114 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games Figure 7.7 Scene from Wolfenstein Enemy Territory regular (top) and with wallhack, glow and names (bottom). Reproduced by permission of Id Software, Inc. request to the graphics driver to draw an opaque object . The 3D graphics drivers render each scene by drawing more distant objects first and then overlaying these with closer objects. When the closer objects are opaque the more distant objects (e.g. other players) are effectively hidden. If the cheater can modify the graphics driver to interpret opaque as mostly transparent (or draw as an open wire frame ), then a wallhack is created without touching the game client software at all. Client software can sometimes also be tricked into revealing information without modifying or patching the client software itself. Many clients have configurable video display options that players can tweak to optimise their client s relationship with the player s choice of video/graphics card. Sometimes these settings can, themselves, cause discontinuities in how map elements are rendered creating holes and gaps between walls and doors where they should not be. Client-side cheats are problematic. Players must trust that every other player s game client is obeying the game s rules and laws of physics, yet the potentially malicious player controls the entire hardware and software environment within which their game
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services
112 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games who play on your behalf to execute basic strategies 24 hours a day while you are off elsewhere (or sleeping). In all interactive game types, there are advantages to being able to execute certain sequences of moves (e.g. jump, spin and shoot). Thus people develop software tools to automate (and thus augment) the player s execution of such moves. One of the most common augmentations in FPS games is the Aimbot automated code to assist in aiming and shooting at other players. Once patched into the actual game client software, an aimbot is fully aware of the precise locations of all other players within the client s view. When the player issues shoot commands, the aimbot rewrites the aiming information with precise coordinates of the target, guaranteeing a hit every time. More sophisticated aimbots may track and automatically shoot targets you have pointed at. Naturally, this is considered an extremely annoying cheat by the victims. Another form of augmentation involves violating the game s laws of physics that would normally impede a player s perception of the world around them. There are a number of examples. Over-riding other players choices of avatar skin (colour schemes), making them clearly stand out visually on screen regardless of ambient lighting Revealing tactically useful information on screen about other players (e.g. their remaining strength, what weapons they are carrying), regardless of whether or not the players are actually visible Eliminating game elements that obscure a player s vision or knowledge of the surrounding map regions. (E.g. fog of war obscuring maps in RTS games, or snow/rain in most games, and opaque walls and boxes hiding enemies in FPS games.) These cheats violate the trust model of the original game server and client. The server provides information about the game (e.g. other players or map layout) that would not normally be revealed to the actual player until such time as the player reaches a point in the map where the information would naturally become apparent. Augmentation cheats remove such restrictions. One source of tactical advantage is to simply augment the skins of other players with a glow or colours that otherwise dramatically contrast with the textures, colours and lighting around the map. This defeats the ability of other players to blend into the shadows and camouflage colouring around a map. To a certain degree you can see into shadows by adjusting the brightness and gamma on your video card and monitor. However, this can create eye strain as the rest of the map becomes too bright and washed out. A client-side software modification targeting only the other players is much more effective. Figure 7.5 shows two versions of the same scene in Quake III Arena with normal and glowing skins. The glow cheat makes the character on the left easy to see even in a fast-paced game. Another form of cheat involves revealing information about the character in view. Figure 7.6 shows a close-up of a Quake III Arena character with its name and current weapon overlaid (this example also shows the glow hack). The most useful cheat is the wallhack . This refers to any cheat that allows you to see through opaque objects (such as walls). There is incredible tactical benefit to being aware of an oncoming opponent before they are aware of you. Figure 7.7 shows an example in the context of Wolfenstein Enemy Territory. With the cheat activated, you become aware of the medic and soldier hidden from view at the end of the corridor well before they become aware of you.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services
