HyVVE
A Voronoi based Hybrid Architecture for Massively Multiplayer On-line Games
Laura Ricci, Luca Genovali and Barbara Guidi
Department of Computer Science, University of Pisa, Pisa, Italy
Keywords:
Distributed Virtual Environments, Voronoi, Delaunay, Load Balancing.
Abstract:
Massively Multiplayer Online Games (MMOGs) have gained a lot of popularity in recent years. The problem
of defining a proper architecture supporting MMOGs is still a research challenge because the classical client
server architecture mainly adopted in commercial applications presents several drawbacks like unsatisfactory
scalability and limited fault tolerance. This paper presents HyVVE (Hybrid Voronoi Virtual Environments),
an hybrid architecture exploiting both server and P2P nodes. HyVVE exploits a Voronoi tessellation of the
virtual world to distribute the load for the management of the virtual entities among the server and the peers.
The paper presents a set of experimental results proving the effectiveness of our approach.
1 INTRODUCTION
Distributed Virtual Environments (DVE) (Ohnishi
et al., 2005a; Bharambe et al., 2006; Yu and Vuong,
2005; Knutsson et al., 2004; Carlini et al., 2013) en-
able geographically distant users to communicate, in-
teract and collaborate within a virtual environment. In
particular, online gaming entertainment has acquired
lots of popularity in the last years from both industry
and research communities. The market size of online
gaming has received a 5 billion evaluation in 2010,
while the number of total users have reached around
20 million worldwide.
Currently, most commercial Massively Multi-
player Online Games (MMOGs) rely on a client
server architecture which supports a straightfor-
ward management of the main functionalities of the
MMOG, such as user identification, management of
the state of the virtual world, synchronization be-
tween players, and billing. However the most im-
portant drawback of these architectures is their lim-
ited scalability which prevent a satisfactory playabil-
ity when huge amounts of concurrent users are play-
ing simultaneously.
Recently several solutions based on P2P based
MMOG have been presented. The main advantage
of these solutions is that they are inherently scalable
because when the amount of users grows, more re-
sources are added to the infrastructure. However, a
pure P2P-based approach is barely feasible. As a mat-
ter of fact, the lack of a centralized authority makes it
complex to enforce security, consistency of concur-
rent updates to the state of the virtual world and per-
sistence of its state when a few players are present in
the virtual world.
An interesting alternative exploiting the advan-
tages of both the client/server and of the P2P archi-
tecture is to define an hybrid solution which properly
distributes the functionalities of the MMOG among
the server and the peers. For instance, the state of
the virtual world may be partitioned and distributed to
the server and to the peers by exploiting the locality
property characterizing most MMOGs. As a matter of
fact, an avatar generally interacts with other entities
(avatars and passive entities) located in its proximity,
i.e. in its Area of Interest, AOI. A simple solution is to
define a partition of the state of the MMOG where a
peer manages the entities in its AOI, while the server
manages entities located in areas not covered by the
AOI of any peer. This solution is simple, however it
does not resolve the problem of the ownership of the
entities which are located in the intersection of the
AOI of a set of peers.
In this paper we propose HyVVE, Hybrid Voronoi-
based Virtual Environment, an hybrid architecture
based on a Voronoi Tessellation (Aurenhammer,
1991) of the virtual world which exploits the local-
ity of MMOGs.
Given n sites in an euclidean space, a Voronoi tes-
sellation partitions the virtual world into n areas such
that the area corresponding to a site n includes all the
points which are closer to n with respect to any other
site.
15
Ricci L., Genovali L. and Guidi B..
HyVVE - A Voronoi based Hybrid Architecture for Massively Multiplayer On-line Games.
DOI: 10.5220/0004531900150023
In Proceedings of the 4th International Conference on Data Communication Networking, 10th International Conference on e-Business and 4th
International Conference on Optical Communication Systems (DCNET-2013), pages 15-23
ISBN: 978-989-8565-72-3
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
In a Voronoi based MMOG, the position of each
peer in the virtual world is exploited to define the
Voronoi tessellation of the world which is exploited
to assign entities to the peers/server. In HyVVE the
entities assigned to a peer are those belonging to the
intersection of its CCAOI, a circular area represent-
ing a super-set of its AOI, and of its Voronoi area.
