Measurement and Estimation of Network QoS Among Peer Xbox 360 Game Players
Youngki Lee1 , Sharad Agarwal2 , Chris Butcher3 , and Jitu Padhye2
KAIST Microsoft Research 3 Bungie Studios

1 Introduction

The research community has proposed several techniques for estimating the quality of network paths in terms of delay and capacity. However, few techniques have been studied in the context of large deployed applications. Network gaming is an application that is extremely sensitive to network path quality [1,2,3]. Yet, the quality of network paths among players of large, wide-area games and techniques for estimating it have not received much attention from the research community. Network games broadly fall into two categories. In some games (e.g. MMORPGs, web-based casual games, Quake) with a client-server architecture, players communicate with a large, well-provisioned, and dedicated game server [4,5]. In some games with a peer-to-peer (P2P) architecture, players communicate with each other directly or via a dynamically chosen peer at some players house. In Ghost Recon, Halo series, and others for the Xbox and Xbox 360 consoles, a server assists players in discovering other peers to host and play with. Accurate and scalable estimation of the network path quality (NPQ) between peer game players is especially critical for games with a P2P architecture. These players need to have good network connectivity to each other, so accurate NPQ data is essential for matchmaking - i.e. to determine which players should play with each other. Furthermore, NPQ estimation needs to be done in a scalable manner. If the number of peers is large, it may not be not feasible to probe all of them. Prior research on P2P games has used data from only a small number of players [6]. We study a much larger data set, from Halo 3 : a popular Xbox 360 game. We cover 5.6 million unique IP addresses that played 39.8 million game sessions in 50 days. Peers in each game session gather NPQ data and report it to the central Xbox server for matchmaking purposes. This paper makes the following contributions: We present data from a large P2P gaming application. The population is several orders of magnitude larger, and far more geographically diverse than any previously reported study. Given the number and geographical diversity of players, we consider this to also be a large-scale study of path quality over the wide-area Internet. We study temporal and geographical correlations in the NPQ data, and propose three different predictors that can provide a rough estimate of NPQ between a pair of players, without requiring any probing. There can be millions of game players
M. Claypool and S. Uhlig (Eds.): PAM 2008, LNCS 4979, pp. 4150, 2008. c Springer-Verlag Berlin Heidelberg 2008

Y. Lee et al.

on-line at any time, and any techniques that can avoid having to perform network probes between all of them can not only reduce network overhead but also reduce the amount of time players have to wait before starting a game over the Internet.

2 Background

The Microsoft Xbox 360 game console supports on-line game play with the Xbox Live subscription service. The Halo series of First Person Shooter (FPS) games has sold over 15 million copies worldwide. We focus on the latest edition, Halo 3. Each Halo 3 Internet game session can include up to 16 players. One console in each game session is selected to be the game host or server. All game communication between other players is relayed through this host console. The Xbox Live server provides accounting and matchmaking services. Therefore, the NPQ between consoles and the Xbox Live server is less important to the overall gaming experience than the NPQ between the consoles themselves. An excellent Halo 3 experience has under 50ms of latency and 50Kbps to 70Kbps of bandwidth between each client console and the host console. Note that the host console may consume up to 1Mbps ((16-1)*70Kbps) of bandwidth. A good experience can be achieved with 150ms latency and 30Kbps of bandwidth. Hence, it is important to group consoles so that they each have good NPQ to the host console. This is critical in this architecture because the host is a fellow players console, typically on a consumer broadband connection, and not a well provisioned, dedicated server. The Xbox Live server helps with matchmaking - setting up such groups, of up to 16 players, from among the hundreds of thousands on-line at any time. A player who starts an Internet game session will sign on to the Xbox Live service and run Halo 3 on her console. She will select Internet game play and can specify several criteria for the session, such as the type of game (e.g. free for all or team objective). With some probability, this console will become a peer game host, instead of a game client. This probability depends on the chosen type of game. If the console is a game host, it will wait for other consoles to discover it, probe their NPQ to it, and join the game. If it is a game client, Xbox Live will send it IP addresses for the other consoles on the Internet that are hosting games of the specied type. This console will send and receive several packet pairs with each IP address. The Xbox 360 networking stack implements the standard packet pair estimation technique [7]. Packet pairs are performed serially and do not overlap with each other. The console will then have an estimate of the round-trip latency (RTT), and the upstream and downstream network capacity with each candidate game host. While being very lightweight, packet pair measures bottleneck link capacity but not available bandwidth. These values are logged by the Xbox Live service. The user is shown those host consoles that it has the best NPQ to. For conciseness, we leave out several details such as NAT traversal. Little is known about the population of on-line P2P game players. Their geographic diversity, diurnal behavior, typical network delay and capacity are useful parameters to network models of game systems for future research. This information can help build estimators of NPQ between any two game consoles on the Internet. Even merely identifying the pairs of consoles with extremely poor NPQ can signicantly reduce the total number of probes, thereby reducing network overhead and user wait time.

3 Data and Methodology

Xbox Live stores information about every Internet game session for Halo 3. In a typical week ending on 29 January 2008, we nd that 72.5% of Internet game sessions required matchmaking; when weighted to account for players per game, it is 83.5%. By a session, we mean an attempt to search for an Internet game - the user may have eventually joined a game or decided not to. The log has the UTC time and the public IP address of the console searching for a game. This console may have probed several other consoles that were hosting games of the requested type - for each probe to a candidate host console, we have the host IP address, median round trip time (RTT), and average capacities upstream to host and downstream from host. We use the term probe to mean 4 packet pair tests from the client console to a host console and 4 in the reverse direction. We use player, console and IP address interchangeably.

Table 1. Data sets

Halo 3 Phase Internal alpha Internal beta Public beta Release Start 11/30/2006 05/08/2007 05/22/2007 11/14/2007 End Distinct IPs Matchmaking games Hosts probed 01/23/2007 4,025 314,606 207,595 05/21/2007 732,487 20,747,695 33,338,060 06/11/2007 903,782 23,182,323 38,453,035 01/03/2008 5,658,951 39,803,350 126,085,887
Fig. 1. Geographic distribution of players
Fig. 2. Game sessions per hour
Table 1 lists our data sets. For conciseness, we focus on the Release data set for Halo 3. Due to the extremely large number of game plays, we limit the data set in two ways - we consider a 50 day period and we only consider a randomly selected 20% of the matchmaking game sessions. The resulting data set covers over 126 million probes between over 5.6 million IP addresses. For geographic analysis, we use the commercial MaxMind GeoIP City Database from June 2007. It was able to provide the latitude and longitude for over 98% of these IP addresses.
4 Player Population Characterization
In this section, we analyze the basic characteristics of the player population, such as the geographic distribution of the players, when and how often they play the game. We also look at the overall NPQ data such as distributions of RTT and capacity.

