A Dynamic Area of Interest Management and ... - Semantic Scholar

A Dynamic Area of Interest Management and Collaboration Model for P2P MMOGs Dewan Tanvir Ahmed and Shervin Shirmohammadi Distributed and Collaborative Virtual Environments Research Laboratory School of Information Technology and Engineering University of Ottawa, Ottawa, Canada {dahmed, shervin}@discover.uottawa.ca

Abstract In this paper, we present a dynamic area of interest management for Massively Multiplayer Online Games (MMOG). Instead of mapping the virtual space to the area of interest (AOI), we scheme AOIs to the virtual space. This zoneless MMOG is the consequence of dynamic AOIs that redeems the necessity of inter-AOI communication. In addition, the AOI maintenance cost is reduced significantly by assigning the maintenance responsibility to a subset of players for each AOI. Due to the integration of peer-to-peer communication model to the system, the scalability has improved. To satisfy the timing constraints, the projection of the underling network topology to the overlay network is more fruitful than building the overlay on the fly unintelligently. In response to this fact, the proposed communication model adapts the gift wrapping algorithm which is usually used for minimax problem. The model is evaluated and justified through proper simulation.

1. Introduction The networked virtual environments (NVE) with many simultaneous users distributed over a wide-area network introduce an interesting research challenge. There are many applications of NVE such as distributed training, education, virtual meeting, multiplayer game, etc. Massively Multiplayer Online Game (MMOG) is a kind of NVE where hundreds of thousands of players interact simultaneously in a virtual world. Once considered a niche business, the prospects of multiplayer online gaming are growing rapidly. It has enviable business hope because of recurring revenues, competitive nature and time engaged and many others. Interestingly, the most valuable Internet Company in China is MMOG, clicking in at US$2.0B in market cap. Currently,

Second Life and World of Warcraft are the two most popular MMOGs having around 10 and 11 million subscribers, respectively. The commercial MMOGs use client-server architecture which is expensive to deploy and maintain. For example, Second Life has approximately 5000 servers to support the virtual space. The development and the maintenance of such server-clusters are really expensive that cost millions of dollars. Recently, the hybrid P2P architecture is considered as an alternative solution for both vendors and end users [23][12][13]. We strongly believe that development of online games over the peer-to-peer model is advantageous in terms of deployment cost and performance, in some sense, through reduced latencies. In order to ensure a consistent game space for the players, in MMOG, each player must maintain a clone of the relevant game states on his computer. When a player performs an action or generates an event affecting the virtual space, the game state of all other players influenced by that action or event must be updated. The amount of information required to exchange between players roughly depends on the population of in the interested area. However, the capacity is bounded by at least two technical limitations: network bandwidth and processing power [19][20]. There are two extremes to model AOI among the users for distributed simulations. The first one is the static geographical partitioning implemented at the initialization phase of a game or a simulation. This is useful as it defines the structure of a virtual world. For example, a virtual world consists of multiple cities where each city represents a geographical partitioning: it is the area where most of the interactions take place, and in most cases, the interacting parties are not interested in what is happening in other cities. Second Life has adopted such approach [22]. To make up for delay introduced by migration, the virtual world is designed such that the boundaries between AOIs are uninteresting in nature: cities are separated by forests or wilderness, where

parties do not tend to stay long. Thus, a migrating party will not experience others’ interactions at the boundaries. Although this works for Second Life, it is not an acceptable solution for all virtual simulations. The second extreme for modeling interest is behavioral. In a military simulation, two different units such as a tank and a fighter jet have different behaviors in terms of how fast they move, how far they can see, and the size of the area they can interact with (a jet launching a missile has a larger area of influence than a jeep patrol). Lu et al. argue that, due to the ease with which mapping processing resources or servers can be applied in geographic regionalization, little effort has been invested in mapping the behavioral approach [16]. The mapping of geographic regionalization to server allocation in a bid to prevent inter-server communications is not the appropriate path to take. This is because geographic regionalization does not afford the greatest potential for player interaction, and the processing overhead of allocating resource. Even though behavioral modeling is the ultimate goal for managing the interest of the parties, geographic regionalization is not without its merits and can be coupled with behavior-based communications. In this paper, we present a new concept to scheme area of interest. Instead of mapping the virtual space to the area of interest (AOI), we scheme AOIs to a virtual space. This pure dynamic AOI does not need to maintain a regular shape and redeems the necessity of inter-AOI communication. The maintenance cost is reduced by assigning the shape tracking responsibility to a subset of players in each interested area. On the other hand, due to the integration of peer-to-peer communication model to the system, the scalability has improved. To satisfy timing constraints, the projection of the underling network topology to the overlay network is more demanding rather than building an overlay on the fly. The proposed communication model incorporates the gift wrapping algorithm in this response. The road map of the paper is as follows. We outline the related work in Section 2. The theory of AOI and the relation of AOI to a game space are covered in Section 3 and Section 4, respectively. The proposed AOI concept is portrayed in Section 5. In Section 6, we present P2P communication model for an AOI. The requirement of inter-AOI is discussed in Section 7. The simulation results and corresponding analysis are mentioned in Section 8. Finally, we conclude the paper in Section 9.