We consider a superset of the AOI of a peer to imple-
ment a prefetching of the entities close to the border
of the AOI of the peer. The server manages all the
entities belonging to areas not covered by the CCAOI
of any peer. The solution where a peer manages all
the entities in its Voronoi area is not feasible, because
when a few peers are present in the virtual world,
the number of entities assigned to a peer would be
huge, and the peer would not be able to support such
a load, especially if its computational capability is
not high. HyVVE supports a load distribution mech-
anism which scale in a natural way. When the world
is scarcely populated, most entities are mapped to the
server. However, the server load is not high in this
case, because a few entities are accessed and modi-
fied by the peers. When the number of the peers in-
creases, the amount of entities assigned to the peers
increases proportionally so that the load is distributed
between the peers. In a crowding scenario, i.e. when
a huge amount of peers gather in the same region of
the world, the entities located in that region are man-
aged by these peers, while the server manages entities
belonging to inhabited regions of the virtual world.
In this scenario, where peers are close to each other
and their AOIs intersect, the ownership of the entities
is defined by considering the Voronoi regions of the
peers. When a conflict occurs because an entity is
located in the CCAOI of a set of peers, the Voronoi
tessellation is exploited to determine the owner of the
entity. Since each point of the space is mapped to a
unique Voronoi Area, each entity of the virtual world
is assigned to a unique node peer.
The paper is organized as follows. Section 2 dis-
cusses the state of art in the area of Voronoi based
virtual worlds. Section 3 introduces the mathemati-
cal notions which are the basis of our approach, while
section 4 discusses how Voronoi tessellation can be
exploited for the definition of virtual environments.
Section 5 describes the architecture of HyVVE. A
set of experimental results are presented in Section
6, while Section 7 reports some conclusion and dis-
cusses future works.
2 RELATED WORKS
Voronoi diagrams and Delaunay-based overlays are a
well know solution to maintain network topology for
P2P virtual environments.
One of the first works along this line is VON (Hu
et al., 2006; Hu et al., 2008; Jiang et al., 2008).
VON exploits a Voronoi division of the DVE in order
to manage event dissemination in a scalable manner.
VON defines an overlay such that each peer maintains
a direct connection with all the peers within its AOI.
In order to maintain overlay connectivity, each peer
also has a direct link with peers that may also be out-
side of the AOI. To reduce bandwidth consumption,
VON has been further upgraded with an enhanced
event dissemination system (Jiang et al., 2008) and
state management (Hu et al., 2008).
VoroGame (Buyukkaya and Abdallah, 2008; Cav-
agna et al., 2008; Cavagna et al., 2009), proposes an
hybrid architecture for the management of passive en-
tities. Their architecture combines a Voronoi-based
network and a Distributed Hash Table (DHT). Two
different peers, one for each overlay, are responsible
for each entity in the DVE. Voronoi nodes are respon-
sible and maintain a copy for any of the entities that
are in their Voronoi area. They also maintain, for each
of these entities, a list of peers that have to be notified
for a state change of the given entity. This list is peri-
odically sent to the corresponding DHT node, whose
task is to broadcast state updates.
The work in (Varvello et al., 2007) proposes a
solution to deal with cluster of players in Delaunay-
based topologies. They employ a flooding messaging
strategy to spread notification inside the AOI. How-
ever, when a peer detects the message rate to exceed
its maximum capacity, it triggers a procedure for clus-
ter management. This procedure logically collapses
the cluster to a single point, allowing communications
to temporarily skip many small neighbouring Voronoi
regions, which helps reduce the communication over-
head. The approach has been proved effective with
realistic movement traces from Second Life.
The definition of a P2P overlay for MMOG based
upon Voronoi Diagrams has been investigated in
(Bonotti et al., 2007; Ricci and Salvadori, 2007; Gen-
ovali and Ricci, 2009; Genovali and Ricci, 2008a;
Genovali and Ricci, 2008b; Ricci et al., 2011) (Jiang
et al., 2008; Lee and Lam, 2008; Ohnishi et al.,
2005b; Hu et al., 2006). Some recent proposals
(Ohnishi et al., 2005b; Varvello et al., 2007; Hu et al.,
2006) have discussed the benefits of defining an over-
lay where the P2P connections correspond to the links
of a Delaunay Triangulation generated by considering
the locations of avatars of the DVE (Ghaffari et al.,
DCNET2013-InternationalConferenceonDataCommunicationNetworking
16
2009). According to this proposal, each peer is paired
with a site of a Voronoi diagram defined on the vir-
tual space and the position of the peer is exploited
to define the space partition. In this way, the area
corresponding to a peer P includes all the points of
the DVE which are closer to P with respect to any
other. The Delaunay Triangulation corresponding to
the Voronoi tessellation defines the P2P overlay con-
necting the peers.