cumulative freq. (IPs)

cumulative freq.(IPs)

latitude

longitude
Fig. 3. Latitude and longitude density of players

1048576 65536

16384 cumulative freq. (sessions) 262144 1

# of IP addresses 1024

# of sessions

# of probes

Fig. 4. Game sessions per IP address (log-log)
Fig. 5. Probes per session (log-log)
Figure 1 shows the geographic locations of all 5,658,951 unique IP addresses, which correspond to 68,834 unique latitude and longitude coordinates. To examine the density of players in each region, we present Figure 3. Almost 85% of players are in USA longitudes -130 to -60, and latitudes 30 to 50. Roughly 15% are in western Europe. Since players are spread across this large geographic region, it is quite possible for consoles that are too far apart to probe each other. This partly motivates us to consider estimation techniques that will identify such cases before probing. To see when games were played, Figure 2 plots the number of game sessions in each hour over a representative week. We notice a very strong diurnal pattern with peaks in the evenings in North American time zones - this is not unexpected given the high density of players in USA. We examine game playing behavior in more detail in Figure 4. The number of games attempted from each IP address almost follows a Zipf distribution. In the far right of the graph, one IP address attempted 5438 sessions - over a 50 day period, this is a huge number of games for any 1 individual! We suspect that the IP addresses in this area of the graph are for proxies with many players behind them. Figure 5 shows a CDF of the number of consoles hosting a game that were probed in each session. While there are many sessions that probed few consoles, there are some that probed as many as 400 consoles. This number depends on how many game hosts the Xbox Live server gives a console requesting a game, which in turn depends on how many consoles are available at the time and the type of game requested. Now we consider overall NPQ data. Figure 6 shows the CDF of RTT across all probes. Over 25% of the measurements are above 150ms, which is an upper bound for a responsive user experience in typical FPS games [1]. We want to pre-determine in which cases the RTT will be above 150ms and skip probing altogether, thereby
1 cumulative freq. (probes) frequency (x1,000,000)

192 Kbps 10Mbps

1.6Mbps 5.8 Mbps

100 delay (ms)

capacity (kbps)
Fig. 6. RTT delay reported by probes (log-log) Fig. 7. Downstream capacity reported by probes
potentially reducing the total number of probes by 25%. Figure 7 shows the distribution of measured capacity across all probes, in the direction from the console hosting a game to the console requesting to join it. The graph for upstream capacity is similar. We see peaks around typical capacities for broadband access in USA (e.g. 192Kbps, 1.5Mbps), within some marginal error due to the packet pair estimation technique.

5 NPQ Prediction

The NPQ probing technique that Halo 3 uses consists of 16 packets per console being probed (4 packet pairs in each direction). However, there can be many candidate hosts for a game. For scalability and to minimize user wait time, we want to reduce the total number of probes and hence propose the use of NPQ predictors. Our goal is to estimate apriori if a console has good NPQ to a remote candidate host console, without doing a probe. If bad NPQ is predicted, then this candidate should not be probed. If good NPQ is predicted, then limited probing can be done (e.g. only 1 packet pair). If no prediction can be made, standard probing should ensue. Based on our analysis of the NPQ data, we now propose and evaluate three NPQ predictors. 5.1 IP History Predictor We hypothesize that a probe between a pair of consoles at time t1 produces an NPQ estimate that is still valid at a later time t2. This may be true if the median RTT and average upstream and downstream bottleneck capacities do not vary signicantly over a period of = t2 t1. To test this hypothesis, and estimate how large can be, we examine NPQ data for pairs of IP addresses over different periods of time. Figure 8 shows the CDF of the coefcient of variation (CV) in RTT for pairs of IP addresses over different time windows. For instance, the Within 5 min line shows the CV of RTTs from probes between the same pair of IP addresses within any 5 minute window. To draw meaningful conclusions, we consider only those IP pairs that probed each other at least 5 times during that period. We have plotted similar lines for 30 minutes, 6 hours, 1 day and the entire 50 day trace (the line labeled no constraints). The lines for all 5 time windows overlap with each other. For over 90% of IP address pairs that probed each other at least 5 times, the variation in RTT estimates was minuscule (CV under 0.2), even up to a of 50 days. For comparison we plot the baseline instead of considering CV for each pair of IPs, we consider the CV for each single IP.
1 cumulative freq. (src-dst IPs) 0.8
cumulative freq. (src-dst IPs) 0.9 0.8

0.9 0.7

0.6 0.5 Within 5 min Within 30 min Within 6 hr Within 1 day No Constraints Baseline

0.7 0.6

0.5 0.4 Within 5 min Within 30 min Within 6 hr Within 1 day No Constraints Baseline 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.More coefficient of variation

0.3 0.2

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.More coefficient of variation
Fig. 8. RTT variation for IP pairs
Fig. 9. Capacity variation for IP pairs
That is, for each IP address, the CV across all RTT estimates where this IP address was involved, across the entire trace. This line is well below the others, indicating that the RTTs are spread across a wide range. We conclude that the IP History Predictor can perform quite well for predicting RTT, and the can be as large as 50 days. Figure 9 has the same graph for downstream capacity. The upstream capacity graph is very similar and is omitted for conciseness. Again, does not affect the NPQ prediction. While the CV is larger, it is under 0.65 for 90% of the IP pairs, and is still much higher than the baseline. Thus we believe the IP History Predictor adequately predicts the NPQ between a pair of consoles based on an NPQ estimate from a prior probe. 5.2 Prex History Predictor We have shown the IP History Predictor to work only when pairs of consoles have probed each other in the past. This may reduce the number of probes in only a limited set of cases. Thus we now propose the Prex History Predictor - this is similar to the IP History predictor, except it uses IP prex pairs. We hypothesize that a probe between a pair of consoles A1 and B1 at time t1 produces an NPQ estimate that is still valid at a later time t2 for a different pair of consoles A2 and B2 , as long as A1 and A2 belong to one BGP prex, and B1 and B2 belong to one BGP prex. This predictor may be accurate if consoles in the same prex share similar last mile access. However, broadband ISPs offer several access packages (e.g., 192Kbps DSL or 1.5Mbps DSL), and the prex may indicate geographic location more than link speed. Thus, predictions for capacity may be less accurate than for RTT. We now analyze NPQ data for pairs of IP prexes that probed each other at least 5 times. We nd a consoles prex by a longest prex match against the 12/27/2007 RouteViews BGP table [8]. Figure 10 shows the performance of this predictor for delay, and can be compared to Figure 8. When considering prex pairs, has a bigger impact - the older the original probe, the worse the prediction. Since CV is a relative metric, small variations in small RTTs (e.g. 5ms versus 10ms) can produce a large CV. Thus in Figure 11 we look at the semi-interquartile range (SIQR) of RTT estimates for prex pairs for no limit on (i.e., the no constraints case). For 90% of prex pairs, the SIQR is under 40ms. Thus it is the outliers beyond the 25%-75% SIQR that contribute to this additional variability. Figure 12 shows the performance of this predictor for downstream capacity estimation. For beyond 5 minutes, it is a very poor predictor. We suspect this is due to different subscription levels of last mile capacity within the same prex.