2. Related Work Knutsson et al. describe the P2P support for Massively Multiplayer Games by integrating Pastry and Scribe features [14] where the game space is divided into fixed-size regions. Each region is managed by a coordinator; the root of a multicast tree. The players inside the same region sub-

scribe to the root node to receive updates from other players, so neighbors are discovered via the coordinator. The coordinators maintain links with each other, facilitating player transitions to other regions. In Knutsson et al., due to discrete AOI, the players can not see across regions. So, if a player decides to listen to more regions, as suggested in the paper, the unnecessary messages beyond the AOI will be received. A considerable performance penalty can be introduced as the overlay does not consider the area of interest; the messages may need to be relayed by other nodes (1 to 2 hops for most cases, but in some cases it goes beyond 50 ”virtual hop”, so more delays happen at the physical level). In short, the architecture does not fully utilize the power of direct connections. The delaunay network is a good solution to the networked virtual environments which organizes the players according to their position in the virtual world [21]. Indeed, it is theoretically a good interest management approach. But, the maintenance cost of the delaunay network increases against the players’ density and movement pattern. Thus, each player has a considerable volume of traffic. To address this issue, they propose a dynamic clustering algorithm where each peer in the network monitors maintenance cost and initiates a cluster formation phase as soon as the volume of traffic exceeds a certain limit. The members of the cluster then expand their coordinates to increase their reciprocal distances. In this way, by reducing the concentration of the players, the opted approach tries to keep the maintenance cost within the affordable limit. But, the centralized architecture and the high maintenance cost are its two big limitations. Hu et al. propose, in some sense, a fully-distributed peer-to-peer architecture to solve the scalability problem of networked virtual environment [11]. This method considers player’s locality of interest intrinsic to the system. The mathematical construct, voronoi diagram, is incorporated in the design where the game space is dynamically partitioned based on players’ position: called the voronoi graph partitioning algorithm. Thus, the players in the same region directly exchange game events. The update interval of the game states is small since they allow the player nodes with the same sub-state to exchange game events among those nodes directly or through a management node. However, as the number of player nodes in a sub-state increases, the volume of messages from each player or its management node also increases massively. So, the communication and the message exchange rate can not be regulated entirely. Shun-Yun et al, present a game state management scheme named Voronoi State Management (VSM)[10]. It mainly works in a centralized manner, i.e. client-server architecture, in each zone. But to balance arbitrators’ load, it adjusts region boundaries and promotes capable clients as new arbitrators. In that sense, they called it a P2P based

reliability, while others use sender-initiated approaches to transmit key updates with guaranteed reliability [18]. The IEEE DIS standard [1] has also been successfully used in a controlled environment with vast resources, mostly for military simulations. These approaches are all based on IP multicast, although it achieves good results in Intranet environment, it is not readily deployable on the Internet. So the projection of the underling network topology to the overlay network is more demanding rather than building an overlay on the fly unintelligently. In the proposed communication model, such points are taken care of. Figure 1. The aura-nimbus model