In a MMOG any event generated by a peer should
be notified to any other peer in its AOI. This no-
tification may be implemented through a AOI-cast
mechanism (Albano et al., 2009), i.e. an applica-
tion level multicast constrained within the boundary
of the area of interest. Flooding the heartbeat through
the Voronoi links generates a large amount of redun-
dant messages and presents evident scalability prob-
lems. A more refined approach dynamically com-
putes a spanning tree on the Voronoi links including
all the peers of the AOI and exploits this tree to notify
the heartbeat. Both solutions are based on forward-
ing, i.e. any heartbeat is routed to the peers in the AOI
through a sequence of intermediate peers. An obvious
drawback of these solutions is the high latency in the
deliveryof an event, especially in crowding scenarios,
i.e. when a set of peers lie close to each other in the
virtual world and their AOI overlap. In this case sev-
eral routing hops may be required to notify an event
due to the large amount of peers located in the AOI.
The resulting latency may be not tolerable in MMOG
and may compromise the interactivity of the applica-
tion. On the other hand, since the number of Voronoi
neighbours is 6 on average (Aurenhammer, 1991), in
this solutions a peer manages a small number of con-
nections.
An alternative solution (Hu et al., 2006) defines
direct links between a peer and any other one in its
AOI. The resulting overlay includes these links be-
sides the Voronoi ones, which have to be maintained
to guarantee the connectivity of the overlay. This so-
lution minimizes the latency, but increases the num-
ber of connections of each peer. In a crowding sce-
nario like the one previously described, a peer should
manage a large number of connections, since a large
amount of peers are located in its AOI.
The problem of maintaining the Voronoi structure
of the overlay in a dynamic P2P environment, like a
MMOG, where the positions of the peers change con-
tinously and no centralized coordination entity does
exist, is a challenging issue. It is worth noticing
that several distributed algorithm for the management
of the Voronoi overlay have been recently proposed.
(Baraglia et al., 2012; Lee and Lam, 2008; Ohnishi
et al., 2005a; Kato et al., 2006) propose protocols
to build and maintain Delaunay triangulation-based
overlay networks. The approach present in (Baraglia
et al., 2012) is based on a gossip approach which re-
sults particularly suitable in a dynamic environment
like a DVE.
3 VORONOI DIAGRAMS AND
DELAUNAY TRIANGULATIONS
This section introduces the basic mathematical con-
cepts we have exploited for the definition of HyVVE.
A Voronoi diagram, (Aurenhammer, 1991) also
referred as Voronoi tessellation, is a special kind of
decomposition of a metric space determined by the
distances of the points of the spaces to a specified dis-
crete set of entities in the space, i.e. the sites.
Let us denote the Euclidean distance between two
points p and q by dist(p,q).
Definition 1. Let S = {s
1
, s
2
, ..., s
n
} be a set of n dis-
tinct points in the plane, i.e. the sites. The Voronoi
Diagram of S is a partition of the plane into n cells,
one for each site in S, such that the point q belongs
to the cell corresponding to a site s
i
if and only if
dist(q, s
i
) < dist(q, s
j
)s
j
S, i 6= j.
Figure 1: A Voronoi diagram.
In the following, we will denote the Voronoi Dia-
gram of S by Vor(S) and the cell corresponding to a
site s
i
by V(s
i
). Fig. 1 shows the Voronoi Tessellation
defined by the set of sites represented by black dots.
Each colored region represents V(s
i
), where s
i
is the
site corresponding to the black dot belonging to the
region.
A Delaunay Triangulation is a mathematical
structure dual with respect to the Voronoi Tessellation.
A Delaunay triangulation Dt(P) for a set P of sites
in the plane is a triangulation, i.e. a partition of the
HyVVE-AVoronoibasedHybridArchitectureforMassivelyMultiplayerOn-lineGames
17
plane into a set of triangles, such that the circumcir-
cle of any triangle in Dt(P) is empty, i.e. it does not
include any other point in P.
Given a set of n sites S = {s
1
, s
2
, ..., s
n
} of the
plane, the Delaunay triangulation is the dual struc-
ture of the Voronoi diagram, where the sites corre-
spond to the vertexes of the triangles, and an edge of
a triangle connects two vertexes s
1
, s
2
if and only if
V(s
1
) and V(s
2
) share a common edge, i.e. s
1
and s
2
are Voronoi neighbours.
Figure 2 shows a Delaunay Triangulation on the
top of s Voronoi diagram, where the borders of the
Voronoi regions are shown by dotted lines and the cor-
responding Delaunay Triangulation links are shown
by continuous lines.