Playability versus Network Conditions and Cheats A broad range of cheating becomes possible when the server administrator(s) cannot be trusted. Administrators can meddle in a God mode within the game, completely subvert the game-state by modifying the game s database from outside the game itself, or modify the game server s internal game-play rules in subtle ways. One of the authors tried exactly this kind of server-side cheat with a simple Quake III Arena mod. In this FPS, the server tells clients the rules of game physics, and clients fall in line. The author s server-side modification gave extra flight time to rockets fired by the author. Conversely, the rockets of another specific player (the victim) were given truncated flight times (so they often self-destruct after flying a shorter-than-normal distance) proportional to how many rockets the victim had left. (The victim a relatively good Quake III Arena player at the time did not notice the imbalance for a number of days.) The more usual annoyances imposed by server administrators include changing configurable game physics parameters (such as gravity), capriciously banning or kicking off players and generally being a disruptive god character. Such abuses tend to be self-limiting though, as the administrator soon ends up with an unpopular and empty server. Finally, it is worth noting a particular style of game play, known as camping,thatis often described as cheating. Particularly in FPS games, there may be places in a map where a player has a good shooting position while at the same time being hard to kill. When so positioned, a player can rack up a large number of kills without moving around (i.e. they camp ). On the one hand it may be cheating to use a camping position that exists only because the map was poorly designed. On the other hand, in real life you cannot complain that the enemy is being unfair if they find a location with admirable camping potential and to the extent a game tries to emulate real-life scenarios, camping is not cheating. The decision ultimately depends on the game players themselves on a case-by-case basis. 7.2.3 Client-side Cheats A huge degree of trust must be placed in the client side of an online game. The game-play experience is entirely mediated by the combination of game client software and the operating system, and hardware environment on which the game client runs. Unfortunately, this is almost precisely the wrong place to put much trust because the client software runs on physical hardware entirely under control of the player. (Whilst this is trivially true for PC games, it is often also true of console games where the console s internal security mechanisms are reverse-engineered by third parties.) Cheats in games using peer-to-peer communication models are essentially variations of client-side cheats, made possible because the local rendering of game-state occurs on the player s own equipment. Most client-side cheats involve manipulating the software context within which the game client operates to augment a player s apparent reflexes and presence, or augment a player s situational awareness by revealing supposedly hidden information. In FPS games, the most important augmentation is aiming of one s gun. In real-time strategy and role-playing games, augmentation may be in the form of automated bots
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services
110 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games actions. Second, the human administrator(s) of the game server may not be trustworthy (or be subject to external influence sufficiently to elicit corrupting behaviour). One example of game-play corruption is the escape cheat. Upon realising that they are losing the game, a player withdraws at the last minute rather than risk having their final score logged for public record and display. Their withdrawal (escape) prevents their own reputation from being damaged and, in some games, denies their opponent a legitimate win. Escape cheating can occur in any game where the server forgets about a player as soon as they leave. The partial solution involves tracking escapees as well as winners and losers, and publicising a player s ranking in all three categories. Where the game server keeps comprehensive activity logs it may also be possible to reconstruct, and publish, the cheater s lack of success up to the point where they left. This solution works because it modifies the metrics by which success is measured socially, putting pressure on players to be seen to play games through to the end. For example, at the end of a map, Quake III Arena only prints on screen the scores of players who remained connected to the server. The server logs do contain all kill/death information, so a cheater s actual success can be determined by an external program. Similar problems have been reported with online strategy games such as Go and Bridge [YAN2003, YAN2002]. Alternative methods of pressuring players to stay have their own weaknesses. For example, a person escaping a game may simply be declared the loser if they do not rejoin and continue in a short period. Unfortunately, such a server-side rule invites malicious players to frame another player use a network DoS attack (discussed later) to force someone else s disconnection in such a way that the victim cannot reconnect in time and thus is declared the loser. (This works better with games such as Go and Bridge, rather than fast-paced FPS games.) Another example of incompletely specified in-game behaviour was the skin cheat of early Counter-Strike servers. Counter-Strike is a team-play mod of Half-Life, where two competing teams ( terrorists and counter-terrorists ) do battle. The original servers did not expect players to request a change in their skin (the image they project to other players) after the beginning of each game. Because it wasn t expected, the server did not block such requests either. So unscrupulous players would switch skins in-game to infiltrate