1 cumulative freq. (src-dst prefixes) cumulative freq. (src-dst prefixes) 1 0.9 0.8 0.7 0.6 0.5
Within 5 min Within 30 min Within 6 hr Within 1 day No Constraints Baseline 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.More

0.4 0.3 0.2 0.1 0

SIQR(25%-75%)

10%-90%

coefficient of variation
Fig. 10. RTT variation for prex pairs
cumulative freq. (src-dst prefixes) 0.9 0.8
Fig. 11. SIQR of RTT for prex pairs
0.5 0.4 0.3 Within 5 min Within 30 min Within 6 hr Within 1 day No Constraints Baseline 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.More
cumulative freq. of pairs

0.3 0.2 0.10

Fig. 12. Capacity variation for prex pairs
Fig. 13. Number of repetitive probes
Based on these results, we believe the Prex History Predictor can adequately predict the RTT between a pair of consoles based on an NPQ estimate from a prior probe between consoles in the same pair of prexes. However, this prediction is not as accurate as the IP History Predictor - so we suggest rst applying that predictor, and only if it cannot make a prediction, then using the Prex History Predictor. To show in how many cases this would apply, we present Figure 13. We plot the CDF of the number of repeated probes in the entire trace between the same pair of IP addresses, and the same pair of prexes. Only about 5% of pairs of consoles probed each other more than once, while about 39% of prex pairs probed each other more than once. Note that we have clipped the horizontal axis at 30 for presentation purposes - the IP pair line continues to 114 probes, and the prex pair line continues to 14,513. 5.3 Geography Predictor While 39% of prex pairs is still a signicant fraction of the number of consoles, and has the potential to reduce a far larger portion of probes those prexes probed each other several times, there is still about 61% of prex pairs left. We now consider the Geography Predictor - we hypothesize that the geographic distance between two consoles has a strong correlation with their RTT, and that current databases for mapping IP addresses to geographic coordinates are reasonably accurate for this. This may be true if distant IP addresses traverse longer links and more router hops to communicate. This predictor does not consider past history, and thus could be applied to any pair of consoles.
More size of range (ms) IP Pair Prefix Pair # of probes 25 30
cumulative freq. (probes) 0.8 0.7

0.4 0.3

distance (miles) 10000 12000
Fig. 14. Probe distance distribution
Fig. 15. Distance-RTT correlation
It is not clear why geographic distance would be correlated with down/up stream bottleneck capacity - our analysis indicates the correlations are -0.075 and -0.097, respectively. Thus we omit detailed results on capacity for conciseness and focus on RTT. We use a MaxMind location database to convert the source and destination IP addresses in a probe to the geographic coordinates, and apply the great circle distance algorithm [9] to calculate distance. The distribution in Figure 14 shows a wide range of distances between probe hosts. About 14% of probes traversed over 5,000 miles, which indicates there is room for optimization by ltering out these console pairs from the probe list. The graph also shows that we have enough samples to examine the correlation between distance and delay. Figure 15 plots the correlation between distance and RTT for 100,000 randomly selected probes (we were unable to plot over 126 million points on the same graph different samples of 100,000 points gave us very similar graphs). We see a very strong correlation between the geographic distance and minimum RTT. However, there is a lot of noise above that minimum, which may be due to queuing delays and sub-optimal routes. We conclude that the Geography Predictor is useful for ltering out pairs of IP addresses that are too far apart to have a low RTT. 5.4 Using Predictors in Matchmaking Incorporating these three predictors into matchmaking is not difcult. For the IP History Predictor, each console will keep a history of previous probes that it was involved in. It can look up this history before attempting any future probes, and decide which candidate game hosts to ignore. For the Prex History Predictor, the Xbox Live server can lter the set of candidate game hosts it provides to each console based on their prexes and past probe history. The server already has the past NPQ estimates, and it can easily keep fresh BGP tables to look up prexes. The Geography Predictor requires an IP to geographic coordinate database, on either the Xbox Live server or on each console.

6 Prior Work

Most prior work on network gaming has focused on games with a client-server architecture [10,4,5] where the server is well-provisioned and dedicated. The literature onP2P games is very limited. In [6], the authors examine game clients deployed in three access networks: dial-up, cable and ADSL. However, their experiments are limited to one

client per access network, and use only one cable and one ADSL link. The game trafc of Halo 2 is analyzed in [11] in a LAN environment for trafc modeling and not for end-to-end network characteristics between real Halo 2 players. There has been much prior work on efcient and accurate NPQ prediction. For conciseness, we identify those done in the context of network gaming. As before, most of this work is for client-server games. In [12], a simple server-side method is proposed to improve server location discovery and reduce probe trafc. Our NPQ prediction methods focus also on reducing overall probe time since that directly affects user wait time. Also, we not only utilize the geographic location of consoles but also previous probe results. A ooding-style server discovery mechanism is proposed in [13] to quickly locate local servers and prevent single directory server failure. That does not scale to P2P games, since in our case several hundreds of thousands of consoles can be on-line at any time. A server selection algorithm is proposed in [14] for distant game players who want to play specically with each other. Our work considers the general case of joining players to any acceptable game, and thus considers NPQ data and correlators across all on-line consoles. The geographic distribution of game servers and players is used in [15] to redirect players to close game servers. While [16] does not consider on-line games, they correlate geographic location and network delay to nd a host, and their experimental result about the correlation complements ours. Outside network games, there has been a lot of research on characterizing NPQ over the Internet. Many of these [17,18] use PlanetLab nodes. They are mostly located in high-performance and stable academic networks, and thus do not reect the characteristics of consumer access networks. In [19], the constancy of NPQ over time among 31 hosts is studied within a stable academic network. Our work signicantly complements prior work in terms of scale and diversity of network connectivity. Studies of hosts behind consumer broadband lines are rare. It is extremely difcult to build a large testbed of such hosts on the Internet. While [20] characterizes network performance between consumer broadband hosts, they use only 25 hosts. More recently, [21] studies residential network link capacities, RTT, and loss rates through relatively large-scale measurement studies. They use 1,894 hosts behind 11 major DSL and cable providers in North America and Europe. Our study is much larger in scale, involving over 5.6 million hosts. Furthermore, they do not characterize direct network connections between pairs of broadband hosts since they measure from several vantage points located in academic networks. Techniques for estimating NPQ have been studied extensively [7,22]. Our work focuses not on the techniques itself, but on the NPQ data.