Figure 2: A Delaunay triangulation on top of a Voronoi di-
agram.
4 VORONOI-BASED MMOG
In a Voronoi based approach, the position of each peer
in the virtual world is exploited to define a Voronoi
tessellation of the virtual world. Given n sites corre-
sponding to the peers, a Voronoi tessellation partitions
the virtual world into n areas such that the area cor-
responding to a site n includes all the points which
are closer to n with respect to any other site. Two
sites are Voronoi neighboursiff the borders of their ar-
eas overlap. The connected graph defined by linking
neighbour sites corresponds to the Delaunay Trian-
gulation associated to the Voronoi tessellation. A P2P
overlay is defined by connecting peers whose sites are
Voronoi neighbours. In the following, the links of this
overlay will be referred as Voronoi links.
. The adoption of this solution presents relevant
advantages:
Mapping of Entities to the Peers: a straightfor-
ward mapping of entities to the peers assigns each
entity to the peer which manages the Voronoi re-
gion where the entity is located.
Bandwidth Saving: since each site of a Voronoi
tessellation has on the average 6 neighbours (Au-
renhammer, 1991), each peer manages a bounded
number of connections with other peers, i.e. those
corresponding to the Delaunay links.
Overlay Connectivity: the connections corre-
sponding to the Delaunay links guarantee that the
overlay is connected. Even if a peer is located in
an uninhabited region of the virtual world, it re-
mains connected with the rest of the DVE through
the Delaunay connections.
Existence of Routing Algorithms for Delaunay
Networks: compass routing is based on a fast-to-
compute angle argument which exploits the math-
ematical properties of geometric networks and has
been proved to be cycle free for Delaunay net-
works. The algorithm can be exploited to define
efficient AOI-cast mechanisms.
5 HyVVE: THE ARCHITECTURE
Even if the definition of a pure P2P network for
MMOG is a challenging alternative to the classical
client/server solution, several problems should still be
solved for the definition of a comprehensive solution.
One of the main problems still to be solved concerns
the management of the MMOG state when the num-
ber of peers is very low or zero. The main problem
when a low number of peers belong to the MMOG
is related to the high load assigned to each peer. A
further problem is the maintenance of the state of the
MMOG when all the peers have left it, because this
state should be restored later when some peer joins
the MMOG.
To manage the problem of state persistence, we
propose HyVVE, an hybrid architecture including a
small number of ”classical” servers controlling the
state of the MMOG and a huge amount of interacting
peers. Note that this solution differs from a solution
based on the definition of Super-Peer based architec-
ture, because the set of servers is statically defined,
while the Super-Peers are dynamically elected, they
participate to the MMOG as normal peers and support
the further task of routing the eventnotification for the
peers they manage. In HyVVE, the server controls the
join of peers to the MMOG, their authentication and
manages a portion of the MMOG state. For the sake
of simplicity, we consider a system where a single
server S is defined. The server is a supervisor which
does not belong to the P2P overlay and is connected
DCNET2013-InternationalConferenceonDataCommunicationNetworking
18
to all the peers. When a peer enters the MMOG or up-
dates its position, it notifies this event to the server so
that the server continuously has a vision of the whole
MMOG and is able to compute a Voronoi tessellation
including all the peers of the MMOG. In this way, it
is able to exploit the tessellation to distribute portions
of the state of the MMOG to the joining peers. On the
other way round, HyVVE differs from client/server ar-
chitectures because a distributed protocol is exploited
to exchange events like positional and object updates
directly between the peers, without any intervention
by the server. This avoid that the server becomes,
like in classical client/server solutions, a bottleneck
for the entire system. The server is involved at a peer
bootstrap and then it receives the positional updates
of the peer, but these are not forwarded to other peers,
instead they are exploited by the server to decide if
it must give up entities to the peers or acquire enti-
ties from them. To implement the distributed protocol
supporting events exchange, in HyVVE the peers are
connected by a Delaunay overlay which is built and
maintained through GoDel, a gossip-based protocol
proposed in (Baraglia et al., 2012). The description
of this protocol is out of the scope of this paper, nev-
ertheless it is worth noticing that it is a light-weighted
protocol which maintains the overlay by a straightfor-
ward application of the equiangular property of the
Delaunay triangulations (Aurenhammer, 1991).