game state management. The fault tolerance of the system has improved by replicating game states in nearby regions at the expensive extra load. It has an interesting localized replication policy. In VSM, the object states are fully replicated at all surrounding regions. It greatly increases load on each arbitrator. That is a big problem to make it scalable. For example, if it has the hexagonal zone shape, each arbitrator must keep and maintain object states of seven regions (six surrounding plus the center zone). From the architectural point of view, our presented approach forms interested area dynamically where its shape changes regularly in a controlled manner. Moreover, in the proposed method, the maintenance cost is reduced to a significant extent by assigning the maintenance responsibility to a subset of players in each interested area. Marios et al. present an approach to support Massively Multiplayer Online Role-Playing Games (MMORPG) using a centralized architecture [2]. To reduce bandwidth requirements, it considers player’s locality of interest for both game servers and clients. But, considering the design presented in their article, it is simply a multiple client/server architecture where performance improvement is flat. There is no guarantee for end-to-end delay as well. In this paper, we use minimax algorithm to relax timing constraints. Network latency and processing delay are critical in this regard. Moreover, the latency tolerance usually varies from application to application and typically lies from 100ms to 1000ms for networked games [9][7]. It affects the performance of the application; especially the gaming experience. When a player participates in a simulation, its interactions with other players must be informed to the all involved players in the vicinity. Because of the networking limitations and the traffic conditions, some of these ’updates’ can be lost or delayed. Much research have been conducted to overcome the networking limitations for better distributed simulators. Some of these studies provide receiver-initiated and selectively-reliable transport protocols [17] that can be used to deliver important messages with a high degree of

3. Area of Interest Management The game space of an MMOG contains plenty of information. But in reality, a single player needs only a small subset of that information. The interest management is the way of determining useful information relevant to each player. Thus, providing relevant information to each player is an effective way for consistency management. The interest management for an MMOG can be abstracted using a publish-subscribe model. The publishers (e.g. objects) produce events; the subscribers (e.g. objects) receive events. In this model, the interest management consists of determining when an avatar subscribes to or unsubscribes from a publisher’s (avatar’s) updates. The interest management, however, can have multiple domains. The most familiar domain is the visibility, but there can be other domains like audible range, radar, sensor arrays, etc. Each interest domain has different properties for message transmission and reception, so different publisher-subscriber models are required. The space-based area of interest management is typically based on proximity, and can be realized in terms of an auranimbus information model as shown in Figure 1. The aura is the area that bounds the existence of an avatar in the space, while the nimbus, i.e. area of interest, is the space in which an object can perceive other objects. At its simplest, both the aura and the nimbus are usually represented by fixedsize circles around the avatar. This model is more appropriate for client-server architectures. The main drawback of a pure aura-nimbus model is its scalability, because of the computing cost associated with the intersection determination between the nimbus and the auras of the avatars. The region-based interest management partitions the game world into several regions. The interest management determines the regions that intersect the expression of interest of the avatar, i.e. subscriber. Thus, the area of interest is the union of the intersected regions with respect to expression of interest. So, it is an approximation of the true expression-of-interest and generally is much cheaper to compute than a pure aura-nimbus model.

load of the system. Each AOI is monitored by a responsible server that provides different maintenance services like tracking AOI scope (explained later). The overlapping conditions of AOIs are checked intelligently at regular intervals to minimize communication overhead and at the beginning of a game session, each player discovers his/her interested area through a server.

5. Zoneless Area of Interest Management

Figure 2. The symmetric and asymmetric relation of interest

4. Game Space and Area of Interest The proper management of AOIs of an MMOG is a challenging task. It is true that the mapping of an area of interest to a fixed size zone, i.e. unification of an AOI to a zone, is straightforward. It does not require forming interested area in real-time. This type of approach supports logical zones which are permanently defined at the beginning of the game. Thus, players moving over a zone are considered to have a common interest confined to that logical space. But, due to the nature of a game, most of the times an area of interest overlaps multiple zones and breaks the significance of the zone formation. In addition, it requires regular inter-zone communication for synchronization. Thus, it makes system more vulnerable to heavy load in terms of players. In this paper, we are presenting a completely different approach for interest management where an area of interest evolves overtime and creates a logical zone as needed. Initially, each player has its own area of interest which is defined by its visibility scope. Say, pi is a player whose area of interest is defined by aoi(pi ). Two players, say pi and pj , have a common are of interest if they fall in each others visibility range. This is a symmetric relation of interest, i.e. if the player x is interested to the player y, then the player y is also interested to the player x (as shown in Figure 2). This symmetric relation is not always true. Sometimes, we need to deal with the asymmetric relations like for radar, sensor arrays, etc. Thus, when more than one player have a common area of interest then the resultant interested area can be defined as AOI(i) = {pi1 , pi2 , ..., pik }, where k is the number of players in that interested area. Thus, the game space consists of many AOIs and can be represented by Gspace = ∪AOIj . In this paper, we have chosen a hybrid MMOG architecture. The system is administered and coordinated by a set of servers which uses the participating players’ resources whenever possible to relay game states to others to reduce

