Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Optimal Overlay Construction for Wireless Multimedia Social Networks

Pietro Cassar a and Tommaso Melodia Wireless Networks and Embedded Systems Laboratory Department of Electrical Engineering University at Buffalo, The State University of New York e-mail: {pietroca, tmelodia}@buffalo.edu

AbstractNew-generation smart phones allow learning, visualizing, and sharing information about the daily activities of the users. Through these devices, users can exchange multimedia information such as digital video, audio and images. Wireless users in physical proximity that want to share these contents can initiate spontaneous social interactions by exchanging user proles. An algorithm is proposed in this paper to create a multimedia social network (MSN) to share video over an overlay topology. The algorithm is developed by following a cross-layer approach that takes into account characteristics of the overlay (application layer) and schedulability of ows at the medium access control (MAC) layer. The algorithm connects new users to the MSN through a link in the overlay topology that corresponds to a multi-hop path in the wireless network. To obtain a scalable network, the overlay topology is created with the objective to equalize the workload among network nodes. Centralized and distributed versions of the algorithm are presented. Simulations are conducted to compare the performance of both algorithms. Also, performance comparisons are provided with an algorithm that links a new user to the overlay following a minimum-hop policy. Results show that the performance of the distributed version is close to the performance of the centralized version, and that the proposed algorithm distributes the workload per node more fairly than the minimum-hop algorithm.

I. I NTRODUCTION Recent advances in sensing, computation, storage, and wireless networking are driving an increasing interest in multimedia [1], [2], [3] and people-centric [4], [5] sensing applications. In the latter, sensing devices, in the form of the ubiquitous mobile smartphones (e.g., Nokia E95, Iphone) are carried by individuals, thus enabling sensing, learning, visualizing, and sharing information about our daily activities. Recent work has also envisioned the convergence between peoplecentric and multimedia sensing and distributed content sharing (e.g., Wikipedia, Blogger) and social networks (e.g., Facebook, MySpace). In this view, smart phones/sensors aggregate participatory as well as sensory inputs from local surroundings. When plugged into the Internet, the collaborative inputs from phones may enable a distributed, detailed, and high-resolution view of the world in space and time [6], [7]. For example, CenceMe [6] leverages sensor-enabled mobile phones to infer people activities and then shares this presence through social network portals such as Facebook. In this context, we consider distributed systems to allow users to ubiquitously sense, store and share digital video, 1

audio, and images, and collectively refer to them as multimedia social networks (MSNs). In wireless MSNs, devices in physical proximity exchange user prole information to initiate spontaneous social interactions. We focus on social networking over handheld mobile devices in locations such as stadiums, conferences, expositions, galleries, bars and restaurants, which we assume to be endowed with wireless interfaces that enable ad hoc connectivity. We assume that network devices can selforganize in a multi-hop wireless network to share multimedia content through peer-to-peer (P2P) technology. Given the scenario described above, we address the key problem of dening a decentralized and asynchronous algorithm to construct an overlay topology to distribute multimedia content over a wireless multi-hop social network. Considerable work has been devoted to study techniques to construct overlay topologies with desirable characteristics for the wired Internet [8], [9], [10], [11]. However, these follow for the most part the traditional layered architecture of the Internet. Accordingly, the overlay topology is constructed independently of the underlying physical topology. However, in wireless networks of resource-constrained embedded devices, the overlay topology should be constructed in light of the topological characteristics of the wireless network and of the broadcast advantage offered by the wireless channel, i.e., a cross-layer design approach is imperative to efciently utilize scarce bandwidth and energy resources [12], [13], [14]. Previous work has also addressed overlay topology construction in wireless networks [11], [15], [16], [17]. However, most previous work considers peer-to-peer le sharing applications, and neglects the problem of providing a real-time service with minimum rate guarantees. Instead, real-time video sharing with resource-constrained devices requires careful topology construction and optimized assignment of resources to provide the quality of experience (QoE) [18], [19] required by the nal users. In particular, network scalability and fair energy consumption for overlay nodes are key requirements. For these reasons, a different approach is attempted in this paper. The difference lies in (i) the objectives/point of view; (ii) the design principles/criteria; (iii) the resulting algorithmic structure. The focus of this paper is to study algorithms for optimized overlay topology construction for real-time multimedia content sharing in spontaneous social networks. The key objectives of the algorithm are to guarantee