Let us consider now the Voronoi-based tessella-
tion enabling a natural mapping of the entities of the
MMOG to the peers. Each entity is mapped to the
peer whose Voronoi region includes it so that each en-
tity is managed by a single peer of the MMOG. The
problem of this approach is that, when the MMOG in-
cludes a small number of peers, a large number of en-
tities may be associated to a single peer which may be
not able to manage all them. As a matter of fact, when
a few peers are present in the MMOG all entities are
partitioned between them and a single peer may result
overloaded. To avoid peer overloading, we associate
with each peer a new circular area, the CCAOI, Cen-
tered Coordination AOI, whose center is the position
of the peer and whose radius is larger than of the AOI.
The goal of the CCAOI is to reduce the number of
entities assigned to a peer when its Voronoi Area is
too large. As a matter of fact, in this solution each
peer manages the entities located inside the Intersec-
tion Area, IA, i.e. the area corresponding to the inter-
section between Voronoi Area and its CCAOI, while
the entities located outside the Intersection Area of
any peer are assigned to the server. Both these areas
change dynamically when the peer moveslike its AOI.
As Figure 3 shows, the area managed by the peer
is only the blue one while the yellow area does not
Figure 3: AOI and CCAOI.
Figure 4: Each Entity is managed by a peer.
belong to any Interaction Area and the entities located
in this area are managed by the server.
The server initially owns the state of the whole
MMOG. When a peer joins the DVE, it first contacts
S for the authentication, then it receives from S, and
from its Delaunay neighbours the set of entities be-
longing to its Intersection Area.
It is important to note that when the the CCAOI
is totally included in the Voronoi area of the peer, the
IA overlaps the CCAOI. On the other way round, if
the CCAOI is a super-set of the Voronoi Area, the IA
overlaps the Voronoi Area of the peer. In a third sce-
nario the IA is the portion of the CCAOI overlapping
the Voronoi area of the peer.
The first case corresponds to a scenario where a
very small number of peers are present in the MMOG.
Note that in this case the Voronoi Area is much larger
than the CCAOI. In this scenario, the introduction of
the Interaction Area enables each peer to take care
only of the coordination of the closer entities, i.e. the
entities located inside its IA while the server manage
the entities located within its Voronoi Area, but not
belonging to its Intersection Area, i.e. the entities lo-
HyVVE-AVoronoibasedHybridArchitectureforMassivelyMultiplayerOn-lineGames
19
cated far from it.
Note that in this scenario the server does not be-
come a bottleneck for the system, even if a large num-
ber of entities are mapped to it because of the presence
of a few peers. As a matter of fact, the probability
that the peers update the entities mapped to the re-
gions managed by the server, i.e. the yellow regions
in Fig. 3, is low, because these entities are located
far away the peers. Note that as the number of peers
decreases, the entities are assigned back to the server
that, in a natural way, acquires the total control of the
system, when the last peer leaves the MMOG. In this
case the server acts as a backup server for the MMOG
state, and when each peer exits the MMOG, the state
of the MMOG will be stored by the server to be re-
stored later. Furthermore, in this way the load of the
peers is reduced.
When the number of peers increases, the number
of entities owned by the server decreases, because it
delegates the managementof the entities to the joining
peers. In this scenario, as in a crowding situation, the
Interaction Area of each peer may overlap its Voronoi
region and the management of the state of the MMOG
may be delegated entirely to the peers, as shown in
Figure 4, where the management of the MMOG is to-
tally delegated to the peers and the server owns no
entity. In this scenario the only task of the server is to
control and authenticate the peer joining the network
since all the entities are managed by the peers.
Consider now a crowding scenario, for instance
one where peers fight against each other and a large
number of peers is concentrated in a small portion
of the virtual space, as showed in Figure 5. In this
case, the Voronoi Area of each peer is included in its
CCAOI, hence the IA of the peer overlaps its Voronoi
Area and, despite the large number of peers, the area
that the server must manage is very large. Even in this
situation, the server does not become a bottleneck,
because it does not receive updates for the entities it
owns since they are located far away from the peers.
Again its task is to store the state of the entities and to
decrease the load of the peer.
Figure 5: A Crowding Scenario.
Figure 6: Entity delegation between peer and the server.
In our solution, the server itself becomes, com-
pared to the classical client/server model, both a
backup and a load distribution mechanism.
5.1 Entities Delegation
When a new peer enters the MMOG or the overlay is
modified due to the movement of the peers, the server
checks if some entity it owns falls within the IA of a
peer and, in this case, it sends the entity to this peer.