opposition positions, achieving a substantial tactical advantage in the process [CHO2001]. For games that rely on players logging in with a user name and password, there can be issues associated with protection of the password system from attack. The simplest cheat would be to launch a dictionary attack on the online login process of a game server cycling through thousands of possible words as passwords for your intended victim. On the face of it, the game server can compensate by enforcing a minimum delay to every password attempt and then freeze the account after a certain number of failed login attempts. Unfortunately, this leads to another type of cheat freeze another player out of the game by deliberately logging into the game server as the victim, using passwords you know are not right. Once you hit the server s limit for bad passwords your victim can also not login. This would be a major problem for the victim if the freeze was triggered just before they were going to login for some critical events in an RTS or MMORPG game.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services
108 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games Creating artificial packet loss is relatively easy make a random decision, and either drop the packet or queue it according to whatever latency or rate-cap rules would otherwise be applied. Artificial jitter, on the other hand, is harder to get right. One might think that it is sufficient to simply randomise the latency added to each packet in a flow of packets. Unfortunately, if per-packet latency is implemented naively, this can result in packet re-ordering. Packet re-ordering is highly undesirable, both in real network paths (where it is possible but rare) and in your controlled network testbed. It can cause TCP to aggressively shrink its flow control window, and cause unpredictable reactions from UDP- based applications. To understand the potential for re-ordering, consider the following scenario. You have a router capable of adding a fixed latency to packets flowing through it. The fixed latency is being updated (changed) at intervals of tens to hundreds of milliseconds, and set to values randomly distributed between 90 ms and 110 ms. Re-ordering can occur in the following way: Packet P1 arrives at time T1, and P1 is assigned a delay of 110 ms. The jittering process now modifies the router s fixed latency to 95 ms. Packet P2 arrives at time T2, and is assigned a delay of 95 ms. Packets P1 and P2 are transmitted (forwarded) when their assigned delay periods expire. If T2 T1 > 15 ms packet P2 will emerge after P1 (normal). If T2 T1 < 15 ms packet P1 will emerge after P2 (re-ordered). The key is step 3. If the router s artificial delay mechanism transmits packets at a fixed delay after the packet s arrival, independent of packets which have arrived before or after, re-ordering can occur. Versions of NISTnet introduce re-ordering when used to create artificial jitter in the manner described above. Dummynet may be used to introduce artificial jitter without re-ordering because packets are queued for transmission in the order they arrive [ARM2004B]. [In the example above, P2 would be queued behind, and transmitted after, P1 even if (T2 T1) was less than 15 ms.] One final issue with artificial jitter is the choice of randomised fixed-latency values applied to the dummynet-enabled router. If your jittering function uniformly distributes latency, the resulting RTT can end up having a triangular probability distribution rather than uniform. To a statistician this is obvious the consequence of independent jittering in both directions through the router is the convolution of two uniform distributions, which is a triangular distribution. However, if you use a normal distribution for latency values (probably a more reasonable approximation of real-world jitter) the convolution is also normally distributed [ARM2004B]. 7.2 Communication Models, Cheats and Cheat-Mitigation Wherever a set of rules governs human interactions, some participants will be inclined to ignore, bend or break those rules. Online games are no different in this regard. Human participants bring various personal motivations to the game, not all of which involve playing fair or according to the rules . And when the rules themselves are enforced
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services
Playability versus Network Conditions and Cheats inadequately or incompletely by the computer, there is a temptation to play by what is possible rather than what is intended. This leads us to the difficult definition of cheating. Broadly speaking, cheating could be described thus: Any behaviour that a player uses to gain an advantage over his peer players or achieve a target in an online game is cheating if, according to the game rules or at the discretion of the game operator (i.e. the game service provider, who is not necessarily the developer of the game), the advantage or the target is one that he is not supposed to have achieved. [YAN2005]. Although many cheats do not directly involve the underlying network, all forms of cheating are of interest in the context of online games. When we play human-mediated games face-to-face (for example, a card or dice game), great skill is required to hide cheating from the other participants around the table. When playing LAN-based computer games it is also difficult to hide cheating from your opponents since they will usually be within sight and earshot. However, when playing online, the participants are invariably nowhere near each other and thus unable to verify that everyone s game play experience is consistent with the game s notional rules. Combined with the de-personalising influence of interacting with people through avatars or characters, the temptation to cheat can be substantial. In this section, we will briefly review how cheating can be classified and then discuss examples of cheats involving misuse of server-side, client-side and network-layer technologies. Finally we will review possible methods of cheat-mitigation. 