the scalability of the video streaming service, while at the same time providing a fair sharing of resources among the wireless devices involved in it. Our approach does not make assumptions on the underlying routing protocol. Instead, it assumes that a routing protocol is able to connect wireless nodes that belong to the overlay network. Based on the topological characteristics of the network, and on the routing protocol, the algorithm determines how to connect a new user that wishes to join the social network. A tree-shaped overlay topology is considered. For the new node, the algorithm selects the appropriate parent in the overlay topology, based on two key considerations, the rst of a user-centric nature, the second of a network-centric nature. First, the algorithm tries to provide the minimum bit rate needed by the video streaming application. From a network perspective, the algorithm tries to guarantee future scalability (i.e., possibility of adding new nodes) and fair energy consumption by equalizing the load among different devices. Since each link of the overlay network corresponds to a potentially multi-hop path in the underlying wireless network, we seek to create an overlay topology whose underlying wireless topology shows a good level of fairness in the transmission load among all wireless nodes. There are three fundamental characteristics that inform and characterize the design of the proposed algorithm: The algorithm creates a tree-shaped overlay to multicast a multimedia content; The algorithm is designed to guarantee the scalability of the overlay network, i.e., avoid bottlenecks. This is obtained through fair sharing of the workload among wireless nodes. The algorithm follows a cross-layer approach to jointly determine the overlay node to connect the new overlay peer based on considerations on schedulability of the generated ows at the medium access control layer, to ensure that the generated topology is able to support the generated trafc.

ows are derived in the Section V, and the proposed algorithm is presented in Section VI. Finally, performance evaluation results and conclusions are presented in Section VII and Section VIII, respectively. II. R ELATED W ORK Multicast in wireless networks is a well-studied problem [15], [16], [17]. However, traditional network-layer multicast suffers from high overhead, since information on the multicast groups has to be stored at each wireless node, which limits the level of scalability [20]. Moreover, network-layer multicast suffers from low robustness to disconnections caused by node mobility. In [20], the authors propose to realize multicast at the application layer through an overlay. The benets of application-layer multicast against a network-layer approach are also discussed. Application-layer multicast is easy to deploy since it does not require protocol, interface, or architectural changes at the network layer. Furthermore, if the network topology changes, links among overlay users remain unchanged, since an overlay link can be implemented through different (and potentially time-varying) wireless multihop paths. This limits the message overhead to maintain the multicast tree. Finally, application-layer multicast can exploit the capabilities of lower layer protocols in providing reliability, congestion control, and ow control. Previous work has also addressed the problem of creating Peer-To-Peer (P2P) overlay over wireless networks [21], [22]. In [23] the authors classify approaches to design P2P overlay protocols over wireless networks. These works offer a good overview about the benets of a P2P overlay in wireless networks. Among others, a P2P network allows sharing resources (e.g., CPU, bandwidth) among a large number of users in a decentralized fashion. However, none of these papers are focused on P2P overlay networks to guarantee end-to-end data rates to allow real-time multicast streaming. In [24], [25], [26] the authors present the protocols ALMA, EOMP and a new overlay multicast protocol to achieve rapid adaptation of overlays when nodes move, respectively. In these papers, the authors discuss why an overlay is a valid solution to provide multimedia streaming services. In [24], by exploiting the characteristics of the application layer, the authors design a receiver-driven protocol through which the nodes nd their connections according to application requirements. The protocol is highly adaptive in reconguring the tree in response to mobility or congestion. In [25], the authors develop an efcient multicast protocol (EOMP) through unicast routing information, based on an application overlay. Multicast state information of each multicast-group in EOMP is created by its unicast routing table. EOMP is able to limit the number of reconguration messages for the overlay, even if nodes move quickly. In [26], the authors propose an overlay multicast protocol to achieve rapid adaptation of the overlay structure when nodes move, and reduced delay. In this algorithm, to join or leave a multicast group, it is only sufcient for a member node to inform its rst upstream member node. This updates the tree structure more rapidly when nodes move. Finally, in 2

Fig. 1.