As the players’ position and scope, i.e. AOIs, change constantly, the logical definition of space is very important to reduce AOI maintenance cost. The key motivation in this regard is to find out a subset of players in each logical AOI and assign AOI scope tracking responsibility to them. These responsible players can properly reshape AOIs when necessary with the least cost without involving all players in the vicinity. This section describes a way to AOI tracking and presents a mechanism to effectively expose the logical area of interest. Say, an AOI has k players represented by AOI(i) = {pi1 , pi2 , ..., pik }. We form a convex hull to present such an AOI. A convex hull for a set of points is the minimal convex set covering that set. The simplest algorithm in the plane was proposed by R.A. Jarvis in 1973. It is also called gift wrapping algorithm with a time complexity of O(nh), where n is the number of points in the set (the players, in our case), and h is the number of points in the hull (responsible for tracking an AOI, in our case). The algorithm starts with a point p0 in the convex hull. One way to discover such point is to find out the left most point in the set. It then selects the next point in the convex hull, say point pi+1 , such that all points are to the right of the line pi pi+1 . This point may be found by comparing the angles of all points with respect to the point p0 taken for the center of coordinates. It then starts processing with respect to the point pi+1 . These steps continue until it reaches p0 again. In this way, we define the scope of interest. Let Sh is the subset of Sn , i.e. Sh ⊆ Sn , that forms the convex hull. The size of the convex hull changes due to the players’ random movement in the game space. All players in an AOI know who are at the convex hull periphery, i.e. Sh . Thus, each player independently can determine its position with respect to the convex hull. So, if a player is inside the convex hull, it has no role to redefine the convex hull at that instant, otherwise it is a candidate to redefine it. Considering the nature of the game at hand, we do not redefine the convex hull at that instant as we do not know whether the player is returning back soon. That is why we use the term candidate. Two attributes are incorporated to make the decision while really redefining the convex hull for each candidate. These attributes are time-span and safety-edge space. A safety-edge space is defined for each edge of a

Figure 3. The safety margin of a convex hull convex hull. It gives a safety margin describing the player has moved far enough and it is unlikely that it will return back soon. The time-span acts like a temporal reference and used to avoid premature decision. Thus, these two conditions ensure mature decision in terms of time and space. The convex hull is changed when a candidate player satisfies both conditions. Thus, a candidate player informs all members with the modified set of convex hull. The motivation of forming a convex hull is to define a confined game space for a set of players having a common interest. As the players move frequently at random directions, the servers require continuous tracking of players’ positions to precisely detect each others interest in the game. When a convex hull is formed for a set of players, the server can roughly discover their area of interest by monitoring and tracking those players who define the convex hull. The players who are inside the convex hull do not require regular interactions with the server provided that, for game-play, a peer-to-peer architecture manages the intra-zone communication. So, the presented concept should reduce communication load of the server to a significant extent because of not only for P2P nature but also for the confined area of interest.

6. P2P based Communication for each AOI

spectives (i.e. First-person or Third-person), the game genres (i.e. racing or role playing game), and the sensitivity of actions as well. In this section, we present an approach for P2P intra-zone communication using minimax theorem: minimizing the maximum latency. To minimize overlay latency, we need a good approximation about the physical position of each player. In the process, each player determines its relative distances with respect to the geographical landmarks. The distance to the landmark is used to estimate its position. It is intuitive that three landmarks are good enough for the approximation. As the end-to-end delay itself is not invariant, the intersections may not be explicitly available. Thus, the decision about player’s position in the global co-ordinate system is approximated. The responsible server uses these facts to configure proper routing table to satisfy the timing constraints. Say, there are n players in an AOI. So, given n distinct points Pi = (ai , bi ) in the plane, the problem is to find a point X = (x, y) that minimizes the maximum Euclidean distance from X to the given points. It should be noted that, here, the logical position of the players in the game space has no role for the effective routing table. It is the physical position that needs to be considered as the packet follows the physical layout of the real network. Let f (X) = max1≤i≤n latency(X, Pi ). The problem is to minimize f (X), i.e., min max1≤i≤n latency(X, Pi ). A standard transformation to write the problem is as follows: min z subject to latency(X, Pi ) ≤ z, f or 1 ≤ i ≤ n This version of the problem has the geometric interpretation of finding a circle with the center X and the minimum radius z so that all the given points Pi are in the circle, called the minimum covering circle problem. We have followed Elzinga and Hearn’s the geometric algorithm for solving the center of a set of points with Euclidean distances [8]. The modified algorithm for our case is as follows. 1. Select any two points Pi and Pj in the AOI space 2. Construct the circle having the diameter L(Pi Pj )