7.2.1 Classifying and Naming Methods of Cheating There is no general agreement on classification and naming of cheats, and we will not attempt to force a particular approach here. It has been observed [YAN2005] that cheats and cheating can be described using three orthogonal attributes. What is the underlying vulnerability (what is exploited?) What are the consequences of cheating (what type of failure can be inflicted?), and Who is doing the cheating? Vulnerabilities can occur in server or client software (bugs), incompletely specified game-state machines (e.g. the game rules programmed into the clients and servers do not cover all possible scenarios), instability of the communication medium (transmission of game traffic can be quite unreliable) or the incomplete hiding of internal game-state information (e.g. ostensibly hidden server-side information leaking to a player via sniffing of network traffic or un-sanctioned modification of the client). Every cheat brings different benefits and risks to the cheater, influencing the likelihood of a given cheat being used. Risk occurs when the cheat creates a change in game play that is easily noticed by other players, or automated cheat-detection schemes are coupled with effective sanctions against players caught cheating. Finally, cheats may be implemented by a single player without anyone s knowledge, they may include other players in collusion, or even include trusted nonplayers who have access to the game server or central game-state databases. 7.2.2 Server-side Cheats Two broad vulnerabilities exist on the server side. First, the server itself may implement game-play rules in a manner that incompletely predicts the space of possible player
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services
Playability versus Network Conditions and Cheats client and getting 62 people to give up their afternoon to play the same map repeatedly. One game for every permutation of controlled network conditions. For example, six possible latencies and three possible loss rates give you 18 repetitions of the same map. Play each map for a realistic 10 15 minutes. Now keep the 62 volunteers focused, alert, fed and somehow independent throughout the entire period. It is hardly surprising that most published lab-based studies of FPS games have yet to come close to this level of statistical accuracy. Finally, one might be tempted to consider a game s own lag-compensation mechanisms to be a source of uncertainty, because it hides the consequences of latency. Some games allow players to turn off lag-compensation on a game-by-game basis. However, if the game is normally shipped with lag-compensation enabled, this will be part of the typical player experience in the face of network latency. Controlled lab trials should mimic the real-world situation as closely as possible, which means leaving in-game lag-compensation enabled. 7.1.3 Considerations for Creating Artificial Network Conditions Controlled test environments often look a lot like Figure 7.3, using either Linux + NISTnet [NISTNET, CAR2003] or FreeBSD + dummynet [DUMMYNET, RIZ1997] for the router that actually instantiates desired latency and loss characteristics. Both approaches have been widely tested and utilised, but they must be used with clear understanding of their underlying limitations. One of the first things to be recognised is that neither FreeBSD nor Linux (in most incarnations) are real-time operating systems. NISTnet and dummynet instantiate bandwidth limits as variable per-packet delays to meet a long-term average rate-cap. They instantiate delays by queuing packets for later transmission. The definition of later is whenever the operating system next schedules the relevant process to check its queues. Thus, an NISTnet-based or dummynet-based controlled latency router may be introducing its own timing errors into the packet streams. For example, dummynet wakes up and processes queued packets once every tick of the FreeBSD system s software clock. At least up to FreeBSD 5.4 the software clock defaulted to 100 ticks per second. In other words, dummynet s activities are quantised to discrete multiples of 10 ms. Configure a 42 ms delay, and packets will experience between 50 60ms of delay (relative to their arrival time) depending on how soon the kernel s software clock ticks after each packet s arrival. Dummynet s accuracy is improved by adjusting the FreeBSD kernel s tick- rate preferably to at least 1000 ticks per second (one tick every 1 ms, the default for FreeBSD 6.0 onwards). This can be achieved either by recompiling the kernel [FBKERN] or adding the following line to the file /boot/loader.conf and rebooting the machine. kern.hz= 1000 Modern motherboards will handle this happily and it reduces dummynet s error to 1 ms. Under these circumstances dummynet is quite a useful and accurate tool [VAN2003]. (Linux kernels have a similar issue with default internal tickrates. At the time of writing, some distributors were beginning to ship products with the default tickrate set to 1000 ticks per second.)