Overlay Graph and Wireless Graph

The remainder of this paper is structured as follows. Related work is discussed in Section II, while the system model is introduced in Section III. The problem denition and the optimization problem are described in Section IV. The necessary and sufcient condition for the schedulability of

[27], the authors address the problem of allocating hardware device resources for a multimedia overlay network assuming that end-users devices have different hardware capabilities. In [24], [25], [26], [27], the focus is on designing an overlay to provide real time multicast services such us IPTV or VideoOn-Demand. Consequently, these protocols are designed for an architecture with a central server and do not consider the impact of the MAC layer. III. S YSTEM M ODEL We model the multi-hop wireless network by a directed connectivity graph GW (VW , EW ), where VW = {v1 , ..., vN } is a nite set of wireless transceivers (nodes), with |VW | = N , and (i, j ) EW represents a unidirectional wireless link from node vi to node vj (referred to also as node i and node j , respectively, for simplicity). Let Si {j : (i, j ) EW } be the set of neighbors for node vi , and Ni its cardinality. We assume that all links are bidirectional, i.e., (i, j ) EW iff (j, i) EW . The overlay topology is represented by a tree-shaped graph GO (VO , EO ) over the wireless network, with |EO | = K . Each overlay node vi VO is an element of VW as well, while a link o(i, j ) EO between vj and vi VO corresponds, in general, to a path p(i, j ) in GW . Let the indicator function i(vi , vj ) equal to 1 iff o(i, j ) EO , and let it be 0 otherwise. A path p(i, j ) between two wireless nodes vi and vj is a loop-free sequence of wireless nodes, connected by a direct wireless link {vi , (i, k ), vk , , vj }. The path is obtained through a suitable wireless routing protocol, e.g., OLSR [28], AODV [29]. Hence, the two ends of a link of the overlay graph correspond to source and destination of a path in the underlying wireless network. Each node vi has a radio interface with maximum data rate ui [bit/s], which is assumed to be substantially higher than the average data rate of the multimedia content. We assume that nodes operate on a time-division periodic schedule. Each period is organized in NT time slots. The number of time slots per period depends on the offered trafc load and on the wireless topology. The duration of a time slot is referred to as ts [s]. A node vi that belongs to the overlay can transmit in multicast towards a group of users, called the multicast group. Node vi is linked to the users of a multicast group by wireless unicast multi-hop paths. Finally, we introduce the denition of conict between two wireless nodes. Since, we consider multicast transmissions, we assume a worst-case denition of conict, which will be then used to derive a lower bound (sufcient condition) for ow schedulability in Section V. Denition. There is a conict between two nodes vi , vk VW if there exists a node vj VW such that (i, j ) EW and (k, j ) EW exist. IV. P ROBLEM D EFINITION Our objective is to design an algorithm to construct an overlay tree GO (VO , EO ) for multimedia content distribution with a high level of scalability, for a given wireless network 3

