NEW HYBRID P2P COMMUNICATION MODELS FOR REMOTE

TERRAIN INTERACTIVE VISUALIZATION SYSTEMS

Ricardo Olanda, Mariano P´erez and Juan Manuel Ordu˜na

Departamento de Inform´atica, University of Valencia, Burjassot, Valencia, Spain

Keywords:

Interactive Terrain Visualization, Real Time Graphics, Peer to Peer.

Abstract:

Over the last years, there has been a great development on real time terrain visualization applications us-

ing remote databases. One of the main problems that these applications must face is the system scalability.

These applications usually use a client-server model that cannot support a large number of concurrent requests

without using a considerable number of servers. In this paper, we propose a new hybrid P2P models for ter-

rain interactive visualization systems. The comparative performance evaluation results show that the system

throughput achieved by these strategies can be more than three times higher than the hybrid P2P strategy

proposed in the literature, signiﬁcantly improving the scalability of these systems.

1 INTRODUCTION

Real-time terrain visualization is a very active re-

search ﬁeld in the area of computer graphics. We can

ﬁnd a large number of applications, where one of the

main tasks is to display virtual terrain models at inter-

active frame rates.

This virtual terrain information is usually stored

in large remote server databases. Since users visual-

ize only a small portion of this information, usually

they prefer to download only the required informa-

tion, rather than store the whole databases in theirs

computer local disks. Nowadays, the terrain visual-

ization applications usually use a client-server model

(C/S model) to access these remote databases over the

Internet due to its ease of management. However, this

model has a limited system scalability.

In order to solve this problem, peer to peer (P2P)

models could be used. There are two main types of

P2P models (Scholmeier, 2001): pure P2P models,

where each computer node act as client and server

simultaneously, and hybrid P2P models, where ad-

ditional computer nodes act as system server exclu-

sively. A pure P2P model presents some important

disadvantages in a terrain visualization application

with respect to C/S models, than can be avoided using

hybrid P2P schemes.

In this paper, we propose new hybrid P2P models.

Using them, the number of required servers can be

signiﬁcantly reduced with regard to the classical C/S

model, while still providing the same performance.

2 RELATED WORK

P2P models refers to a network where communica-

tions take place with direct connections between peer

nodes, without using any dedicated server. Most net-

works and applications usually contain some non-

peer elements which can act as clients, servers or

both (Zhu, Gong, Liu, Song and Zhang, 2007). De-

spite that, these networks and applications are usu-

ally called P2P, but it would be called hybrid P2P

(Scholmeier, 2001).

One of the most important problems to be solved

in P2P communication models is which other nodes

must be classiﬁed as “neighbors” for each node, in

order to require them the information needed. We can

assume that two users placed in a similar location in

the virtual scene will require a similar terrain infor-

mation. Therefore, the list of neighbors of a given

visualization node could be formed by the nodes that

are closely located to it in the virtual scene. Solipsis

(Frey et al., 2008) or Vorogame (Byukkaya, Adbal-

lah and Cavagna, 2009) are examples of Distributed

Virtual Environments (Singhal and Zyda, 1999) that

use this strategy. Nevertheless, the location of users

in the virtual world is not related at all with the physi-

cal location of their computing nodes. As a result, the

network latency among distant computing nodes may

prevent this strategy to reach the desired efﬁciency.

Other works also considers additional parameters

in order to obtain the list of neighbor nodes, like trans-

mission data time, data availability or network band-

413

Olanda R., Pérez M. and Manuel Orduña J..

NEW HYBRID P2P COMMUNICATION MODELS FOR REMOTE TERRAIN INTERACTIVE VISUALIZATION SYSTEMS.

DOI: 10.5220/0003812404130418

In Proceedings of the International Conference on Computer Graphics Theory and Applications (GRAPP-2012), pages 413-418

ISBN: 978-989-8565-02-0

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

width (Royan, Gioia, Cavagna and Bouville, 2007).

However, all these works do not deal with large ter-

rain databases, and therefore, they do not focus on the

problems associated with the transmission of this type

of information.

Zheng, Yu, Li and Gau in PeerTR (Zheng, Yu, Li

and Gau, 2007) propose a speciﬁc model for real time

terrain visualization. This model use a mixed solu-

tion where clients can obtain information from others

clients or from a server, reducing the system server

workload compared with classic C/S model. A de-

tailed description of this work is done in section 3.1.