As mentioned earlier, the communication within each AOI follows a peer-to-peer scheme. But, the construction of an appropriate routing table for each AOI member is not easy to follow. It is has been discussed that MMOGs have a set of requirements like consistency and responsiveness. Most importantly, a networked game requires observing the effect of an action in time. But, the network latency and the processing delay make the synchronization difficult. Moreover, the latency tolerance usually varies from game to game and stays in between 100ms to 1000ms for networked games [7]. Obviously, the value contrasts on the game per-

• The center of this circle is optimal if all points are inside the circle • Otherwise, we choose a point Pk outside the circle 3. Steps to follow: • If the triangle determined by Pi , Pj and Pk has a right triangle or an obtuse triangle, rename the two points opposite to the right angle or to the obtuse angle as Pi and Pj , and move to step 2.

each wedge in parallel. As mentioned before, the message within each sector is disseminated to the sector members directly by the sector leader.

7. Do we need Inter AOI Communication?

Figure 4. The coronas, wedges and sectors for an AOI

• Otherwise, we construct the circle passing through the three points • If the circle contains all the points, the algorithm is complete, else follow the step 4. 4. Choose some point Pl not in the circle, and let Q be the point among {Pi , Pj , Pk } that is greatest distance from Pl . Extend the diameter through the point Q to a line that divides the plane into two half planes. Let the point R be the point among {Pi , Pj , Pk } that is in the half plane opposite Pl . With the points Q, R, and Pl , jump to step 3. The center of the AOI with respect to the physical positions is a good approximation to keep the end-to-end delay within the limit. So, the root of the overlay can be the player who is the closest to that center. Let us consider the physical space as a disk D of radius R. Then, we decompose it into disjoint concentric sets named coronas obtained as follows. Consider k concentric circles of radii 0 < r1 < r2 < ... < rk = R centered at X. Now, for every i, (1 ≤ i ≤ k), the corona Ci is the sub-area of D delimited by the circles of radii ri−1 and ri . The width of each corona is R/k. So, here, all coronas have same width. For example, in Figure 4, k = 4 and the area is partitioned into four coronas C1 , C2 , C3 and C4 . For effective state sharing, we decompose each corona into several sectors. Consider an arbitrary wedge W subtended by an angle θ, as shown in the Figure 4. The W is partitioned into k sectors S1 , S2 , ..., Sk by its intersection with k concentric circles. For each sector we select a leader who communicates with other sectors in the same wedge. Within each sector, each player directly forwards messages to the fellow sector members. So, each event is first directly transferred to a member of the center corona. Thus, the message follows via the leaders from sector to sector within

Despite the integration of zoneless area of interest in the design and henceforth the AOI formation, the overlapped AOIs are not uncommon. The overlapped AOI is the result of multiple interests of the players, and sometimes due to the non-uniform scope of interests among the players. So, the question is, do we need inter-AOI communication to endure such circumstances? To realize whether there is a need for an inter-AOI communication; we need to find out those players that belong to multiple AOIs. Say, AOI(i) = {pi1 , pi2 , ..., pik } and AOI(j) = {pj1 , pj2 , ..., pjl } are two overlapped AOIs. So, the simple intersection of these two sets is good enough to uncover the overlapped members. From the design it is evident that each member of the overlapped AOIs must stay at a corona in each AOIs and being a member of the respective overlays. This ensures each player receives all messages from the all AOIs where it belongs and redeems the necessity of explicit inter-AOI communication. This is the real benefit of zoneless AOI.

