3. Approach
Our approach is to use Voronoi diagram
to solve the neighbor discovery problem. We first describe what Voronoi is, and
then explain the designs of the P2P network.
3.1 Voronoi
explained
Given n points (referred to as sites)
on a 2D plane, we can partition the plane into n non-overlapping regions
such that all points within a region are closer to the region’s site
than to any other site (see Figure 4a). The region can be seen as the sphere of influence of a particular site. Voronoi diagram has certain nice properties: for example, once constructed, it can be used to easily identify the k-nearest neighbors
for any given site. Voronoi diagrams have been studied extensively for nearly a
century and applied in diverse fields. Voronoi diagram construction is beyond
the scope of this work and readers may refer to [10] for existing algorithms.
(a) (b)
Figure 4: Voronoi diagram
(a) The dots indicate sites, and lines define boundaries for regions.
(b) Squares (▓) represent enclosing neighbors; triangles (▲)
represent boundary neighbors; stars (★) are both
enclosing and boundary neighbors; circle (●) represents a regular AOI-neighbor.
3.2 Voronoi-based
P2P
As stated earlier, the ideal P2P design
for NVE is to let each node connect directly and exchange messages with only
its AOI neighbors. The key problem, however, is to discover new neighbors to
connect to, as nodes move around.
In our design, we require each node to
maintain direct connections with all of its AOI neighbors (i.e. nodes within
the AOI of a given node). Given the coordinates of its neighbors, a node may
then construct a Voronoi diagram containing all of its AOI neighbors. Discovery
of new neighbors is done with the help from boundary neighbors (defined
as AOI neighbors whose Voronoi regions overlap with the AOI boundary) as they may
know what other nodes exist beyond the AOI (see Figure 4b). As each node moves, it sends updates to all of its AOI neighbors. If a boundary neighbor receives the update, it would check for potential new neighbors on behave of the moving node and send out notifications if new
AOI neighbors are found. To ensure that the P2P topology remains
fully-connected even when a node has no AOI neighbors, we also require each
node to minimally maintain its enclosing neighbors (defined as the nodes
whose Voronoi regions immediately surround the given node, see Figure 4b).
We call this scheme Voronoi-based Overlay Network (or
VON for short, overlay is the academic term for P2P network, as it is
built on top of the physical Internet). Compared with existing architectures or
other P2P-based NVE designs, VON may achieve better scalability (as each
node only maintains a limited number of nodes, resource consumption is thus
bounded), responsiveness (because latency is minimized by the direct
connections between nodes), and message-efficiency (requests to check
for potential new neighbors are embedded in regular position updates, and
neighbor notification is sent only when necessary, see [6] for a complete
comparison).
3.3 Dynamic
AOI adjustments
To prevent a node from message overload
when crowding occurs (i.e. when many nodes come nearby to each other, see
Figure 5), we also propose dynamic adjustments to AOI boundary when necessary. The rules are simple: 1) AOI-radius is shrunk when the number of connections exceeds a predefined connection limit. 2) AOI-radius grows back
when number of connections has decreased. 3) Mutual awareness should be
maintained. So if node A cannot see node B due to overload, node B should also
shrink its AOI-radius to exclude node A.
Figure 5: Example of a crowding situation
4. Results
To evaluate VON, we run a number of
simulations to test the scalability, topology consistency, and reliability of
the design. Simulation parameters are shown in Table 1 and two metrics are used during the simulations:
Topology consistency [11] - defined as the ratio between the numbers of AOI neighbors that should be seen versus those actually observed. For example, if 5 neighbors exist within a node’s AOI, but only 4 are known, then topology
consistency would be 4/5 = 80%).
Drift distance [12] – Topology consistency does not capture the inconsistencies between actual and observed coordinates of a remote node. The absolute difference in coordinate values for each known node is called drift distance.