The peers instead have a limited knowledge of the
MMOG because of the local information obtained by
the direct connections with peers that fall in their AOI
and with their Delaunay neighbours. When a peer P
receives information from a neighbour V about either
a position update or a new neighbour notification, P
updates its local Delaunay Triangulation with the new
neighbourspositions and check whether any of the en-
tities owned fall in the IA of its neighbours or in the
server area. In both cases, P is no longer the owner of
the entity and sends the entity to the new owner.
For instance, in Figure 6, the peer P moves from
left to right and the entity O, first included in the
IA(P) managed by P, because of the shift of P, en-
ters the area under server competence. If we observe
the movement of P, from left to right, the entity O
would be assigned to the server.
In Figure 7 we see the exchange of an entity be-
tween peer P and P
1
. Dotted lines show the CCAOI
and the Voronoi region of P before its movement,
when the entity O just falls in IA(P). When P moves
following the arrow, then O enters IA(P
1
) and P
1
be-
comes the new owner of O by receiving from P all the
information.
DCNET2013-InternationalConferenceonDataCommunicationNetworking
20
Figure 7: Entity delegation between peers.
6 EXPERIMENTAL RESULTS
This section describes a set of experiments whose
goal is to evaluate the load of the server and of the
peers and the number of transfers between the server
and the peers for management of the entities. The ex-
periments have been conducted by varying the num-
ber of peers, the radius of the CCAOI and the speed
of the peers.
We have exploited PeerSim (Montresor and Je-
lasity, 2009), an highly scalable simulator for P2P
networks. In the experiments we have considered a
2-dimensional virtual environment of size 600× 800
and 1000 simulation cycles. At the start up of each
simulation, peers are positioned at random on the
map, afterwards the the peers move according to the
random way-point mobility model (Bettstetter et al.,
2004).
We consider 1000 entities uniformly positioned in
the virtual world. Initially, all the entities belong to
the server, then the server transfers subset of enti-
ties to the interested peers during the simulation. The
entities may be then exchanged directly between the
peers.
The goal of the first set of experiments is to evalu-
ate the average number of entities owned by a peer
and by the server by considering scenarios charac-
terized by different number of peers, radius of the
CCAOI and avatars’ speed. The speed is the number
of pixels covered by a single movement of a peer. We
fix the number of entities to 1000, while the number
of peers varies in the range [1. . . 1000] with a step of
100. We consider two values for the CCAOI radius,
i.e. 50 and 10 pixels, while the speed of the avatars
varies in the range [1. . . 2, 5] with a step of 0.5.
Fig. 8 shows the average number of entities owned
by the server, while the average number of entities
owned by each peer is shown in Fig 9. Note that,
when the radius of the CCAOI is fixed at 50 or at 10,
the behaviour of the function is the same for differ-
ent avatars’ speeds. For this reason, the lines cor-
responding to different speed are completely over-
lapped. First of all, we observe that the server load de-
creases when the number of peers increases, because
a larger set of peers contributes to the management
of the entities. It is interesting to note that the server
does not own any entity when the radius of the CCAOI
is equal to 50 and the number of peers is larger than
300 because, in this case, the IA of the peers covers
the whole virtual environment, no entity is managed
by the server and the partition of the entities among
the peers is determined by the Voronoi partition of
the virtual world. Also the load of the peers decreases
when the number of peers increases, but, while the re-
duction is remarkable when the radius of the CCAOI
is 50, it is negligible when the radius is 10. As a mat-
ter of fact, in the latter scenario, the size of the CCAOI
is very small and each region of the virtual world is
covered by a single peer. On the other side, when
the size of the CCAOI is larger, some portions of the
virtual world are covered by a set of peers, and this
implies a reduction of the average load of a peer.
Fig. 11 shows the average number of entity trans-
Figure 8: Average Number of Entities of the Server with
1000 objects.
Figure 9: Average Number of Entities for each peer with
1000 objects.
HyVVE-AVoronoibasedHybridArchitectureforMassivelyMultiplayerOn-lineGames
21
fers initiated by a peer and Fig. 10 those initiated
by the server, at each simulation cycle. First of all,
note that the number of transfers is influenced by the
avatars’ speed. As a matter of fact, the difference be-
tween the IA of a peer before and after its movement
is larger when the speed increases and this implies a
larger number of entities to acquire.
The probability that the IA of a peer is a subset
of its Voronoi area is higher when the radius of the
CCAOI is small, even when the number of peers in-
creases. In this case, a peer must acquire entities from
the server at each movement. As a matter of fact note
that the number of transfers from the server is larger
when the size of the CCAOI is 10 with respect to 50.