8. Simulation and Analysis Extensive simulations were carried out to verify the proposed approaches and to evaluate the performance. We have considered simulations for two types of MMOGs: MMORPG (Massively Multiplayer Online Role Playing Game) and FPS (First Person Shooter). We have chosen to work with real-world measurements of traffic for these two types of games due to their varying characteristics. In this section, we go over these simulations. MMORPG and FPS games are similar to some extent as they require low bandwidth and generate small packets periodically. The bandwidth requirement of an MMORPG is even lower because of its strategic nature compared to an FPS. The game traffic of an MMORPG has strong periodicity but sometimes follows the temporal locality. The traffic pattern of a client is a complex function of different factors but in most cases it follows exponential distribution like ShenZhou online [6]. But, the packet interarrival time is spread between 20ms through 600ms and varies with respect to the game genres. In this simulation, a player’s traffic is modeled using the exponential distribution function. The packet interarrival time for the simulated MMORPG, e.g. ShenZhou, was set to 500ms, as measured in other works [6]. It should be noted that, the client packet interarrival time is independent on the number of players

Figure 5. The effectiveness of the proposed approach in terms of maintenance overhead

Figure 6. The maximum distance between players in an AOI

[3]. In Quake, the most computationally expensive part of a cycle is the rendering. It causes slower hosts to have significantly higher and more variable interarrival time, while the fastest hosts generate most packets at 14ms intervals [15][4]. Moreover, the client packet interarrival time also depends on the game map where some maps have very regular packet transmission intervals and for others, packets are sent more randomly [15]. Thus, the client traffic can be modeled as one extreme [4] or two extreme distributions [5]. We considered one extreme distribution. It should be noted that, for Voronoi case, every player keeps track of its voronoi space which ultimately introduces life long maintenance cost. As explained earlier, the maintenance cost of a pure AOI is reduced by assigning zone tracking responsibility only to the members on top convex hull. So, we have estimated how many players are actually involved keeping the track of the interaction space. The performance indeed depends on the players’ position and the size of AOI in terms of the number of players. We have conducted a simulation to estimate the effectiveness of the proposed concept. From Figure 5, it seems that the performance improves as the size of AOI increases in terms of the players. So for a moderate sized AOI, i.e. the size of AOI is 30, only 34% players are involved for AOI maintenance. We also have evaluated the adapted minimax algorithm for P2P MMOGs. For different AOI sizes, we have compared the maximum distance required to forward game states. To be fair, we didn’t add any constraint to the general approach where we assumed that the maximum delay is the pure maximum end-to-end delay between any two players. But, in the proposed method, a message first transferred to the leader of the center corona and then it is relayed to other members. Thus, the general strategy (similar to clientserver architecture) slightly outperforms the adapted min-

imax approach in terms of maximum distance, as shown in the Figure 6, which is quite obvious based on the settings designed for the simulation. The cogent part of the adapted minimax algorithm is the average distance between each player over the P2P model. For the same settings, we measured the average distance of our adapted minimax approach. The end-to-end delay between all pairs of players was calculated and the average was taken. From Figure 7, it is clear that the adapted approach outperforms the generic approach in terms of average distance. This is due to the reflecting of the underlying physical network onto the overlay network to forward game states.

9. Conclusion In this paper, we have presented a new concept to manage area of interest. Instead of mapping the virtual space to the area of interest (AOI), we scheme an AOI to a virtual collaboration space. This pure AOI does not need to maintain a regular shape and redeems the necessity of inter-AOI communication. The system maintenance cost is reduced by assigning the shape tracking responsibility to a subset of players for each AOI. On the other hand, to satisfy the timing constraints, the projection of the underling network topology to the overlay network is more demanding rather than building an overlay on the fly unintelligently. The proposed communication model incorporates the gift wrapping algorithm in this response. So, the presented concept reduces communication load of the server to a significant extent because of not only for its P2P nature but also for the confined area of interest. From simulation results, it is evident that the proposed approach can be a good solution for P2P MMOGs.

