A LOCATION AND BANDWIDTH AWARE P2P VIDEO ON DEMAND

SYSTEM FOR MOBILE DEVICES

Danilo F. S. Santos and Angelo Perkusich

Embedded System and Pervasive Computing Laboratory, Department of Electrical Engineering

Federal University of Campina Grande, C.P. 10105 - 58109-970 - Campina Grande - PB - Brazil

Keywords:

Multimedia, Video on Demand, P2P Network.

Abstract:

Due the increasing number of new wireless technologies and the advent of new mobile devices, new services

and applications are necessary to supply the needs of these devices users. Video on-Demand (VoD) services

have become one of these services. In this paper we present a proposal of a P2P Mobile Video on-Demand

(VoD) System focused on mobile devices where they can act as both, media servers as well as clients. Due

to different possible communication interfaces that are available for mobile devices, we considered that they

can be in networks with different throughput characteristics, and also, they can be distributed through the

Internet. Therefore, our proposal make possible to build content distribution trees in the application layer

considering two major aspects related to mobility: the available bandwidth that the mobile device can share

with the network, and; the relative position of the mobile device in the network.

1 INTRODUCTION

In the last few years, the development of new wire-

less technologies and the advent of embedded mobile

devices, such as PDAs and smart phones, are grow-

ing rapidly. These new devices have embedded cap-

ture media devices, such as video cameras and audio

recorders, therefore, users can create multimedia con-

tent. Also, with the fast growing of peer-to-peer (P2P)

networks in the Internet (Wai-Pun Ken Yiu and Chan,

2007), users may share their content with other users.

As a result of this emerging scenario, new services

and applications are necessary to supply the needs

of the users. Video on-Demand (VoD) services have

become one of these services (Thouin and Coates,

2007). The main objective for VoD is to provide users

with the possibility to watch videos from the begin-

ning at any time.

In the context of mobile devices few works pro-

vide VoD services taking into account mobility re-

quirements. Most of the work related to VoD systems

(Hu, 2001) (Ma and Shin, 2002) relies on the use of

multicasting streams to share the video ﬂow with the

maximum number of clients. However, due to the

nature of the Internet, multicasting is not a realistic

option.

Considering the issues mentioned above, in this

paper we present a proposal of a P2P VoD System fo-

cused on mobile devices where they can act as both,

media servers as well as clients. Thus, they can share

multimedia content, in the context of this paper video,

between nodes in an overlay network. Due to differ-

ent possible communication interfaces that are avail-

able for mobile devices, we considered that they can

be in networks with different throughput character-

istics, for instance a GPRS or a WLAN network,

and also, they can be distributed through the Internet.

Therefore, based on the proposal introduced in this

paper it is possible to build content distribution trees

in the application layer considering two major aspects

related to mobility: the bandwidth available that the

mobile device can share with the network, and; the

relative position of the mobile device in the network.

The rest of the paper is organized as follow. In

Section 2 we present a general overview of our pro-

posal, and describe the proposed tree algorithms. In

Section 3 we present a ﬁrst analysis of the building

tree algorithms. Finally, the conclusion is presented

in Section 4.

2 PROPOSAL ARCHITECTURE

In this section we present a general overview of the

proposed architecture, focusing in the media distribu-

249

F. S. Santos D. and Perkusich A. (2008).

A LOCATION AND BANDWIDTH AWARE P2P VIDEO ON DEMAND SYSTEM FOR MOBILE DEVICES.

In Proceedings of the International Conference on Wireless Information Networks and Systems , pages 249-252

DOI: 10.5220/0002027702490252

 SciTePress

tion protocol and its main characteristics. In the de-

scription that follows we assume that a video V has a

duration D and is divided in T segments of D/T dura-

tion.

2.1 General Overview

The proposed P2P VoD system is divided in two over-

lay networks:

• The control network, which is responsible for

the communication between client nodes and the

server node that stores the original content. For

the architecture a P2PSIP based network is pro-

posed (iet, 2007).