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services
Playability versus Network Conditions and Cheats Local LAN Client 1 Game server Client 4Client 2 Client 3 Internet Router with controllable latency and loss (e.g. NISTnet/Linux or dummynet/FreeBSD) Figure 7.3 A small testbed for controlled latency and packet loss trials latency at the application level. The evidence supports a belief that RTS games are significantly more tolerant of network latency than FPS games success in the game depends more on strategic decisions than sheer reflex and reaction time. Sports games have also been evaluated for latency tolerance. In 2004, a study was published on network latency and EA Sports Madden NFL. Football [NIC2004, MADDEN]. The authors used a small, dedicated lab network (along the lines of Figure 7.3) with two play stations, two consoles and a router running NISTnet for controlled delays. A player s success at running and passing were not noticeably affected until latency reached approximately 500 ms. The authors conclude that this game has latency requirements falling somewhere between the strict demands of typical FPS games and the looser demands of RTS games. 7.1.2 Sources of Error and Uncertainty A number of limitations plague most published research to date in this area. Public-server studies usually lack any real knowledge about every player s external environmental conditions, and experience a limited range of network conditions. Lab-based studies usually only control network conditions and often collect an insufficient number of data points to be statistically rigorous. Rather than being insurmountable problems, these are simply issues to be considered in future work. It is important to test player behaviour under a broad mix of network and environmental conditions. One key limitation of public servers is that you do not get much choice in the range of latency, jitter and loss rates experienced by the players. Some public-server studies have seen network-induced jitter to be almost always less than 20 % of the latency [ARM2004A]. This limits one s ability to draw conclusions about player behaviour under low-latency/high-jitter conditions because such conditions rarely occur on public servers. Public-server trials may also suffer from variability in the RTT estimation algorithms used within each game server. The simplest way to track client latency is to modify each server to log the server s own per-client, dynamically updated RTT estimates.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services
106 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games Relative frequency versus latency 0 5 10 15 20 25 30Relative frequency (%) Ping perceived network latency Server perceived network latency 195 200 205 210 215 220 225 Latency (ms) Figure 7.4 Measuring RTT with ping versus a Quake III Arena server s internal mechanism Unfortunately, the game server s RTT estimation occurs in the application space rather than kernel space, and is not necessarily strictly designed for accuracy. Figure 7.4 shows an example where a Quake III Arena server was separated from a client by a router providing configurable network delay [ARM2004A]. Regular pings (ICMP echo/reply) were run concurrently over the same link. The ICMP echo/reply technique reveals path RTT tightly sitting between 199 200ms. In contrast, the Quake III Arena server s internal RTT estimates range between 198 214ms (with most between 200 210ms) the median RTT reported by the game server is 5ms higher than the network path s RTT, and suffers from some self-inflicted jitter. Creating statistically relevant lab-based studies can also be rather problematic. It is not uncommon for researchers to believe that they, and a handful of close friends or colleagues, are representative of all players. Unfortunately, the rules of statistics say that many tens, if not hundreds, of independent repeat trials are required before one can draw solid conclusions. For example, consider a generic study of player tolerance for latency and loss. A game is played with a fixed latency and loss, the player experiences are recorded, and then a new game is played again with a different latency and loss. If the estimated standard deviation of measurements for any given latency and loss settings was one (1) and we wanted to achieve 95 % confidence intervals with an error of 0.25, we would need at least 62 independent samples [ZAN2004]. For each and every latency and loss setting, that is a lot of samples. Particularly when you realize that, strictly speaking, independent samples means using different players, not simply the same player repeating the game a number of times. Practical considerations come into play when aiming for statistically valid lab-based studies. Consider the logistics of setting up 62 PCs with 62 licensed copies of an FPS game