Figure 10: Average Transfers of the Server with 1000 ob-
jects.
Figure 11: Average Transfer of a Peer with 1000 objects.
Figure 12: Average Number of Entities for the Server: Vari-
able CCAOI.
When the radius of the CCAOI is 50 and the num-
ber of peers is larger than 300 no transfer between
the server and the peers occurs since, as observed be-
fore, the server does not manage any entity. As far as
concerns the transfers initiated by the peers, the av-
erage number of transfers for each peer does not de-
pend from their number when the size of the CCAOI is
small, because, as observed before, the number of en-
tities managed by each peer is nearly constant. When
the CCAOI is large, the average number of transfers
of a peer at each cycle decreases proportionally to the
number of peers, because when the number of peers
is high, no entity is transferred between peers and the
server.
The last set of experiments evaluates the aver-
age number of transfers initiated by the server/by the
peers by considering different radius of the CCAOI.
Note in Fig 12 that, when the radius of the CCAOI
is equal to 20, the average transfers initiated by a the
server increases until the radius of the AOIis 400, then
it decreases. As a matter of fact, the IA of each peer
is included in its Voronoi Area, when the number of
peers is lower than 400 and the number of transfers in-
creases with the number of peers. When the number
of peer is larger than 400, the peers starts to exchange
entities so that the number of transfers with the server
decreases. This phenomenon occurs for smaller num-
ber of peers when the radius of the CCAOI is larger.
7 CONCLUSIONS
This paper presents an hybrid architecture for MMOG
based on a Voronoi tessellation of the virtual world.
Our approach allows a dynamical distribution of the
load for the management of the entities of the virtual
world among the server and the peers. We have in-
troduced the concept of CCAOI to avoid peer over-
loading. The CCAOI allows to reduce the number of
objects assigned to each peer. We plan to consider
further mobility models to evaluate our approach in
different scenarios. Further, we are considering sev-
eral distributed algorithms for the management of the
consistency of the entities managed by the peers.
ACKNOWLEDGMENTS
The authors thank Katia Monni for her support in de-
veloping the experiments.
DCNET2013-InternationalConferenceonDataCommunicationNetworking
22
REFERENCES
Albano, M., Quartulli, A., Ricci, L., and Genovali, L.
(2009). Aoi cast by tolerance based compass routing
in distributed virtual environments. In Proceedings
of the 8th Annual Workshop on Network and Systems
Support for Games, NetGames ’09, pages 13:1–13:2.
Aurenhammer, F. (1991). Voronoi diagrams - a survey of a
fundamental geometric data structure. ACM Comput.
Surv., 23(3):345–405.
Baraglia, R., Dazzi, P., Guidi, B., and Ricci, L. (2012).
Godel: Delaunay overlays in p2p networks via gos-
sip. In IEEE P2P, International Conference on P2P
Computing, pages 1–12.
Bettstetter, C., Hartenstein, H., and P´erez-Costa, X. (2004).
Stochastic properties of the random waypoint mobility
model. Wireless Networks, 10(5):555–567.
Bharambe, A., Pang, J., and Seshan, S. (2006). Coly-
seus: a distributed architecture for online multiplayer
games. In Proceedings of the 3rd conference on Net-
worked Systems Design & Implementation - Volume 3,
NSDI’06, pages 12–12, Berkeley, CA, USA. USENIX
Association.
Bonotti, A., Genovali, L., and Ricci, L. (2007). A publish
subscribe support for networked multiplayer games.
In Proceedings of the Third IASTED European Con-
ference on Internet and Multimedia Systems and Ap-
plications, EurolMSA ’07, pages 236–241.
Buyukkaya, E. and Abdallah, M. (2008). Efficient trian-
gulation for p2p networked virtual environments. In
Proceedings of the 7th ACM SIGCOMM Workshop on
Network and System Support for Games, NetGames
’08, pages 34–39.
Carlini, E., Ricci, L., and Coppola, M. (2013). Flexible
load distribution for hybrid distributed virtual environ-
ments. Future Generation Comp. Syst., 29(6):1561–
1572.
Cavagna, R., Abdallah, M., Buyukkaya, E., and Bouville,
C. (2008). A Framework for Scalable Virtual Worlds
using Spatially Organized P2P Networks”. In IEEE
ICPADS workshop on Peer-to-Peer Network Virtual
Environment, P2P-NVE.
Cavagna, R., Abdallah, M., Buyukkaya, E., and Bouville,
C. (2009). “VoroGame: A Hybrid P2P Architecture
for Massively Multiplayer Games”. Ieee.