• The distribution network, which is responsible to

distribute the content (video segments) between

the nodes using point-to-point connections in a

P2P based way. The content is distributed using

a single source application layer tree, then using a

different overlay network than the previous one.

The option to use two different overlay networks

was done to make possible use other mechanisms (be-

yond SIP) to control the distribution network. This

approach is shown in Figure 1.

Figure 1: Overall Architecture.

2.2 Obtaining Network Information

In the proposal presented in this paper it is necessary

to collect network context information to distribute

the media content between mobile nodes. Speciﬁ-

cally, our building algorithm relies on two types of in-

formation: the Available Bandwidth and the Position

information. To collect this information, we propose

the use of two widely-used and light-weight tools:

• Vivaldi algorithm (Dabek et al., 2004) to calculate

the relative coordinates of the host in the network;

• Pathload (Jain and Dovrolis, 2002) tool to esti-

mate the available bandwidth of a path.

2.3 Forming the Distribution Tree

After receive all requests of one layer, the server node

arranges the incoming nodes in a single source tree

application layer. We propose three tree building al-

gorithms:

• A simple bandwidth aware algorithm;

• A bandwidth and position aware algorithm;

• A fast bandwidth and position aware algorithm.

2.3.1 Simple Bandwidth Aware Tree - SBA

This algorithm is based on the Breadth-First Search

algorithm (Knuth, 1997). It uses two data struc-

tures: one which holds nodes not visited by the algo-

rithm (whiteList ), and other that holds visited nodes

but without adjacent (children) nodes (grayList). The

nodes that have already been visited and have children

nodes are called black nodes.

The algorithm starts inserting the source node in

the grayList, initializing all other nodes with it dis-

tance from the source dsrc as inﬁnity and adding those

in the whiteList. It proceeds by getting one node X

from the grayList. Node X share bandwidth to support

other N nodes as it children. These N nodes are in-

serted in the grayList and removed from the whiteList.

Node X is set as a black node. The operation is re-

peated until the grayList is empty, so until all nodes

are visited and have assigned nodes as children, if

possible. This algorithm is showed in the following

pseudo code.

Listing 1: SBA Pseudo Code.

1 SBA( AllNo d es , sr c N o d e )

2 f o r e a c h n o d e n i n Al lN od es :

3 c o l o r [ n ] : = WHITE

4 d s r c [ n ] : = i n f i n i t y

5 end f o r

6 c o l o r [ s r c N o d e ] : = GRAY

7 d s r c [ src No de ] : = 0

8 p u sh ( g r a y L i s t , sr c N o d e )

9 w h i l e ( g r a y L i s t != 0 )

10 u : = p u l l F i r s t ( g r a y L i s t )

11 fo r i = 1 t o N[ u ] :

12 n : = p u l l F i r s t ( w h i t e L i s t )

13 c o l o r [ n ] : = GRAY

14 p a r e n t [ n ] : = u

15 d s r c [ n ] : = d s r c [ u ] + 1

16 pu s h ( g r a y L i s t , n )

17 end f o r

18 c o l o r [ u ] : = BLACK

19 end w h i l e

2.3.2 Bandwidth and Position Aware Tree - BPA

This algorithm is based on the Dijkstra’s algorithm

(Dijkstra, 1959). It uses three data structures, the

WINSYS 2008 - International Conference on Wireless Information Networks and Systems

250

same two structures described in the previous algo-

rithm (whiteList and grayList), and also a list that hold

gray nodes with available bandwidth to support other

children (almostBlackList).

The algorithm starts inserting the source node in

the grayList, initializing all other nodes with it dis-

tance from the source dsrc as inﬁnity and adding

those in the whiteList. It proceeds by getting (and

removing) the closest node X to the source from the

grayList. This node is inserted in the almostBlack-

List. Then, for each node in the grayList and whiteList

(let’s merge these two lists in one whiteAndGrayList)

it is calculated the new distance from the source ele-

ment through each element in almostBlackList. This

is the “relax” process of the Dijkstras algorithm. All

visited nodes are inserted (or updated) in the grayList.

