]. He sees game ideas as dreams, vague, blurry images, consisting only of the outlines, and these ideas are represented in metaphors like dreams of a dragon.

He considers games to be a unique medium because of their interactivity — they can fundamentally change the ways we communicate en masse. Instead of being a static observer who listens, as in classical media, games allow us to interact.

Crawford calls for a set of technologies that can help interactivity, these boil down to the simulation of human behaviour through AI and a visual-audio parity with the real world.

Since the simulation of human behaviour (in a fully dynamic sense, where a simulated human can react to any possible situation) is extremely difficult, if not impossible to execute, we cannot rely on it as a basis for a truly interactive medium. The only obvious solution to this problem is to network people so they can interact together. This is the basic idea behind networking games, so this series is focused on the same.

Crawford invokes that a game is a communication between its creator and the player. This concept in networked games translates to the creator setting up an environment in which players will communicate with each other — that is, the creator sets the rules of the world while the players interact in it. The limit on the number of possible interactions changes from the limitations of a set algorithm, to the practically unlimited number of ideas of a human being. One of the most important factors in any networked game is the number of interactions, as Adam Millard put it in his video “Putting The Players Back In Multiplayer” [15].

We will consider visual and auditory parity with the real world through the principle of realism, as precision, detail or fidelity of a simulation — players are accustomed to the real world, and expect the game to behave like it.

To conclude, interaction and a simulation’s fidelity play an important role in networked games. It logically follows that the environment with more players has more interaction, and it is therefore important that the number of players in a networked game is expanded to the technological bleeding edge.

Technical limitations of the current generation

Let’s now examine the current generation of networked games and their technical limitations.

Fortnite

Fortnite, made in UE4, contains 100 players, with 40-50 thousand replicated characters [18], that is, interactive elements. These interactive elements are destructive objects (trees, walls), pick-up weapons, artificial intelligence and the like.

The world is refreshed very quickly — the server sends updates to clients 30 times a second (30 Hz), while the client sends its data to the server at 60 Hz. Refreshing can be variable, so at the beginning of the round where all 100 players are in one place, the refresh rate is dynamically reduced to 10 Hz [19]. This results in an average data rate of 45.3 Kbit/sec for download, and 42.7 Kbit/sec for upload [20], which is more than acceptable for a wired connection.

The only flaw with this game is the lack of physical objects — these are just vehicles and objects that can be picked up, which gives the impression of a somewhat static world.

Fortnite is the standard of the current generation: 100 players in an interactive world with a high refresh rate.

Battlefield 4

One of the more advanced games in the field of networking technology is Battlefield 4. It consists of 64 players, in a very dynamic world, with destructible objects, deformable terrain, physically based vehicles and projectiles.

It also uses an advanced technique called high frequency networked spheres, an advanced distance-based replication technique [21]. It takes into account:

• the proximity of other characters, so that closer characters refresh faster (if they are within a certain distance)
• and the field of view of the client, so that players inside the FOV get refreshed the fastest

These fields are created on the basis of three classes — pedestrian, ground vehicle and air vehicle — due to their different speeds and fields of view. This method has enabled scaling of the networking system to 120 Hz, which keeps pace with competitive games like Counter-Strike: Global Offensive. [22] Chris “Battle(non)sense”, “BF 120Hz Tickrate: (How) Does it Work?”

The only drawbacks of this game are the number of players, which does not follow the current trend of 100+ players; its destruction is non-dynamic (the environment can be destroyed in only one way); and it has a lack of AI.

This game has one of the best networking configurations, and will serve as a good technological goal for this paper.

DayZ Standalone

DayZ Standalone is a game that, through its development, serves as a perfect argument for server-authoritative logic. It started life as a modification of the game “ArmA 2”, and later switched to a standalone version, in the “Real Virtuality” engine.

DayZ as a game is very punishing — it is played in a persistent world, and if the player ever dies, they start from the beginning, losing all progress. It motivates individuals to look for ways to cheat, and since “Real Virtuality” as an environment was very client-oriented, it was very easy.

When we say client oriented, we mean that the user was trusted when they said something to the server. They could tell the server they were on one end of the world, and the next moment, at a completely different end, or that they never received damage, that they had countless bullets, and so on.

For these, and many other reasons, the server-authoritative model had to be implemented, through the development of the “Enfusion” engine, where, through a long campaign of reengineering the engine itself, they eliminated most methods of cheating [25]. Today (2021) DayZ is a very impressive networked game, with hundreds of thousands of networked objects, vehicles, a lot of AI and a lot of players (60-120 depending on the server configuration).

The conclusion we can draw from DayZ is that we need to create games from the ground up with networking in mind, and with a server-side authoritative model.

Planetside 2

Planetside 2 is a slightly older game, but it has a configuration concentrated on maximising the number of players. It is a set of connected servers into a single experience — one server in this connected set is called a shard. One of these shards can host up to 2000 players, representing the upper limit of the current generation and single-server architectures. It should be mentioned that it sacrifices fidelity to achieve these numbers, such as not replicating projectiles to players, etc.

Specifications of the next generation

Obviously networked games are very technically demanding — they are limited mostly due to processor performance and the current single-server architecture, and are designed around these constraints. They are often just repurposed from local games to networked ones, without too much thought given to the consequences. The result is a very sparse simulation that only gives the appearance of realism, and provides a limited number of interactions.

Which brings us to the idea of the next generation of games. We want to completely separate ourselves from the limitations of the single-server architecture; we want more players in a bigger, more detailed and more persistent world. In short, we want games with less compromise to computing limitations.

As a specification for next-generation games, in a networked sense, we will use the average global internet speed [27], which we estimate at 100 Mbps. From the server we assume a 10 Gbps connection [28]. Furthermore, the game must be server-authoritative, to eliminate the possibility of cheating, and it must exceed the current generation in player numbers and fidelity.

We will achieve this through a change in architecture, and as an extension of that we will work on networked physics, as well as other networked features. Lastly, we will provide the rules of the game for as much interaction as possible through game and match functionality.

A closing thought on local limitations

It should be mentioned that, in a way, these limitations are also present due to the limited budget of local machines, often last-generation consoles — for example, the Battlefield series of games is known for working across many generations, but it is also limited in design and simulation detail. The limitations of local computers, and graphics optimisation, are not the subject of this work and will not be addressed, but it should be mentioned that these problems are being actively worked on, and UE5 with its technologies and methods like Lumen and Nanite will open up even more possibilities [29].

Continue to Part 2 — Networked Physics Overview, covering deterministic lockstep and state sync. The full six-part series and source project are on GitHub.