Table 1 Simulation parameters
|
|
Simulation Type
|
|
Parameters
|
Scalability
|
Reliability
|
|
World dimension (units)
|
1000x1000
|
1000x1000
|
|
Initial AOI-radius (units)
|
150
|
150
|
|
Packet loss rate*
|
0%
|
0% ~ 100%
|
|
Simulation time-steps
|
1000
|
1000
|
|
Velocity (units/time-step)
|
5
|
5
|
|
Maximum connection (in dynamic AOI)
|
10
|
10
|
|
Number of nodes (in increment of 20)
|
10 ~ 250
|
150
|
*
Packet loss rate indicates the loss in all messages regarding position updates
and neighbor discovery, which makes up 95% of all transmission. Other control
messages (5%) are still delivered reliably.
4.1 Scalability
A system remains scalable as long as each system
component does not deplete its resource with the addition of new nodes. In the
case of P2P-based NVE, as long as bandwidth use at each node is kept within the
node’s limit, the system may scale continuously. Therefore, to
qualitatively evaluate a NVE system’s scalability, a simple way is to
observe its bandwidth use at each node as a function of system scale.
Figure 6: Average transmission size per node per
second
Figure 6 shows that as the number of nodes increases, the average transmission size for each node increases linearly in the basic mode (i.e. fixed AOI radius) and logarithmically in the dynamic AOI model (i.e. adjustable
AOI-radius). This shows that resource consumption of VON is at least as good as
existing client-server architecture’s O(n) for the basic model, and may
achieve potentially much better scalability with the sub-linear consumption growth
under dynamic AOI adjustments.
In the dynamic AOI model, average transmission
size per second for each node approaches 3kb/second at 250 nodes (assuming 10
updates are generated per second), this is well within the bandwidth provided
by current broadband (which operates at about 16 ~ 32kb / second for upload and
64 ~ 128kb / second for download). However, the significance here is not the actual
transmission size, which depends heavily on the implementation, but that with
careful design, transmission may become bounded, which is the
characteristic for highly scalable systems.
This bounded transmission is explained by Figure 7, which shows the average number of connected neighbors and AOI neighbors for each node. It is clear that the number of neighbors becomes bounded due to the pre-defined connection limit.
Figure 7: Average neighbor size for basic and dynamic
AOI models
Figure 8: Comparison of transmission size between VON and client-server
If we compare VON against client-server architecture with equivalent
function, we see that a single node in VON utilizes much less bandwidth than a
typical server, by distributing message transmission to all participating nodes
(see Figure 8). This not only lowers system cost by removing the need for a powerful, high-bandwidth server, but also increases system robustness as no single point of failure exists.
4.2 Topology
consistency
Scalability would not be useful if
consistency is greatly sacrificed. We evaluate VON’s topology consistency
by observing topology consistency and drift distance matrices. Figure
9 shows that the topology consistency is close to 100% for both basic and dynamic AOI models, though there is a slight drop in dynamic AOI. However, consistency still remains high (above 99.70%) for up to 250 nodes.
Drift distance is likewise quite low, and remains close to 0 (see Figure 10). The slight inconsistency is likely caused by a temporary asymmetric understanding of the topology after AOI adjustments. However, inconsistency is bound to occur in a real network environment due to latency, the
more important question therefore is how quickly can it recover from inconsistency?
Figure 9: Topology consistency
for basic and dynamic AOI models
Figure 10: Average drift distance for basic and dynamic AOI models
This question is answered by Figure 11, which shows the number of steps to recover from inconsistency under dynamic AOI. Here VON recovers within an average of 1.5 steps, which is fairly fast. We can thus see that VON keeps fairly good consistency in a
robust manner.
Figure 11: Average number of steps to recover from inconsistency
4.3 Reliability
How good does VON perform when the
underlying network is not 100% reliable? We answer this question by performing
another set of simulations where packet loss ranges from 0% to 100%. As certain
messages are essential to maintain the topology (such as initial connection handshakes),
we keep them to be delivered reliably, but allow other types of messages to be
sent unreliably (mostly movement notifications and neighbor discovery messages,
which occupy about 95% of all messages).
In Figure 12 we see that VON still maintains fair consistency for a loss rate up to 50%, and only begin to drop after a loss rate of 60%. Figure 13 shows the corresponding average recovery steps, where VON can still recovers from inconsistency within 4 steps for a loss rate up to 50%.