Genovali, L. and Ricci, L. (2008a). “JaDE: A JXTA Sup-
port for Distributed Virtual Environments”. In 13th
IEEE Symposium on Computers and Communications
Program, ISCC, Marrakesh,Morocco.
Genovali, L. and Ricci, L. (2008b). “Voronoi Models for
Distributed Environments”. In ACM CoNEXT Student
Workshop, Madrid, Spain.
Genovali, L. and Ricci, L. (2009). Aoi-cast strategies for
p2p massively multiplayer online games. In Proceed-
ings of the 6th IEEE Conference on Consumer Com-
munications and Networking Conference, CCNC’09,
pages 1317–1321.
Ghaffari, M., Hariri, B., and Shirmohammadi, S. (2009). A
delaunay triangulation architecture supporting churn
and user mobility in mmves. In Proceedings of the
18th international workshop on Network and oper-
ating systems support for digital audio and video,
NOSSDAV ’09, pages 61–66.
Hu, S., Chang, S., and Jiang, J. (2008). Voronoi state man-
agement for peer-to-peer massively multiplayer online
games. In Consumer Communications and Network-
ing Conference, 2008. CCNC 2008. 5th IEEE, pages
1134–1138. IEEE.
Hu, S.-Y., Chen, J.-F., and Chen, T.-H. (2006). Von: a scal-
able peer-to-peer network for virtual environments.
Netwrk. Mag. of Global Internetwkg., 20(4):22–31.
Jiang, J., Huang, Y., and Hu, S. (2008). “Scalable AOI-cast
for Peer-to-Peer Networked Virtual Environments”. In
ICDCSW The 28th International Conference on Dis-
tributed Computing Systems Workshops.
Kato, H., Eguchi, T., Ohnishi, M., and Ueshima, S. (2006).
Autonomous generation of spherical p2p delaunay
network for global internet applications. In Creat-
ing, Connecting and Collaborating through Comput-
ing, 2006. C5’06. The Fourth International Confer-
ence on, pages 184–191. IEEE.
Knutsson, B., Lu, H., Xu, W., and Hopkins, B. (2004). Peer-
to-peer support for massively multiplayer games. In
INFOCOM 2004. Twenty-third AnnualJoint Confer-
ence of the IEEE Computer and Communications So-
cieties, volume 1, pages 96–107. IEEE.
Lee, D. and Lam, S. (2008). “Efficient and Accurate Pro-
tocols for Distributed Delaunay Triangulation under
Churn”. In ICNP.
Montresor, A. and Jelasity, M. (2009). PeerSim: A scalable
P2P simulator. In Proc. of the 9th Int. Conference on
Peer-to-Peer (P2P’09).
Ohnishi, M., Nishide, R., and Ueshima, S. (2005a). In-
cremental construction of delaunay overlaid network
for virtual collaborative space. In Proceedings of the
Third International Conference on Creating, Connect-
ing and Collaborating through Computing, C5 ’05,
pages 75–82.
Ohnishi, M., Nishide, R., and Ueshima, S. (2005b). “In-
cremental Construction of Delaunay Overlay Network
for Virtual Collaborative Space”. In ”Third Int. Conf.
on Creating, Connecting and Collaborating through
Computing”.
Ricci, L., Genovali, L., Carlini, E., and Coppola, M. (July
2011). ”AOI-cast by Compass Routing in Delau-
nay Based DVE Overlays”. In The 2011 Interna-
tional Conference on High Performance Computing
and Simulation, HPCS 2011, Istanbul, Turkey.
Ricci, L. and Salvadori, A. (2007). “Nomad: Virtual En-
vironments on P2P Voronoi Overlays”. In 1th Int.
Work. on Peer to Peer Networks, PPN’07, Vilam-
oura,Portugal.
Varvello, M., Biersack, E., and Diot, C. (2007). Dynamic
clustering in delaunay-based p2p networked virtual
environments. In Proceedings of the 6th ACM SIG-
COMM workshop on Network and system support for
games, NetGames ’07, pages 105–110.
Yu, A. P. and Vuong, S. T. (2005). Mopar: a mobile peer-
to-peer overlay architecture for interest management
of massively multiplayer online games. In Proceed-
ings of the international workshop on Network and
operating systems support for digital audio and video,
NOSSDAV ’05, pages 99–104.
HyVVE-AVoronoibasedHybridArchitectureforMassivelyMultiplayerOn-lineGames
23