7 Conclusions

We studied the quality of network paths among Xbox 360 game consoles playing Halo 3. We focused on network delay and capacity measured between players prior to each Internet game match-up. We studied the general characteristics of the player population such as geographical diversity and diurnal patterns of game play. We leveraged our understanding of these characteristics to propose three predictors for determining path quality without additional probe trafc : IP and prex history-based and geographybased. Our evaluation of these predictors showed that they can signicantly reduce the
number of probes and hence user wait time during matchmaking. For future work, we plan on comparing the initial NPQ estimate to actual in-game network performance.

References

1. Dick, M., Wellnitz, O., Wolf, L.: Analysis of factors affecting players performance and perception in multiplayer games. NetGames (2005) 2. Quax, P., Monsieurs, P., Lamotte, W., Vleeschauwer, D.D., Degrande, N.: Objective and subjective evaluation of the inuence of small amounts of delay and jitter on a recent rst person shotter game. NetGames (2004) 3. Armitage, G.: An experimental estimation of latency sensitivity in multiplayer Quake 3. In: ICON (2003) 4. Feng, W., Chang, R., Feng, W., Walpole, J.: Provisioning on-line games: A trafc analysis of a busy Counter Strike server. In: IMW (2002) 5. Kim, J., Choi, J., Chang, D., Kwon, T., Choi, Y., Yuk, E.: Trafc characteristics of a massively multi-player online role playing game. NetGames (2005) 6. Jehaes, T., Vleeschauwer, D.D., Coppens, T., Doorselaer, B.V., Deckers, W.N.E., Spruyt, J., Smets, R.: Access network delay in networked games. NetGames (2003) 7. Carter, R.L., Crovella, M.E.: Measuring bottleneck link speed in packet-switched networks. Technical report, Boston University (March 1996) 8. University of Oregon: Routeviews project page http://www.routeviews.org/ 9. Hexa software development center: Distance calculation method between two latitude and longitude coordinates, http://zipcodeworld.com/docs/distance.pdf 10. Chambers, C., Feng, W., Sahu, S., Saha, D.: Measurement-based characterization of a collection of on-line games. In: IMC (2005) 11. Zander, S., Armitage, G.: A trafc model for the Xbox game Halo2. In: NOSSDAV (2005) 12. Zander, S., Kennedy, D., Armitage, G.: Server-discovery trafc patterns generated by multiplayer rst person shooter games. NetGames (2005) 13. Henderson, T.: Observations on game server discovery mechanisms. NetGames (2003) 14. Gargolinski, S., Pierre, S., Claypool, M.: Game server selection for multiple players. NetGames (2005) 15. Chamber, C., Feng, W., Feng, W., Saha, D.: A geographic redirection service for on-line games. ACM Multimedia, New York (2003) 16. Fdida, S., Duarte, O.C.M.B., de Rezende, J.F., Ziviani, A.: Toward a Measurement-Based Geographic Location Service. In: Barakat, C., Pratt, I. (eds.) PAM 2004. LNCS, vol. 3015, pp. 4352. Springer, Heidelberg (2004) 17. Lee, S.-J., Basu, S., Sharma, P., Banerjee, S., Fonseca, R.: Measuring Bandwidth Between PlanetLab Nodes. In: Dovrolis, C. (ed.) PAM 2005. LNCS, vol. 3431, pp. 292305. Springer, Heidelberg (2005) 18. Banerjee, S., Grifn, T.G., Pias, M.: The Interdomain Connectivity of PlanetLab Nodes. In: Barakat, C., Pratt, I. (eds.) PAM 2004. LNCS, vol. 3015, pp. 7382. Springer, Heidelberg (2004) 19. Zhang, Y., Dufeld, N., Paxson, V., Shenker, S.: On the constancy of Internet path properties. In: IMW (2001) 20. Lakshminarayanan, J., Padmanabhan, V.N.: Some ndings on the network performance of broadband hosts. In: IMC (2003) 21. Dischinger, M., Haeberlen, A., Gummadi, K.P., Saroiu, S.: Characterizing residential broadband networks. In: IMC (2007) 22. Dovrolis, C., Ramanathan, P., Moore, D.: Packet-dispersion techniques and a capacityestimation methodology. IEEE/ACM Transactions on Networking (December 2004)

Fire Weapon Reload/Action Switch Weapons Melee Attack Jump Swap Grenades Flashlight Zoom Scope (Click)

ONLINE ENABLED

Throw Grenade E-brake (Warthog) Boost (Vehicles) Crouch (Click) Score Pause Game Settings
Get the strategy guide primagames.com

0904 Part No. X10-96235

SAFETY INFORMATION
About Photosensitive Seizures
A very small percentage of people may experience a seizure when exposed to certain visual images, including ashing lights or patterns that may appear in video games. Even people who have no history of seizures or epilepsy may have an undiagnosed condition that can cause these photosensitive epileptic seizures while watching video games. These seizures may have a variety of symptoms, including lightheadedness, altered vision, eye or face twitching, jerking or shaking of arms or legs, disorientation, confusion, or momentary loss of awareness. Seizures may also cause loss of consciousness or convulsions that can lead to injury from falling down or striking nearby objects. Immediately stop playing and consult a doctor if you experience any of these symptoms. Parents should watch for or ask their children about the above symptomschildren and teenagers are more likely than adults to experience these seizures. The risk of photosensitive epileptic seizures may be reduced by sitting farther from the television screen, using a smaller television screen, playing in a well-lit room, and not playing when you are drowsy or fatigued. If you or any of your relatives have a history of seizures or epilepsy, consult a doctor before playing. Other Important Health and Safety Information The Xbox Instruction Manual contains important health and safety information that you should read and understand before using this software.