Figure 7. The average distance between players in an AOI

References [1] IEEE standard for distributed interactive simulation - application protocols. IEEE Std 1278.1a-1998, pages –, 19 Aug 1998. [2] M. Assiotis and V. Tzanov. A distributed architecture for mmorpg. In NetGames ’06: Proceedings of 5th ACM SIGCOMM workshop on Network and system support for games, page 4, New York, NY, USA, 2006. ACM. [3] R. Bangun, E. Dutkiewicz, and G. Anido. An analysis of multi-player network games traffic. Multimedia Signal Processing, 1999 IEEE 3rd Workshop on, pages 3–8, 1999. [4] M. Borella. Source models of network game traffic, 2000. [5] C. Chambers, W. chang Feng, S. Sahu, and D. Saha. Measurement-based characterization of a collection of online games. In IMC ’05: Proceedings of the 5th ACM SIGCOMM conference on Internet measurement, pages 1–14, New York, NY, USA, 2005. ACM. [6] K.-T. Chen, P. Huang, C.-Y. Huang, and C.-L. Lei. Game traffic analysis: an mmorpg perspective. In NOSSDAV ’05: Proceedings of the international workshop on Network and operating systems support for digital audio and video, pages 19–24, New York, NY, USA, 2005. ACM. [7] M. Claypool and K. Claypool. Latency and player actions in online games. Commun. ACM, 49(11):40–45, 2006. [8] D. J. Elzinga and D. W. Hearn. Geometrical solutions for some minimax location problems. Transportation Science, 6:379–394, 1972. [9] T.-Y. Hsiao and S.-M. Yuan. Practical middleware for massively multiplayer online games. IEEE Internet Computing, 9(5):47–54, 2005. [10] S.-Y. Hu, S.-C. Chang, and J.-R. Jiang. Voronoi state management for peer-to-peer massively multiplayer online games. In Consumer Communications and Networking Conference, 2008. CCNC 2008. 5th IEEE, pages 1134–1138, 2008.

[11] S.-Y. Hu, J.-F. Chen, and T.-H. Chen. Von: a scalable peerto-peer network for virtual environments, 2006. [12] T. Iimura, H. Hazeyama, and Y. Kadobayashi. Zoned federation of game servers: a peer-to-peer approach to scalable multi-player online games. In NetGames ’04: Proceedings of 3rd ACM SIGCOMM workshop on Network and system support for games, pages 116–120, New York, NY, USA, 2004. ACM. [13] I. Kazem, D. T. Ahmed, and S. Shirmohammadi. A visibility-driven approach to managing interest in collaborative virtual environments with dynamic load balancing. In Proceedings of IEEE International Symposium on Distributed Simulation and Real Time Applications, 2007. [14] B. Knutsson, H. Lu, W. Xu, and B. Hopkins. Peer-to-peer support for massively multiplayer games, 2004. [15] T. Lang, P. Branch, and G. Armitage. A synthetic traffic model for quake3. In ACE ’04: Proceedings of the 2004 ACM SIGCHI International Conference on Advances in computer entertainment technology, pages 233–238, New York, NY, USA, 2004. ACM. [16] F. Lu, S. Parkin, and G. Morgan. Load balancing for massively multiplayer online games. In NetGames ’06: Proceedings of 5th ACM SIGCOMM workshop on Network and system support for games, page 1, New York, NY, USA, 2006. ACM Press. [17] J. M. Pullen. Reliable multicast network transport for distributed virtual simulation. In DIS-RT ’99: Proceedings of the 3rd International Workshop on Distributed Interactive Simulation and Real-Time Applications, page 59, Washington, DC, USA, 1999. IEEE Computer Society. [18] S. Shirmohammadi and N. D. Georganas. An end-to-end communication architecture for collaborative virtual environments. Comput. Netw., 35(2-3):351–367, 2001. [19] S. Singhal and M. Zyda. Networked virtual environments: design and implementation. ACM Press/Addison-Wesley Publishing Co., New York, NY, USA, 1999. [20] J. Smed, T. Kaukoranta, and H. Hakonen. Aspects of networking in multiplayer computer games. In L. W. Sing, W. H. Man, and W. Wai, editors, Proceedings of International Conference on Application and Development of Computer Games in the 21st Century, pages 74–81, Hong Kong SAR, China, 2001. [21] M. Varvello, E. Biersack, and C. Diot. Dynamic clustering in delaunay-based p2p networked virtual environments. In NetGames ’07: Proceedings of the 6th ACM SIGCOMM workshop on Network and system support for games, pages 105–110, New York, NY, USA, 2007. ACM. [22] B. D. Vleeschauwer, B. V. D. Bossche, T. Verdickt, F. D. Turck, B. Dhoedt, and P. Demeester. Dynamic microcell assignment for massively multiplayer online gaming. In NetGames ’05: Proceedings of 4th ACM SIGCOMM workshop on Network and system support for games, pages 1–7, New York, NY, USA, 2005. ACM. [23] A. P. Yu and S. T. Vuong. MOPAR: a mobile peer-to-peer overlay rrchitecture for interest management of massively multiplayer online games. In NOSSDAV ’05: Proceedings of the international workshop on Network and operating systems support for digital audio and video, pages 99–104, New York, NY, USA, 2005. ACM Press.