Listing 2: BPA Pseudo Code.

1 BPA( ALLNodes , s rc No de )

2 f o r e a c h n o d e n i n ALLNodes :

3 d s r c [ n ] : = i n f i n i t y

4 c o l o r [ n ] : = WHITE

5 end f o r

6 c o l o r [ s r c N o d e ] : = GRAY

7 d s r c [ src No de ] : = 0

8 i n s e r t ( g r a y L i s t , sr c N o d e )

9 w h i l e ( g r a y L i s t != 0 )

10 u : = e x t r a c t −min−d s r c ( g r a y L i s t )

11 al m o s t B l a c k : = a l m o s t B l a c k U {u }

12 p : = p a r e n t [ u ]

13 i f p i s n o t u :

14 N[ p ] : = N[ p ] − 1

15 i f N[ p ] = 0 :

16 rem o v e [ a l m o s t B l a c k , p ]

17 c o l o r [ p ] : = BLACK

18 fo r e ac h node v i n w h i t e A n d G r a y L i s t :

19 i f p a r e n t [ v ] = p :

20 f o r e a c h n o d e b i n a l m o s t B l a c k :

21 RELAX( b , v )

22 end f o r

23 end f o r

24 e l s e

25 f o r e a ch node v i n wh i t e A n d G r a y L i s t :

26 RELAX( u , v )

27 en d f o r

28 end w h i l e

30 RELAX( u , v )

31 i f ( d i s t a n c e O f ( u , v ) + d s r c [ u ] < d s r c [ v ] )

32 d s r c [ v ] : = d i s t a n c e O f ( u , v ) + d s r c [ u ]

33 p a r e n t [ v ] : = u

34 i f ( c o l o r [ v ] = WHITE)

35 c o l o r [ v ] : = GRAY

36 i n s e r t ( g r a y L i s t , v )

37 e l s e

38 u p d a t e ( g r a y l i s t , v )

When one node in the almostBlackList uses all N

connections (do not support more children), then it

is removed from that list and all nodes in the white-

AndGrayList must update its dsrc information (relax

process). We check this every time a parent is as-

signed to a node. The operation is repeated until the

grayList is empty. This algorithm is described in the

following pseudo code.

2.3.3 Fast Bandwidth and Position Aware Tree -

FBPA

This algorithm is based on the ﬁrst algorithm pre-

sented, but using the “relax” process of the previous

algorithm. Therefore, it uses two data structures, the

whiteList and grayList.

As the previous algorithms, this one starts insert-

ing the source node in the grayList, initializing all

other nodes with it distance from the source dsrc as

inﬁnity and adding those in the whiteList. It pro-

ceeds by extracting the closest node X to the source

node from the grayList. Node X share bandwidth to

provide other N nodes as it children. For all nodes

in the whiteList it is calculated the distance from X.

Then, the N closest nodes from X are inserted in the

grayList. Node X is set as a black node. The operation

is repeated until the grayList is empty. This algorithm

is described in the following pseudo code.

Listing 3: FBPA Pseudo Code.

1 FBPA ( AllN o des , sr cN od e )

2 f o r e a c h n o d e n i n Al lN od es :

3 c o l o r [ n ] : = WHITE

4 d s r c [ n ] : = i n f i n i t y

5 end f o r

6 c o l o r [ s r c N o d e ] : = GRAY

7 d s r c [ src No de ] : = 0

8 p u sh ( g r a y L i s t , sr c N o d e )

9 w h i l e ( g r a y L i s t != 0 )

10 u : = p u l l F i r s t ( g r a y L i s t )

11 fo r e ac h node v i n w h i t e L i s t :

12 d s r c [ v ] : = d i s t a n c e O f ( u , v ) + d s r c [ u ]

13 end f o r

14 fo r i = 1 t o N[ u ] :

15 n : = e x t r a c t −min−d s r c ( w h i t e L i s t )

16 c o l o r [ n ] : = GRAY

17 p a r e n t [ n ] : = u

18 pu s h ( g r a y L i s t , n )

19 end f o r

20 c o l o r [ u ] : = BLACK

21 end w h i l e

3 PRELIMINARY ANALYSIS

In this section we analyze the three building algo-

rithms, comparing the mean distance from the source

node against the elapsed time during the calculation

of the tree. The nodes were inserted in a network

topology and its coordinates were normalized to be

in a Cartesian plan between (0,0) and (1000,1000).