3 HYBRID P2P MODELS

Visualization delays in a terrain visualization applica-

tions context are unacceptable, because clients require

a ﬂuid navigation experience over the scene. Hence,

continuous availability of all the terrain information

must be guaranteed. In a pure P2P model it cannot

been guaranteed, since nodes can join or leavethe sys-

tem at any moment. Also, an initial terrain database

distribution among the existing nodes is required in

order to avoid a single point of failure and a possi-

ble bottleneck. However, this distribution is infeasible

because it cannot be predicted when these nodes will

join into the system. In order to solve these problems,

we have deﬁned several new hybrid P2P models.

3.1 Strategy 1: PeerTR-based Model

PeerTR (Zheng, Yu, Li and Gau, 2007) is an hybrid

P2P model proposed for real time terrain visualiza-

tion. A given client node can obtain terrain informa-

tion from the nodes in its list of neighbor clients. In

order to guarantee permanent data accessibility to the

whole terrain database, there is a server which appears

in the list of neighbor clients of all the clients. Each

client connected to the system stores the downloaded

terrain information in its local storage, periodically

reporting to other clients about the availability of the

information stored.

We havedeveloped a new hybrid P2P model based

on the PeerTR model. This model is composed of

two layers (Figure 1): service layer and visualization

layer. The service layer provides to the users a con-

tinuous terrain data access, guaranteeing a lower re-

sponse time compatible with a real time terrain visu-

alization. This layer is composed of memory nodes

(terrain database and support server nodes) and con-

nectivity nodes. The Visualization layer groups the

user computers that perform a terrain visualization.

Figure 1: Connexion scheme of hybrid P2P strategies.

In the original PeerTR system, there is a dedicated

server which stores the complete terrain database,

and it also manages the topology of the system. In

order to avoid a single point of failure and a pos-

sible bottleneck, we have split this server into two

kind of servers: memory nodes, that serve the ter-

rain database, and connectivity nodes, that manage

the system topology.

We also have deﬁned two new types of memory

nodes: terrain database nodes, which stores the whole

terrain database, and support server nodes, which pro-

vides terrain information to the visualization nodes

when they cannot obtain it from their neighbor nodes.

Typically, users are often interested in a reduced por-

tion of the terrain database. For this reason, support

servers could store a reduced region instead the whole

database, acting as “cache” nodes. In order to guaran-

tee the continuous access to whole database informa-

tion, this node will access to the terrain database node

when terrain data are not found in its cache.

The Connectivity Node is a special support server

that is able to manage the topology of the system. It

sends periodically to each visualization node a list of

neighbor visualization nodes and a support server in

order to require them the terrain data.

A Visualization Node refers to a client which per-

forms the interactive visualization of terrain data. The

terrain data to be displayed are downloaded from

other neighbor visualization nodes or from the sup-

port servers, and they are stored in a local disk cache.

Although this node already existed in the original

PeerTR model, it has been modiﬁed: ﬁrst, the list

of neighbor visualization nodes size has been lim-

ited, in order to add the same computing overhead

regardless of the number of neighbor visualization

nodes connected to the system. Second, the cache

size has been limited, so it only stores the most re-

cently downloaded terrain data, reducing the visual-

ization node storage requirements. Third, the visual-

ization node will periodically report which terrain in-

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

414

formation stores only to the visualization nodes in its

neighbor list, instead to all visualization nodes con-

nected to the system, reducing signiﬁcantly network

and computing overhead.

3.2 Strategy 2: Neighbor Nodes Query

The process of reporting other nodes about the ter-

rain information stored in a given visualization node

cache may produce a signiﬁcant overhead. In or-

der to avoid this overhead, we have developed a new

strategy that uses the same scheme of the strategy 1

(Figure 1), but where terrain information reporting

is avoided. In order to obtain the terrain data, a vi-

sualization node sends a message with the required

data to each node of its list of neighbor visualization

nodes. Each neighbor node answers back indicating

which required data are available in its cache. The

report also provides other parameters like the node

workload status or the network transmission latency.

These parameters are used by the visualization node

to estimate the response time of each neighbor node.

According to this information, the visualization node

selects one o more neighbor nodes in order to perform

a concurrent download of the terrain data required. If

none of the neighbor nodes stores the terrain data re-

quired or the estimated response time is too high (it

can be a serious problem for a ﬂuid terrain visualiza-

tion in real time), the visualization node requests the

data terrain to its assigned support server.

The use of this strategy may supposes a signiﬁ-

cant reduction in the number and size of the message

exchanged, reducing the overall processing time.