TABLE OF CONTENTS

Secret Transmission... 2 Master Chief.... 3 Breakdown of Known Covenant Units.. 4 Controller.... 6 Mjolnir Mark VI Battle Suit HUD... 8 Battle Tactics.... 9 Current UNSC Weaponry...10 Covenant Weapons...13 UNSC Ground Transport...16 Intel on Covenant Vehicles....18 Settings.....20 Multiplayer Basics....21 Xbox Live....24 Gaming with Friends!....26 Multiplayer Strategy....28 Bungie.net...29 Credits....30 Warranty....32 Technical Support...33
Avoid Damage to Your Television
Do not use with certain televisions. Some televisions, especially front- or rear-projection types, can be damaged if any video games, including Xbox games, are played on them. Static images presented during the normal course of gameplay may burn in to the screen, causing a permanent shadow of the static image to appear at all times, even when video games are not being played. Similar damage may occur from static images created when placing a video game on hold or pause. Consult your television owners manual to determine if video games can be played safely on your set. If you are unable to nd this information in the owners manual, contact your television dealer or the manufacturer to determine if video games can be played safely on your set.
Choose your weapon, nd your friends, save the Earth!
Dont miss the best selling novel series from Del Rey, www.delreydigital.com, based on the hit Xbox and PC games. Halo: The Fall of the Reach, by author Eric Nylund, explores the history of the heroic super-soldier, the Master Chief. Halo: The Flood, by author William C. Dietz, continues the story as the Master Chief struggles to unlock Halos secrets. The Human-Covenant war rages on as the alien juggernaut sweeps inexorably toward its nal goal: destruction of all human life in Halo: First Strike, by author Eric Nylund. Go beyond the game with action gures from JoyRide Studios, www.joyridestudios.com; Xbox strategy guide from Prima, www.primagames.com; and the ocial game soundtrack from Sumthing Else MusicWorks, Inc., www.sumthing.comor look for em at www.bungiestore.com.

<known casualties, incurred losses, unaccounted records encrypted/enclosed#####>
Re: SPARTAN-117 MASTER CHIEF Spartan 117, the Master Chief, is a member of the SPARTAN-II project. He is a genetically, biologically, and technically enhanced ghting unit, standing seven feet tall and weighing half a ton in his armor. His reexes are unmatched, his strength and endurance quite unlike any other human and his tenacity molded by a lifetime of conscripted military training. The Master Chief is procient in all current ballistic weapons and tactics, incursion, and unarmed combat, and has extensive experience with Covenant military tech.
<information on Flood parasite morph/biology encrypted/ enclosed#####>
Through a connection to Halos deep data cores, I ascertained that the ring was built by an ancient race of beings (referred to by the Covenant as The Forerunners) as a weapon of last resort against the Flood. A sizable population of Flood was in stasis on the ring, and the Covenant, either by accident or design, released the parasite. Not their most brilliant maneuver.
<information on Forerunner artifacts, structures, symbols encrypted/enclosed#####>
343 Guilty Spark convinced SPARTAN 117 to activate Halos primary weapon system and eliminate the Flood. The AI neglected to tell him however, that because the Flood consumes any suitable sentient host, Halo would make no distinction between the Flood and other life forms. In short: if Halo red, it would destroy every thinking being in the galaxy. Human, Covenant, everything.
BREAKDOWN OF KNOWN COVENANT UNITS
Covenant society is highly segmented, consisting of a confederation of races. While an overall socio-political review is important for understanding the nuances of Covenant society, including the role of Prophets, the key for combating our enemy is to review the Covenant ghting classes which are comprised of Grunts, Jackals, Hunters, Elites, Brutes, Drones, and Prophets.

Elites

The Elites are the core of the Covenant military. Excellent soldiers, brilliant tacticians, and disciplined, aggressive ghters, they are the primary strength of the Covenant force. Faster, stronger, and tougher than Humans, they ght in relatively small numbers but often lead squads of Grunts. Armor color seems to indicate rank, and we believe Elites are promoted based on numbers of casualties they inict.

Grunts

The basic infantry unit of the Covenant, Grunts are dangerous in groups but present little threat individually. Short, stocky, and relatively slow, they will often panic when faced with superior forces. However, if they are being led by an Elite, they will stand and ght.

Brutes

Not as readily understood as Elites, Brutes ght together in a pack and are physically stronger. Brutes demonstrate similar battleeld abilities to Elites, and their numbers have demonstrably increased since the conict began. They carry a ballistic explosive weapon with an attached bayonet device.

Main Menu

From the Main Menu, select Campaign to begin a solo or cooperative game. Select Xbox Live, Split Screen, or System Link to start a multiplayer game. Select Settings to customize player proles or modify game types.

Campaign

You can play a campaign as a single player in a solo campaign or with a friend in cooperative campaign. You play a cooperative campaign in split screen mode. A campaign is associated with a player prole. To start a new campaign, you need to create a new player prole. To start a solo campaign, select Campaign from the Main Menu. Select New Campaign, select a diculty, and then press A. To start a co-op campaign, select Campaign from the Main Menu, and then select Cooperative. Both players need to select a player prole and then press A.

Weapon Indicators

Mjolnir Mark VI armor can monitor the status of two weapons simultaneously. A left-wielded weapon and ammo display on the left of the HUD, while right-hand or single-use weapons display on the far-right portion of the HUD. When a single weapon is wielded, status on your grenade types displays on the left side. Since SPARTAN-117 can keep a weapon in reserve, it is indicated on the right side.
Saving Progress and Loading a Level
Your progress in a campaign is saved automatically at specic checkpoints throughout a game level. To continue a campaign from your last saved checkpoint, select Campaign from the Main Menu, and then select Resume. To load a specic level, select Campaign, and then select Select Level. You can select only levels youve already conquered.

Motion Tracker

Indicates relative whereabouts of allies and hostile combatants. Located on the lower-left portion of the HUD, it is tuned to detect aggressive or obvious motion and cannot show the location of stationary or slow-moving hostiles.

Shield Indicator

A bar above the motion tracker. Solid blue indicates an optimal state.

Warning Indicators

These display below your reticle to provide valuable information.

Reload

Low Ammo

Low Battery

No Ammo

No Battery

No Grenades

End transmission.

CURRENT UNSC WEAPONRY
Since you all will be required to use non-specialist equipment in these dicult circumstances, we are issuing you a refresher on the current available UNSC arsenal. If you have not used any of these weapons in the eld or in practice, you would do well to familiarize yourself with them before entering a combat zone. Some of these weapons you can dual-wield. enabling you to re them simultaneously. To dual-wield, press and hold Y to pick up the secondary weapon. Use the Left and Right triggers to re.
Ammo Capacity: 12 rounds per magazine.

S2 AM Sniper Rie

Gas-operated, magazine-fed weapon utilizes a smart-linked scope with two levels of magnication (5X and 10X). Firing 14.5mm armor-piercing, n-stabilized, discarding-sabot rounds makes it very powerful. Sheer size and limited magazine capacity dictate strategic use. Devastating from a secured position, has clearly limited use as close-range weapon.

Ammo Capacity: 4 rounds per magazine.

M6C Pistol (dual wield)

Standard UNSC sidearm. Recoil-operated magazine-fed handgun, ring a magazine of six 12.7mm semi-armor piercing rounds. Fired accurately in semi-automatic mode, can be powerful anti-personnel weapon. This strippeddown C variant does not feature a scope.