Figure 12: Effect of loss rate on topology consistency (dynamic AOI model)
Figure 13: Effect of loss rate on average recovery steps
5. Potential Applications
5.1 MMOG
Massively Multiplayer Online Game is the
initial application area in which VON was designed for. By adopting VON, highly
scalable, responsive, and fault-tolerant MMOG may be developed and deployed
affordably. Although VON in its current form does not yet support the full
functional requirements of a MMOG, such as event consistency, data persistency,
or security, VON can be used to offload position updates from the servers (estimated
to be 70% of all message traffics), which can greatly reduce the cost for
server-side bandwidth.
5.2 Military
simulations
Large-scale military simulations where
human trainees go to immersive simulated battlefields were among the first NVE
applications. Security concerns regarding P2P (i.e. that the clients could be
hacked) may postpone the full adoption by game industry. However, in a military
context, as all the simulators are within military’s control, concerns
for client-side hacking is non-existent. It may be quite practical, even in its
present form, to adopt VON in large-scale military applications.
5.3 Scientific
simulations
Viewed from a more general perspective,
VON is in fact not just a platform for games, but a P2P-based environment that
supports spatially-oriented (i.e. 2D or 3D coordinate system) and tightly
synchronized (i.e. state information is exchanged at every time-step)
large-scale simulations. As such, another promising application would be in
scientific simulations where coordinate systems are the basis of the
simulation. For example, molecular dynamics (MD) that simulates physical
systems at atomic scale may be distributed on VON to parallelize the calculations.
References
[1] MMOG Chart. Available at: http://www.mmogchart.com/
[2] IGDA, “2004 Persistent Worlds
Whitepaper,” Available at: http://www.igda.org/online/IGDA_PSW_Whitepaper_2004.pdf
[3] Shun-Yun Hu and Guan-Ming Liao,
“Scalable Peer-to-Peer Networked Virtual Environment,” in Proc.
ACM SIGCOMM 2004 workshops on NetGames '04, Aug. 2004, pp. 129-133.
[4] Shun-Yun Hu, Jui-Fa Chen and Tsu-Han Chen,
“VON: A scalable peer-to-peer network for virtual environments,”
IEEE Network (accepted), 2006. Available at:
http://vast.sourceforge.net/docs/pub/2006-hu-VON.pdf
[5] VAST Project Home Page. Available at: http://vast.sourceforge.net
[6] Shun-Yun Hu, “Scalable Peer-to-peer
Networked Virtual Environment,” Master’s thesis, Tamkang Univ., Taiwan, 2005. Available at: http://vast.sourceforge.net/docs/pub/2005_hu_master_thesis.pdf
[7] Wikipedia, “definition of
peer-to-peer.” Available at: http://en.wikipedia.org/wiki/Peer-to-peer
[8] I. Stoica et al., “Chord: a
scalable peer-to-peer lookup protocol for Internet applications,”
IEEE/ACM Trans. Networking, vol. 11, no. 1, pp. 17-32, 2003.
[9] I. Foster and A. Iamnitchi, “On
death, taxes, and the convergence of peer-topeer and grid computing,” in
2nd Intl. Workshop on Peer-to-Peer Systems (IPTPS'03), Berkeley, CA (2003).
[10] F. Aurenhammer, “Voronoi
diagrams—a survey of a fundamental geometric data structure,” ACM
Computing Surveys (CSUR), vol. 23, no. 3, pp. 345-405, 1991.
[11] Y. Kawahara, T. Aoyama, and H. Morikawa,
“A peer-to-peer message exchange scheme for large-scale networked virtual
environments,” Telecommunication Systems, vol. 25, no. 3-4, pp.
353–370, 2004.
[12] C. Diot and L. Gautier, “A
distributed architecture for multiplayer interactive applications on the
Internet,” IEEE Network, vol. 13, no. 4, pp. 6-15, 1999.
any questions/comments/inquiries? please contact us
Copyright © 2005-2009, VAST Development Team
Last modified: June 30 2006.