3.3 Strategy 3: Specialized Server

Cache

In a terrain visualization application, visualization

nodes usually require data about the region that they

are visualizing. In both strategies 1 and 2, a sup-

port server is assigned to each visualization node ex-

clusively using the criterion of the current workload

in the existing support servers. In order to improve

the use of the support servers, we have deﬁned a

new strategy where the assignment of a given sup-

port server to each visualization node also takes into

account the region of the scene displayed by the visu-

alization node, selecting the support server that cur-

rently stores in its cache the greatest amount of terrain

data required by the visualization node and supports

a low workload. This support server selection is dy-

namic, changing over time in order to satisfy both cri-

teria. The scheme of this strategy is the same that the

one used in the strategies 1 and 2 (Figure 1). Obvi-

ously, this strategy makes sense when there are more

than one support server in the system.

4 PERFORMANCE EVALUATION

A remote terrain visualization system using an hybrid

P2P model can consist of a large number of clients,

requiring a lot of human and material resources that

result unaffordable for a single research team. There-

fore, we have implemented and tested an execution-

driven simulator that can measure the performance of

the hybrid P2P models when they are used in a remote

terrain interactive visualization system.

4.1 Simulator Characteristics

We have implemented a centralized, execution driven

simulator of the P2P system written in C++ which

follows a discrete event simulation methodology

(Sadoun, 2000). This simulator supports multiple

peer-to-peer networks structures and different net-

work characteristics like message transmission time,

network contention, transmission errors, network de-

lays or node saturation. In order to validate the simu-

lator, several tests have been carried out varying the

simulator conﬁguration parameters. Some of these

parameters are: number of visualization nodes, num-

ber of support servers, cache size, request process

time, transmission time, transmission error, etc.

4.2 Simulator Validation

In order to validate a simulated model, it should be

compared with another reference model that can be

either real or simulated (Sargent, 2005). Since there

is not a comparable simulator for terrain visualization

systems in the literature, we also have implemented a

real terrain visualization system to validate it. A re-

duced number of visualization nodes has been used

(between 10 and 35), due to the limited resources

available. With this number of nodes it is not nec-

essary a large number of servers, so only one terrain

database server, one connectivity node and between 1

and 3 support servers have been used.

In a terrain visualization application, the move-

ment of the users in the virtual world can be quite

different over time. Some different initial user’s posi-

tion distribution and movement patterns distributions

are usually used to evaluate DVE systems (Morillo,

Rueda, Ordu˜na and Duato, 2007). We are going to

use similar distributions and patterns to evaluate our

new models. Initially, user’s position on the map is

NEW HYBRID P2P COMMUNICATION MODELS FOR REMOTE TERRAIN INTERACTIVE VISUALIZATION

SYSTEMS

415

randomly selected from an uniform distribution. Af-

ter that, users can move following different movement

patterns: randomly around all virtual world (uniform

distribution), towards only one “hot point” or towards

several “hot points” (Figure 2).

Figure 2: Distributions of users obtained using three differ-

ent movement patterns. From left to right: uniform, clus-

tered with one “hot points” and clustered with multiple “hot

points”.

The terrain database used in the tests is the Puget

Sound database (Georgia Tech College of Comput-

ing, 2011). This database is usually used to test ter-

rain visualization applications due to its varied geog-

raphy. Different cache sizes have been tested for sup-

port servers and visualization nodes, according to this

database size. Although we do not show here the re-

sults due to space limitations, comparative examples

can be found in Olanda’s work (Olanda, 2010). All

the results show that the behavior of the real system

is very similar to the behavior of the simulator, with a

maximum relative error lower than 6%. These results

validate the simulator as a reliable tool for measuring

the performance of the proposed hybrid P2P models.

4.3 Evaluation Results

Different test have been performed in order to com-

pare the new hybrid models. We have used the simu-

lator to test these models with a high number of vi-

sualization nodes. In these tests, several simulator

parameters like cache sizes, request process time or

error time, have been ﬁxed by experimental tuning.

In order to select a transmission time, an interval of

possible message latency values has been ﬁxed us-

ing a study of actual broadband quality developed by

Oxford and Oviedo Universities (Oviedo and Oxford

Universities, 2009).

4.3.1 Movement Patterns Effects

We have ﬁrst studied how the different movement pat-

terns followed by users in the virtual scene may af-

fect the proposed strategies. This study has been per-

formed using two support servers connected. Figure 3

shows the average system response provided to visu-

alization nodes by strategy 3 for different movement

patterns.