topology GW (VW , EW ), and network-layer routing protocol. Each link of the overlay network (i.e., each path of the wireless network) should be guaranteed an average bit rate of r [bit/s]. To guarantee a high level of scalability the workload per wireless node must be distributed fairly among all nodes. We introduce the following denition. Denition. The workload wi of a node vi represents the fraction of time that vi is active to support real-time streaming of the multimedia content, as a receiver or as a transmitter. Hence, I t (vi ) (1) wi = tAi NT where Ai represents the set of time slots t, 1 t NT , during which vi is active, while I t () represents an indicator function, i.e., I t (vi ) = 1 iff node vi is active at time slot t, 0 otherwise. Intuitively, if the workload is distributed fairly, the number of time slots where a node is active is similar for all nodes in the wireless network, which in turn assures a high scalability level for the overlay network. By evenly distributing the workload each time a new node joins the overlay, we can prevent nodes from becoming bottlenecks and depleting their batteries. The notion of workload can be related to the bit rate per time slot provided by a node vi . If a wireless node is active during a given time slot t and there transmits/receives with ui data rate ui , we say that it serves a ow N . Now, suppose T there are K links composing the overlay network tree. Then, these K links correspond to K source-destination pairs in the wireless network, which concurrently share the network infrastructure. Each pair k , k = 1, ..., K is connected through a path p(i, j ), vi , vj VO in the wireless network, so in the wireless network we have p1 (i, j ), ..., pK (l, m). On each path pk (i, j ) a bit rate of r [bit/s] must be served, i.e., the minimum bit rate required by the multimedia content. Hence, each wireless node that is part of the path must serve a bit rate r [bit/s]. Let Pi represent the set of paths where node vi is active. We assume that a node vi that belongs to the overlay can transmit in multicast (and receive in unicast). Hence, during a single time slot, it can serve (by transmitting in multicast) the entire set of paths Pi . We refer to the ow transmitted by a node vi as fi . Similarly, we refer to a ow received by a node vi as fi . If vi is not in the overlay, the ow fi served by node vi is given by the sum of the ows per path fik , fi = kPi fik . In fact, in this case vi can be active in one path per time slot, because it can not transmit in multicast. Hence, the node needs to transmit in multiple time slots for a given period of scheduling. This case is shown in Fig. 2(a), where each outgoing link belongs to a differen path. In this example the node is active in path l at time slot V or in path n at time slot U . Instead, if vi is in the overlay, the ow fi is equal to any fik , fi = fik . In fact, in this case vi can transmit the same data for multiple paths in a single time slot, because it can

multicast to a group. Figure 2(b) shows an example where the node during time slot V transmits in multicast on the set of paths {l, m, n}. In the example, each outgoing link belongs to a different path and hence belongs to a different overlay link. For the received ow Fig. 2 shows that, regardless of whether vi belongs to the overlay or not, the ow received by vi is k k equal to any f i , fi = fi .

j Ni

kPj

k fj

uj

1 ts
(n)

(5)

vi GW , vi / GO k fj j Ni uj 1 ts

(6)

(n) vi GO (n) (n1) i(vnew , vj ) = 1 j VO P k k kPi (fi +fi ) fi = wi = u ui i

(7) (8)

Fig. 2.

Activity per node example

The workload per node vi has been dened as the ratio between the time that a node is active and the total scheduling period time. Hence, the workload can be expressed also as the ratio between the ow offered by vi , for a given scheduling period, and the node capacity ui as follows: wi =
tAi

I t (vi )

NT

fi ui

(2)

Note that Ai is the set of all time slots where node vi is active, for a given path pk (l, m) if vi / GO or for a given set of paths Pi if vi GO . An overlay network is started by a wireless node vi , which represents the source of the multimedia content, and which we o . The overlay topology then evolves in steps. We refer term v1 (n) (n) (n) to the overlay topology at step n as GO (VO , EO ). At each (n) step n, we assume that a new node vnew wishes to join the (n1) (n) (n) W / VO . The problem overlay, where vnew V , and vnew can be formulated as in Problem P1.

The problem above consists of determining the overlay (n) (n) link o(vnew , vj ) to connect vnew to the overlay through a overlay node vj this also determines the optimal overlay topol(n) (n) (n) ogy GO (VO , EO ) at step n, given the overlay topology (n1) (n1) (n1) GO (VO , EO ) at step n 1, the node willing to join (n) (n) the overlay at step n vnew and the set of overlay links Enew by (n) which vnew can be connected to the overlay. The optimality criterion is minimal maximum workload. The minimization (n) (n) takes into account that for each o(vnew , vj ) Enew the routing protocol returns a path. Hence, each overlay link correspond to a path, and which determines the workload per wireless node. Constraints (3-5) are derived from the necessary and sufcient condition for the schedulability of ows for a given wireless topology. These conditions will be derived in the next section. Constraint (7) assures that the overlay graph is connected. Since the solution of Problem P1 may require a (centralized) algorithm with high computational complexity, we will discuss a distributed low-complexity algorithm in the section VI. V. S CHEDULABILITY C ONDITIONS In this section, we derive the necessary and sufcient conditions for the schedulability of ows per node, for a given wireless topology and overlay network. First, we derive a necessary condition for the wireless network to guarantee that a node can transmit or receive an assigned ow. Lemma 1. If a node vi , with capacity ui , can transmit a ow fi and receive a ow fi for given scheduling period, and if I t (vi )T x , I t (vi )Rx are indicator functions that return 1 if the node transmits or receives during time slot t, respectively, and 0 otherwise, then the following inequality must hold:
tAi