M19 SSM Rocket Launcher

Most commonly used light anti-vehicle weapon in UNSC arsenal. Man-portable, shoulder-red, with single 2X level of magnication. Fires 102mm shaped-charge, high explosive tracking rockets. Reticle indicator denotes when launcher has achieved lock on target.
Ammo Capacity: 2 102mm shaped-charged rockets.
Ammo Capacity: 36 rounds per magazine.

BR55 Rie

Battle Rie res 9.5mm rounds from 36-round magazine. Mounted with a 2X optical scope for targeting. Fires in short, automatic bursts of three rounds. Very accurate, relatively high rate of re makes it a useful all-around infantry weapon.

M90 Shotgun

Ammo Capacity: 60 rounds per magazine.
Ammo Capacity: 12 8-gauge shotgun shells.
M7/Caseless Sub Machine Gun (SMG) (dual wield)
Bullet hose res sustained burst of 5mm re from 60-round magazine. SMGs, while not accurate over long distances, can provide withering re at close quarters. Tends to walk upwards as compounding momentum from recoil takes hold. Therefore, careful moderation of aim required to maintain accuracy.
Powerful, loud, pump-action, magazine-fed, ring 8-gauge magnum (3.5) rounds with strong recoil. Devastating at close range and in conned quarters, should be used with caution. Ineective damage ratios at long range mean it should be used appropriately.

COVENANT WEAPONS

Radius: 15-30 feet.
M9 HE-DP Fragmentation Grenade
Basic explosive device has changed little thanks to excellent design and exibility. Well-thrown grenade will kill or stun most opponents. Fuse activates half a second after striking surface or object to avoid accidental detonation. Also allows combatants to bounce grenade into dicult to reach targets.
Recent events have exposed us to yet more Covenant military technology, much of it more powerful than previously encountered, and used in widely diering combat situations. Following is a rough eld guide to currently known Covenant armaments.
Plasma Pistol (dual wield)
Understood better than most Covenant weapons, Plasma Pistol is semi-automatic, directed-energy weapon. Fires rapid bursts of superheated plasma, but holding trigger for extended period can build a powerful overcharged plasma bolt. Becomes temporarily unusable as it discharges excess heat. Both Plasma Pistol and Rie use power core we dont fully understand.

Core Power Output: 100-150 kV : 2-3 dA.
Ammo Capacity: Unlimited.
M41 LAAG (vehicle mounted)
3-barreled, electric-powered, link-less drum-fed, vehicle mounted light anti-aircraft gun. Standard armament on Warthog. Fires 450-550 rounds per minute. Excellent armor penetration capability.

Plasma Rie (dual wield)

Favored by Elites, but used by many Covenant troops, is a directed energy weapon and capable of both automatic and semi-automatic re. Extended bursts of automatic re cause weapon to overheat, temporarily disabling gun and depleting energy core. Once energy core has completely discharged, it is useless.
Rate of Fire: 420-600 rounds per minute.
M68 Gauss Cannon (vehicle mounted)
Asynchronous linear-induction motor produces bipolar magnetic eld to re 25mm projectile at hyper-sonic velocity. Excellent armor penetration capability, but not as eective against multiple infantry.

Needler (dual wield)

Unusual magazine-fed weapon res razorsharp crystalline projectiles, using unexplained homing ability to center on soft organic targets, to pierce esh no matter the angle of impact. Energy shields deect them successfully, and they ricochet from other hard surfaces. Ammunition explodes after it impacts esh, causing further damage.
Ammo Capacity: 30 rounds per magazine.

Covenant Carbine

Rare Covenant projectile-ring rie, a powerful, stocky weapon, and tted with magnifying scope. Fires single rounds with high degree of accuracy and power. In some ways its technology mimics Covenant Fuel Rod Gun, although obviously on a smaller scale, but oers similar penetration to UNSC Battle Rie.

Covenant Energy Sword

Initially thought to be purely ceremonial. Few have been seen in combat, but they are invariably carried by high-ranking Elites. We dont understand how it functions, but it cuts through any armor with ease. Press B for regular melee attack and pull Right trigger for basic undercut attack. Or wait until reticle turns red to lock on an enemy, then pull Right trigger to perform a fatal lunge attack.

Length: 3 feet.

Output Capacity: 18 bursts per charge.

Particle Beam Rie

Very precise, powerful weapon, uses relatively familiar particle beam acceleration method to re devastating beam of energy. Limited battery capacity means weapon can re only 18 bursts before depleting charge. Integrated scope enables two levels of zoom, approximately 5x and 10x. Makes excellent sniper weapon.

Fuel Rod Gun

Launching highly radioactive fuel rod projectiles, weapon is eective against both vehicles and personnel. Weapon is bulky, heavy and carries ve fuel rods per clip. Single 2X level of magnication.
Ammo Capacity: 5 rounds per clip.

Wraith

CREW: 1 Propulsion: Boosted Gravity Propulsion Drive Armament: Plasma Mortar & 2 Auto-ring Plasma Cannons
CREW: 1 Propulsion: Boosted Gravity Propulsion Drive Armament: Twin Plasma Cannons (100-250 kW range)
Slow, bulky, and presenting a huge target, is nonetheless Covenants most destructive mobile armor. Huge bulk is well shielded; covering re it provides from massive Plasma Mortar makes it an inestimably dangerous foe. Piloted by a single occupant, who controls all vehicle and weapons systems, also features limited boost system for enhanced maneuverability.

Spectre

CREW: 1+1 gunner (+ 2 riders) Standard reconnaissance and rapid attack vehicle, deployed by Covenant in all ground combat. Usually, but not always piloted by Elites, is highly maneuverable and res twin bolts of superheated plasma in 100-250kW range. Vehicle is also capable of sustained bursts of speed, although cannot re and appears to be less maneuverable at such speeds. Pull Left trigger to boost speed. Possible that it uses energy from weapons systems to achieve speed increase. Propulsion: Boosted Gravity Propulsion Drive Armament: Plasma Cannon

Banshee

CREW: 1 Propulsion: Boosted Gravity Propulsion Drive Armament: Two Plasma Cannons Secondary Weapon System: Fuel Rod Cannon
Multi-troop armored transport is small and maneuverable and while slow, can move with ease in conned spaces. Main weaknesses are lack of speed and acceleration, and that occupants are fairly exposed. Rear-mounted plasma cannon is dangerous, and vehicle provides multiple ring positions.