Figure 3: Average system responses for different movement

patterns.

Figure 3 shows that, when users follow a uniform

movement pattern, the strategy 3 supports around

2500 visualization nodes. However, it supports up to

4000 nodes when users follow a movement pattern

with one “hot point”, and up to 6000 visualization

nodes when users follow a 4 “hot points” movement

pattern.

These signiﬁcant variations in the system through-

put are due to the fact that users move within a lim-

ited region when there are one or several “hot points”

speciﬁed. As a result, visualization nodes probably

store in their cache the information required from

other neighbor visualization nodes, (because all the

nodes require similar information), and the number

of requests from the visualization nodes to the support

servers decreases. In the case of several “hot-points”,

there are several regions where users tend to crowd

and this produces a better specialization of support

server caches, resulting in a lower response time.

According to this results, we can state that

the movement pattern producing the highest system

workload is the uniform movement pattern. In order

to measure the performance of the proposed strategies

in the worst case, we have used this movement pattern

in all the performance evaluation results shown in the

subsequent sections.

4.3.2 Single Support Server

The hybrid strategies 1 and 2 and the classical C/S

model have been compared in order to evaluate which

one provides a higher throughput (since only one sup-

port server is used, hybrid strategy 3 results are the

same that the strategy 2 ones).

Figure 4 shows the average system response times

for all the requests generated in the simulation as the

number of visualization node increases. As it could be

expected, the plot for the classical C/S model shows

the lowest throughput, reaching saturation when 100

visualization nodes are present in the system. How-

ever, the Strategy 1 supports 600 visualization nodes,

increasing the throughput ﬁve times with respect to

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

416

the C/S model. This improvement is due to the inher-

ent scalability of P2P models with respect to central-

ized schemes. That is, in Strategy 1 the nodes can get

the required information from other neighbor nodes,

avoiding the access to the server. The plot for Strat-

egy 2 shows that this strategy outperforms Strategy 1,

supporting around 800 nodes without reaching satu-

ration. This result shows that the strategy of asking

about terrain data to the neighbors when needed, is

better than periodically reporting about what informa-

tion contains each node. One of the reasons for this

behavior is that the reporting message have a size sev-

eral times greater than the size of the messages used

to transmit the terrain data.

Figure 4: Average system response times for one support

server.

Figure 5 shows the percentage of requests served

by support servers as the number of visualization

nodes increases. It is worth mention the signiﬁcant

decrease of this percentage for both hybrid strategies

for a reduced number of nodes connected to the sys-

tem, and how this percentage slowly decreases for a

high number of nodes. The reason for this behavior

is that as the number of visualization nodes increases,

a given node can obtain the required data from more

neighbors (that explain the quickly initial decrease).

However, there is always a signiﬁcant percentage of

visualization node requests that cannot be found in

a neighbor node (due to the large size of the terrain

database and the limited size of the visualization node

caches), so the support server has to solve them. The

non-signiﬁcant reduction of this percentage for a high

number of visualization nodes explains the server sat-

uration of the support server shown in ﬁgure 4: the

server workload steadily increases as the number of

visualization node increases.

According to these results, we can state that the

hybrid models provides a higher scalability than the

classical C/S model, and the new strategy 2 provides a

higher throughput (it allows a greater number of con-

nected visualization nodes) than strategy 1.

Figure 5: Average percentage nodes requests served solved

by the support server.

4.3.3 Several Support Servers

In order to evaluate the scalability of the different

strategies, we have measured the system performance

using more than one support server connected to the

system. Discarded the C/S model, the three hybrid

strategies have been compared.

Figure 6 shows a representative case of the per-

formance provided by the proposed strategies. Con-

cretely, it shows the average response time provided

to the visualization nodes by each considered strat-

egy when three support servers are used. This ﬁgure

shows that Strategy 2 provides a system throughput

that is 2.5 times greater than strategy 1, supporting

around 4000 visualization nodes without reaching sat-

uration. In turn, strategy 3 provides a throughput 20%

higher than strategy 2.

Figure 6: Average system response times for three support

servers.