P1 : Given : GW (VW , EW ), GO
(n) vnew , (n) Enew (n1)

(VO

(n1)

, EO

(n1)

),

F ind :

(n) {o(vnew , vm )}, (n1) vm VO (n)

o(vnew , vj ) Enew M inimize :


vi VW

(n)

max {wi } 1
(n)

I t (vi )T x + tAi I t (vi )Rx = NT fi + fi 1. ui

Subject to :

kPi

k P k fi + kP f i i ui

(3)

vi GW , vi / GO
k P fi + kP i k f i

Proof. A node can be in transmission mode for nT x time slots, in receive mode for nRx time slots. Hence, it must hold nT x + nRx NT , from which nT x nRx + 1. NT NT 4 (9)

ui

(4)

vi

(n) GO

The terms in the previous inequality represent the percentage of time that a node transmits, and the percentage of time that a node receives, respectively. From the denition of workload in (2), for the transmission and receive mode (9) becomes:
tAi

I t (vi )T x + tAi I t (vi )Rx = NT fi + fi 1. ui

(10)

Substituting (8) in (10), conditions (3) and (4) can be derived. Next, we derive the sufcient condition for ow schedulability. First, we introduce some denitions. Denition. A conict-graph on the network GW (VW , EW ) is a graph GC (VW , EC ), where for each pair of nodes vi , vj VW , the link (i, j ) EC iff a conict exists between the vi , vj . Denition. A multi-conict-graph on a conict graph GC (VW , EC ), is a graph GM (VM , EM ) where each node in GM is a node that belongs to GC with a given cardinality. The cardinality is the number of copies in VM of the node in GC . Each copy has the same link node in GC . Then, set of the i i i , where EC is Er the edge set of GM is EM = vi VM EC i the link set of node vi GC and Er is the link set among the copies of the node vi in GM . Denition. The maximum degree C of a multi-conict graph is max d(vi ), where d(vi ) represents the degree of node vi , i.e., the number of all incoming and outgoing links that belong to vi . Denition. The coloring index iC of the multi-conict-graph, is the minimum number of colors necessary to color all the nodes in the graph, so that two nodes with the same color are not linked by an edge.
vi Vm

The mapping between the multi-conict graph, the conictgraph and the wireless network can be intuitively explained. The nodes in the multi-conict graph are all the nodes in the conict-graph, but for each node in the conict-graph we assign a cardinality. The cardinality is calculated as the slot rate, i.e., the number of times slots per second required to support a ow fi . The slot rate needed is enough to serve the ow fi , avoiding the conicts between any node pair. The number of time slots per scheduling period to guarantee the slot rate per node is then calculated. The multi-conict graph can then be used for this purpose because the nodes that are not connected in the multi-conict graph do not have conicts. Moreover, each node has as many copies as its slot rate. Hence, through the multi-conict graph we can nd sets of nodes that can transmit at the same time for scheduling period to support the ow per node fi . This set represents the minimum number of time slots necessary to transmit all the ows avoiding conicts. The set of independent nodes can be calculated through results from the color graph theory [30], [31] applied to the nodes of the multi-conict graph, i.e., by calculating minimum number of colors necessary to color the nodes of the graph, so that no nodes that have the same color are connected by an edge. In our case, the minimum number of colors represents then the minimum number of time slots required by the network to offer the ow per node. In Fig. 4, we show the coloring of the multi-conict-graph of Fig. 3. In this case, we have a different color for each node because only a node can transmit in each time slot.