Shadow

CREW: 2+8 Propulsion: Boosted Gravity Propulsion Drive Armament: Plasma Cannon
Fast and maneuverable, a formidable aerial assault vehicle. Well shielded against small arms re, Banshee can be crippled or destroyed by heavier weapons. Fuel rod cannon makes it a dangerous bomber as well as fast ghter. Banshee has been observed barrel-rolling and looping in turns and arcs that would be impossible with conventional aerodynamics.
Covenants main mode of moving large numbers of troops around on land. Can hold driver, gunner, and up to 8 occupants, depending on species of Covenant. Seems to be outtted to carry Elites, Brutes, Grunts, and Jackals. Equipped with plasma cannon, but main purpose is to deploy infantry.

SETTINGS

Use the Settings menu to customize a player prole or create a set of rules for a multiplayer game type.

MULTIPLAYER BASICS

Halo 2 multiplayer enables friends to nd one another, to game together, and to move throughout the Halo 2 world as a group. You can play multiplayer games with your friends via Split Screen, System Link, and over the Xbox Live service.

Player Proles

You can have a number of dierent customizable player proles. Change a proles name, modify the controller settings, edit the multiplayer characteristics for a prole, or delete a player prole. To create or edit a player prole, select Settings from the Main Menu, then select Player Proles.

Split Screen

With split-screen play, you can compete with up to three other players, side by side, on a single console. To start a split-screen game, select Split Screen from the Main Menu.
Changing the Control Layout
You can change the control layout for each player prole to better suit your style, including the button scheme and the speed at which you look around. A number of people choose to invert their controls. This means that when you push the Right thumbstick forward, you look down, and when you pull it back, you look up. Experiment to nd which combination works best for you. To change the control layout, select Controls from the Edit Prole screen. You can also modify your layout in a campaign or a multiplayer game by pressing START to bring up the Game Menu.

System Link

With system-link play, you can connect two Xbox consoles with an Xbox systemlink cable, or up to 16 Xbox consoles using an Ethernet hub. For more information on how to do this, see your Xbox console Instruction Manual. To host or join a system-link game, select System Link from the Main Menu.

Xbox Live

With the Xbox Live service, you can play a multiplayer game with people from all over the world, and you can download new Halo 2 maps. To play an Xbox Live multiplayer game, select Xbox Live from the Main Menu (see pg. 24 for details).
Changing the Appearance of Your Multiplayer Character
To customize the appearance of your character in multiplayer games, select Appearance from the Edit Prole menu. You can choose either a Spartan or a Covenant Elite, set primary and secondary colors, and design a custom player emblem.

Game Lobbies

In order to play any type of multiplayer game, you need to go through a game lobby. A game lobby is the gathering place to meet and talk to friends, start a game, or join a game. In the Pregame Lobby you can set up a game; you also can customize your game map and options there, and start a co-op campaign in split-screen play.

Game Variants

Every type of multiplayer game has a set of rules called a variant. You can modify the rules for a game youve created by creating a custom game variant. To customize a game variant, select Options from the Main Menu. To use a variant that youve created, go to Game Setup in the Game Lobby and select Change Rules.
When you play a system link game, go to the Available Games screen to create a new game or join an available game.

Game Types

There are seven multiplayer game types. Each game type has a number of built-in variants that create dierent rules for a game. The Slayer game type has a regular variation called Slayer, Team Slayer, Rockets, and so forth. You also can create a custom variant for a game type to create your own rules for a game. Slayer: Kill the most opponents. Capture the Flag: Score the most points by capturing the other teams ag and bringing it back to your teams base. Assault: Score the most points by carrying, arming, and dropping your teams bomb in the other teams base. King of the Hill: Control the hill for the longest time. Oddball: Find the ball and hold on to it for the longest time. Juggernaut: Only the Juggernaut can winand if you are the Juggernaut, everyone is out to get you. Territories: Earn the most time by nding and controlling territories on the map.

Multiplayer HUD

The multiplayer HUD adds the Multiplayer Scoreboard, which shows the score of the leading player or team above your score. If you are the leader, the score of the second-place team or player is shown underneath. It is located on the lower-right side of the the screen. Hold down the BACK button to see more extensive scores.
Multiplayer Warning Indicators
There are several additional warning indicators that show up below your reticle in multiplayer gamesthese are valuable.

Bomb Dropped

Flag Dropped

Enemy has Bomb

Enemy has Flag

Waypoints

Waypoints are HUD elements used to indicate status on objects in your eld of view. Pay attention to the waypoints in team games and in objective-based games like CTF and Assault.
A map is a self-contained game level designed specically for multiplayer games. Most maps are based on variations of specic campaign levels in Halo 2. Maps come in dierent sizes and have dierent types of buildings, scenery, etc. Large maps work well when you have a lot of people. Some maps have objects with which you can interact. Press X to interact with an object. You can use any game type with any map.

Friends Waypoints

Friend

Firing

Taking Fire

Killed

Talking

Voice Proximity

When playing System Link and Xbox Live games, you can hear other players voices. In Halo 2, the louder the players voices, the closer they are to you on a map. When theyre softer, theyre farther away. Use this to help determine another players proximity. Be careful thoughnearby enemies can hear you talking as well. If you have an Xbox Communicator, you can use voice to direct your team. You can tap the White button to engage your radio and talk to your team, no matter where they are on a map. The radio will remain open until you stop talking.
Has Oddball Has Bomb Has Flag

Objective Waypoints

Capture the Flag

Oddball

Assault

King of the Hill

Territory

Objective

Dropped Flag

Dropped Oddball

Dropped Bomb

XBOX LIVE

Xbox Live!
Take Halo 2 Beyond the Box
Xbox Live is a high-speed or broadband Internet gaming community where you can create a permanent gamer identity, set up a friends list with other players, see when theyre online, and receive invitations to play games. For games with multiplayer mode, invite your friends to play and talk to them in real time while you play. For games with downloadable content, download new items such as levels, missions, weapons, vehicles, and more to your Xbox console.

Custom Game

You can play a game with a specic player, a Party of players, or another Clan. A Custom Game is a private game that you must invite other players to join (see pg. 26 for details). To create a Custom Game, select Create Party, and then select Game Setup to congure the game. When everyone in your Party is ready, select Start Game. Note: A Custom Game has no eect on any of your Xbox Live Halo 2 levels.

Create Party

You can create a Party to game as a group (see pg. 26 for details).

Connecting

Before you can use the Xbox Live service, you need to connect your Xbox console to a high-speed or broadband Internet connection and sign up for the Xbox Live service. To determine if Xbox Live is available in your region and for information about connecting, go to www.xbox.com/connect.

Xbox Communicator

Keep track of your Clan, taunt opponents, or yell at your buddy who doesnt have a clue what stay put means. See the instructions that came with your Xbox Communicator for more information.

Matchmade Game

Quickmatch
Quickmatch picks a random game type, map, and the fastest service for you to start having fun against players with a similar skill level.