Figure 7 shows the average percentage of requests

served by the support servers. This ﬁgure also shows

that the percentage of requests served by the support

servers are very similar for strategies 2 and 3. There-

fore, the greater throughput of the strategy 3 with re-

spect to strategy 2 shown in ﬁgure 6 is due to the “re-

gion caches” achieved in strategy 3, which provides a

higher number of cache hits (it is not needed to access

the terrain database), reducing the response time for

serving these requests. As a result, there is more avail-

able time for serving more requests. In order to prove

NEW HYBRID P2P COMMUNICATION MODELS FOR REMOTE TERRAIN INTERACTIVE VISUALIZATION

SYSTEMS

417

this statement, Figure 8 shows the percentage of vi-

sualization node requests found in the support servers

cache. The plots in this ﬁgure shows that strategy 3

achieves around 90% of cache hits from 2000 visual-

ization nodes up, while strategy 2 hardly reaches 70%

of cache hits.

Figure 7: Average percentage of visualization node requests

served support servers (using three support servers).

Figure 8: Average percentage of cache hits in the support

servers (using three support servers).

These results show that the specialization of the

support server cache achieved by the strategy 3 pro-

vides the highest system throughput of all the strate-

gies considered in this work, therefore providing the

highest level of system scalability.

5 CONCLUSIONS

In this paper, we have proposed a comparative study

of new hybrid P2P strategies for terrain visualization

systems that improve the scalability of the classical

C/S model. In order to measure the performance

achieved by each proposed strategy, we have devel-

oped and validated an execution-driven simulator.

The performance evaluation results show that the

best strategy consists of avoiding the periodical re-

porting among peer nodes about the current informa-

tion contained in each node, and also using the sup-

port servers as cache memories specialized by regions

of the virtual world. The system throughput achieved

by this strategy can be more than 3 times higher than

the hybrid P2P strategy proposed in the literature.

ACKNOWLEDGEMENTS

This work has been jointly supported by the Spanish

MICINN and European Commission FEDER funds

under grants Consolider-Ingenio 2010 CSD2006-

00046 and TIN2009-14475-C04-04.

REFERENCES

Buyukkaya, E., Adballah, M., Cavagna, R., 2009.

Vorogame: A Hybrid P2P Architecture for Massively

Multiplayer Games. IEEE Consumer Communica-

tions and Networking Conference.

Frey, D., Royan, J., Piegay, R., Kermarrec, A. M., An-

ceaume, E., Le Fessant, F., 2008. Solipsis: a decen-

tralized architecture for virtual environments. Interna-

tional Workshop on Massively Multiuser Virtual En-

vironments, 29–33.

Georgia Tech College of Computing. Puget

Sound Database, 2011. Retrieved Octo-

ber 4, 2011 from http://www.cc.gatech.edu

/projects/large

models/ps.html.

Morillo, P., Rueda, S., Ordu˜na, J. M., Duato J., 2007. A

latency-aware partitioning method for distributed vir-

tual environment systems. IEEE Transactions on Par-

allel and Distributed systems, 18 (9), 1215–1226.

Olanda, R., 2010. Visualizaci´on interactiva de terrenos ex-

tensos en entornos distribuidos sobre redes mixtas P2P

y cliente-servidor. Phd. Thesis.

Oviedo University, Oxford University, 2009. Global Broad-

band Quality Study Shows Progress, Highlights

Broadband Quality Gap. Retrieved October 4, 2011

from http://www.sbs.ox.ac.uk/newsandevents /Docu-

ments/BQS%202009%20ﬁnal.pdf.

Royan, J., Gioia, P., Cavagna, R., Bouville, C., 2007.

Network-based visualization of 3D landscapes and

city models. IEEE Computer Graphics and Applica-

tions, 70–79.

Sadoun, B., 2000. Applied system simulation: A review

study. Information Sciences, vol. 124, 173–192.

Sargent, R. G., 2005. Veriﬁcation and validation of sim-

ulation models. 37th Winter Simulation Conference,

130–143.

Schollmeier, R., 2001. A deﬁnition of Peer-to-Peer Net-

working for the classiﬁcation of Peer-to-Peer architec-

tures and applications. IEEE International Conference

on Peer-to-Peer Computing, 27–29.

Singhal, S., Zyda, M., 1999. Networked Virtual Environ-

ment. ACM Press.

Zheng, S., Yu, Z., Li, Z., Gau, L., 2007. PeerTR: a Peer-to-

Peer terrain roaming architecture. Advanced Parallel

Processing Technologies, LNCS 4847, 292–300.

Zhu, J., Gong, J. H., Liu, W. G., Song, T., Zhang, J. Q.,

2007. A collaborative virtual geographical environ-

ment based on P2P and grid technologies. Information

Sciences, 177 (21), 4621–4633.

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

418