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services
104 Networking and Online Games: Understanding and Engineering Multiplayer Internet Games games played by people, which have over 150 180ms latency, reflects an upper bound of tolerance for players of Quake III Arena. Unfortunately, the 2001 Quake III Arena trials lacked any information on player motivations and personal environmental conditions. They also did not measure the network jitter and packet loss rates being experienced by players who experienced particular median latencies. Thus Figure 7.1 and Figure 7.2 are more illustrative of what we might expect, and should not be taken as hard data. Another study in 2001 was based on a public Half-Life server in the UK [HEN2001]. For over three and a half weeks the maps rotated every 60 minutes, and they saw 31,941 instances of a player joining and then leaving the server, attributable to 16,969 unique players. This study concluded that latencies around 225 250ms dissuade players from joining a server, but below that level there was no strong evidence of absolute latency alone being a predictor of a player s likelihood of joining, staying or leaving a server. Another key conclusion was that simplistic correlation of data measured from public servers is fraught with potential errors. Early research results led to the development of more sophisticated testbeds. In a 2003 follow-up to [HEN2001], two identical (with respect to configuration and location) public Half-Life servers were augmented with a Linux-based network switch. Using the Linux iptables and libipq functionality [IPTABLES], this switch could add controlled latency to particular client traffic arriving over the Internet [HEN2003]. Player behaviour over weeks and months was monitored as latency was alternately added to one server and then the other. When one server had 50 ms added there was a definite player-preference towards joining the server with lower latency at any given time. Interestingly, when latency was added to an individual player after they had started playing, the likelihood that the player would subsequently leave seemed to decrease the longer that player had been playing. In another example, the latency and packet loss tolerance of Unreal Tournament 2003 (UT 2003, an online multiplayer FPS game) was publicised in 2004 [BEI2004]. The authors established a controlled network environment of four client hosts and a local UT 2003 server (somewhat like Figure 7.3). A Linux-based network router running NISTnet [NISTNET] was used for introducing configurable latency and packet loss to each client. Results suggested that players were unaware of packet loss rates of up to 5 %, and had only limited latency sensitivity when performing simple or complex movements. However, a distinct latency sensitivity emerged when using weapons, and players using more precise weapons were more impacted by increases in latency. [The paper s authors concluded … even modest (75 100ms) amounts of latency decreases accuracy and number of kills by up to 50 % over common Internet latency ranges. and noted that latencies as low as 100 ms were noticeable and latencies around 200 ms were annoying .] Of course, FPS is not the only game genre of interest. In 2003, a study was published on the latency tolerance of players in Warcraft III a multiplayer, online Real-Time Strategy (RTS) game [SHE2003]. A controlled network environment (rather like Figure 7.3) was established, with two local Warcraft III clients and a local Warcraft III server. Latency was varied over multiple games while monitoring player behaviour during distinct build , explore and combat phases within the game. The authors concluded that … overall user performance is not significantly affected by Internet latencies ranging from hundreds of milliseconds to several seconds. In addition, Warcraft III uses TCP rather than UDP for its underlying transport protocol, ensuring that packet loss is translated into additional
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Mac Web Hosting services