Xbox Live Guests

You can have up to three additional guests play a custom game with you on your Xbox console in split-screen mode. Select Xbox Live from the Main Menu, then press A and select a player prole.

OptiMatch

Choose a matchmaking playlist and youll get matched with other players who select that same playlist and have similar skill levels. They will also be players who have the optimal connection speed to provide you with the best possible experience.

Xbox Live Stats

You can compare your stats to other players worldwide. For even more detailed stats info, go to Bungie.net on the web.

Levels

The outcome of a Quickmatch game or an OptiMatch game aects your Halo 2 Xbox Live level for a particular playlist. For each type of playlist, you can earn a dierent level. As you get better at a playlist, your level increases. The more you play, the better chance you have of playing a player in a similar skill level.

GAMING WITH FRIENDS!

There are several ways to nd and play Halo 2 multiplayer games with friends and people you meet online. With an Xbox Live account, you can access your Xbox Live Friends list to play Halo 2 multiplayer games. You can also create a Party, which is a temporary group of friends for a session of Xbox Live multiplayer gaming. And you can create a group of players to form a Clan.
A Party is a temporary group of friends, or other players youve just met, who are playing an OptiMatch or a Custom Game together on the Xbox Live service. Its like being on a virtual couch with people from all over the world as you travel throughout the Halo 2 universe.

Friends List

When you sign into your Xbox Live account, the Friends list is available in Halo 2. You can press Y at any time to bring up the Friends list. Or you can press START in a campaign or multiplayer game, and then press Y to access the Friends list. Use the Friends list to see friends, Clan information, and a list of players youve recently played against.

Create a Party

To create a Party, select Create Party from the Xbox Live screen, then send out your party invitations! Press Y in the Pregame Lobby to bring up your friends list, highlight the friend you want to invite, and then press A to send a Party invite. Friends who join your Party appear in the Pregame Lobby. As the Party Leader, its up to you where you take your Party. To see where you can take your Party, select Game Setup from the Pregame Lobby. When you move between games, you can bring your Party with you. Note: Parties are session-based, so when everyone logs out, the Party comes to an end.

Friends

You can add up to 100 other gamers to your Friends list. Send a text or a voice message to a friend, send a Party invite, or remove a friend from your list. The Status column shows you who is online, who is in your Party, and the current Party Leader.
A Clan is a semi-permanent organization of Halo 2 players on the Xbox Live service. Each Clan can have up to 100 members, but you can be a member of only one Clan at a time. To create, join, or leave a Clan, press Y, and then go to the Clans tab on the Friends List. Clan levels represent the entire clan. Clans are organized into four roles: Overlord, Sta, Member, and Peon. A Clan must have at least one Overlord. Everyone who joins a Clan is made a Member by default. To read more about your Clan, go to Bungie.net (see pg. 29 for details).
The Clans tab lists all the players in your Clan so you can send them messages. You can create, join, or leave a clan from the Clans tab.

Players List

The Players List displays all current players and up to 100 of the most recent players youve competed with or against. You can view player proles, send text or voice messages, provide feedback about what kind of players they are, or invite them to be a friend.

MULTIPLAYER STRATEGY

Individual Strategy
When you play as an individual player, its important to manage your two dierent weapons correctly. Try to carry a long-range (such as a Battle Rie) and shortrange weapon (such as a Shotgun). Make sure youve always got a few grenades. Running backwards and throwing grenades is a great defensive move when another player surprises you in a map. Use dual wielding in tight spaces at short range, and vehicles when outdoors. Try boarding another players vehicle. A successful boarding ejects another player from their seat. Move slowly by crouchwalking if you want to stay invisible on other players motion trackers. Also, dont charge through the front door. Use a little stealth and planning.

BUNGIE.NET

Bungie.net is the gaming center of the universe for the Halo 2 community, Halo 2 Clans, and the individual Halo 2 player. Bungie.net is the place to get the latest Halo 2 information, to interact with the Bungie team, and to buy cool Halo 2 merchandise at the Bungie Store. Bungie.net is your resource to learn how to be a better Halo 2 player. If you create a Bungie.net account, you can access the Bungie forums and see which of your Xbox Live friends are online. You can review the stats from the most recent Halo 2 games youve played on the Xbox Live service, compare stats with friends, and check your current level on the various playlists.

Microsoft

User Testing John Hopson Randy Pagulayan Xbox Marketing & PR Chuck Blevens Jen Martin Cameron Payne Genevieve Waldman Orlena Yeung Localization Peter Fitzpatrick Kyoung Ho Han Mitsuru Kitamura Robert Shih-Wei Lin Victoria Olson Jason Shirley Xbox Platform & Xbox Live Mei-Mei Bong Michal Bortnik Sam Charchian Tony Chen Michael Courage Brent E. Curtis Tristan Jackson Daniel McGillicuddy Boyd Multerer Anthony Smith Jason Strayer Van Van Additional Test Adrian Brown Jeremy Fischer Domenic Koeplin Matt Shimabuku Justin Jones Manual Design Jeannie Voirin User Experience Caitlin Sullivan Matt Whiting Special thanks: Robbie Bach, Matt Case, Lev Chapelski, FASA Studios, Ed Fries, Nick Gray, Shane Kim, Peter Moore, Stuart Moulder, Alex Seropian.
Animators, Game & Cinematic Bill OBrien Mike Budd John Butkus Nathan Walpole Additional Animation Jeremy Fones Stacey Moore Lead Environment Artists David Dunn Christopher Barrett Single-Player Environment Artists Frank Capezzuto Vic DeLeon Tom Doyle Justin Hayward Paul Russell Michael Wu Mike Zak Additional Single-Player Environment Art Chris Lee Lead Multiplayer Environment Artist Chris Carney
Bungie Princesses Alta Hartmann Amanda Anderson
S & T Onsite Art Source Coresta Siemens

TECHNICAL SUPPORT

Limited Warranty For Your Copy of Xbox Game Software (Game) Acquired in the United States or Canada Warranty
Microsoft Corporation (Microsoft) warrants to you, the original purchaser of the Game, that this Game will perform substantially as described in the accompanying manual for a period of 90 days from the date of rst purchase. If you discover a problem with the Game covered by this warranty within the 90-day period, your retailer will repair or replace the Game at its option, free of charge, according to the process identied below. This limited warranty: (a) does not apply if the Game is used in a business or for a commercial purpose; and (b) is void if any diculties with the Game are related to accident, abuse, virus or misapplication. Technical support is available 7 days a week including holidays. In the U.S. or Canada, call 1-800-4MY-XBOX. TTY users: 1-866-740-XBOX. In Mexico, call 001-866-745-83-12. TTY users: 001-866-251-26-21. In Colombia, call 01-800-912-1830. For more information, visit us on the Web at www.xbox.com