Also, the nodes could share their bandwidth with

A LOCATION AND BANDWITDH AWARE P2P VIDEO ON DEMAND SYSTEM FOR MOBILE DEVICES

251

a maximun of three others nodes. The simulations

were done using topologies with 10, 100, 1000, 2500,

5000, 7500 and 10000 nodes. The ﬁrst graph on Fig-

ure 2 shows the mean distance from the source for

each topology. The second graph on Figure 3 shows

the relative elapsed time to calculate the tree for each

topology. This elapsed time is relative to the calcula-

tion time of SBA, which is the fastest one.

Figure 2: Mean distance from the source node.

Figure 3: Relative calculation elapsed time.

Despite of the fact that the mean distance of BPA

is shorter than FBPA, the relative elapsed time when

calculate a tree with more than 1000 nodes of the BPA

grows faster than the elapsed time of FBPA. Then,

when calculating the distribution tree, a tradeoff be-

tween the number of nodes and complexity of the al-

gorithm must be considered. Especially when dealing

with a big amount of nodes (more than 1000), and

nodes with restricted processing capabilities, such as

mobile devices.

4 CONCLUSIONS

In this paper we presented an architecture for a mobile

P2P Video on-Demand system, which uses the avail-

able bandwidth and position information of the nodes

to efﬁciently distribute the multimedia content. It was

discussed why use two overlay networks, one for the

control and negation between the nodes, and another

to distribute the content.

Also, we have shown three tree building algo-

rithms that take into account both, the available band-

width as well as the position information of the nodes

to build trees with low mean distance from the source.

We compared the algorithms considering the process-

ing time to calculate the tree, and the mean distance

from the source of each node. We have shown that

a tradeoff between the BPA (Bandwidth and Position

Aware Tree) and FBPA (Fast Bandwidth and Position

Aware Tree) algorithms must be considered related to

the number of nodes in each calculation.

The next step in our work is the simulation of the

system as a whole in a more realistic topology. In this

simulation we intend to analyze the delay for video

delivery. Also, we plan to simulate others cases, such

as changes of available bandwidth of a node during

a transmission, leave of a node, failure of a node,

change in the position of a node, and other speciﬁc

cases. Finally, after the simulation stage, a full imple-

mentation of the system is planned in order to validate

the system as a whole.

REFERENCES

(2007). IETF P2PSIP Working Group.

Dabek, F., Cox, R., Kaashoek, F., and Morris, R. (2004).

Vivaldi: A decentralized network coordinate system.

Dijkstra, E. W. (1959). A note on two problems in connex-

ion with graphs. Numerische Mathematik.

Hu, A. (2001). Video-on-demand broadcasting protocols:

a comprehensive study. In Proceedings. INFOCOM

2001. Twentieth Annual Joint Conference of the IEEE

Computer and Communications Societies, pages 508

– 517. IEEE.

Jain, M. and Dovrolis, C. (2002). Pathload: A measurement

tool for end-to-end available bandwidth.

Knuth, D. E. (1997). The Art Of Computer Programming.

Ma, H. and Shin, K. G. (2002). Multicast video-on-

demand services. SIGCOMM Comput. Commun. Rev.,

32(1):31–43.

Thouin, F. and Coates, M. (2007). Video-on-demand

networks: Design approaches and future challenges.

IEEE Network, 21:42 – 48.

Wai-Pun Ken Yiu, X. J. and Chan, S.-H. G. (2007). Chal-

lenges and approaches in large-scale peer-to-peer me-

dia streaming. In IEEE Multimedia, Vol. 14, No. 2,

pages 50–59.

WINSYS 2008 - International Conference on Wireless Information Networks and Systems

252