Fig. 4.

Coloring Multi-Conict-Graph example

The cardinality of the nodes in multi-conict graph nodes can be represented by the slot rate. The slot rate is dened as follows. If a node vi is able to transmit a ow fi with a link capacity of ui , then if the duration of a time slot is ts [s], the slot rate is [30]: fi sri = [slots/s]. t s ui (11)

Now, we are able to state the sufcient condition for the schedulabity of ows, given by the following lemma. Lemma 2. If the the following inequality is satised t tAj I (vj ) = NT j Ni fj 1 ts , uj
j Ni

Since fi and ui are rational numbers, we can nd ts such that sri is an integer for each node. Figure 3 illustrates an example of how to construct the multi-conict graph.

then a node vi , with capacity ui , can offer a ow fi . Proof. Using a result from vertex graph coloring [30], [31] we can assert that
Fig. 3. Multi-Conict-Graph example

i C C + 1 5

1 . ts

(12)

By the denition of multi-conict-graph, C can be rewritten as C = max sr(vj ) . vi


j Ni

(3)

Among the overlay links o(vnew , vj ) Enew select the one with associated path pk that minimizes the max workload per wireless node in the path.

Then, (12) becomes iC max


vi


j Ni

1 sr(vj ) + 1 . ts

From (11), we can write j Ni


j Ni

iC max
vi

t I ( v ) t j s tAj

max
vi

NT fj 1 +1 . uj ts ts

+1=

(13)

Substituting (8) in (13), we can derive (5) and (6). From these, the minimum number of time slots NT per scheduling period can be calculated, which will be equal to the index coloring iC , since, the scheduling period needs at least of C + 1 time slots. VI. D ISTRIBUTED A LGORITHM In this section, we describe an algorithm to create a tree shaped overlay that minimizes the maximum workload per node as in Problem P1. We developed an distributed algorithm where each overlay node makes a decision only based on local (1-hop neighborhood) information. In fact, we take into account unstructured networks, such as ad-hoc networks, where the only information available for the nodes is that about their neighbors. We name the algorithm Distributed Minimum Max Workload Algorithm (DMMWA). The algorithm, starting from the root of the overlay tree, creates a tree-shaped overlay, in which a new nodes (generated in random order) join the overlay step-by-step. In the DMMWA overlay nodes share information among them because they do not know the wireless topology. The information shared is the node capacity of the neighbors, the number of neighbors, the overlay topology and the vector with the workload per wireless node. The vector of the workload per wireles node collects the workload offered from the node and its neighbors to achieve the paths at which they belong to. The main steps followed by our algorithm to achieve the overlay tree are summarized in the following: (1) (2) Check if the new overlay node can be joined to the overlay with a link achieved through 1-hop path; Check if the new overlay node pk (i, j ) o(i, j ) EO vi = vj VO ; 6

In the rst step, DMMWA checks if the new node can be joined to the overlay with a link achieved through a 1-hop path, because the overlay nodes can transmit in multicast. So the growth of workload due to an additional node for a overlay node is zero. Instead, in the second step the algorithm checks if the new overlay node at step n, belongs to any path which achieves any overlay link at step n-1. In fact, if the new node belongs to any path to achieve any overlay link, the overlay link to join new node to the overlay already exists. Then we need only to refresh the new nodes entry in the vector of the workload per wireless node, to consider that new node can transmit in multicast. If the previous cases are not veried, the algorithm chooses the link solving the optimization problem. DMMWA in this case veries which nodes belong to the overlay. Through the routing protocol, DMMWA receives all the possible minimum paths which link the new node to each overlay node. DMMWA chooses the path that minimize the workload per node. When the path is chosen, the vector of workload per wireless nodes, which belong to the path, is refreshed. Finally, the DMMWA returns the overlay topology. We have developed also the centralized version of the algorithm named Minimum Max Workload Algorithm (MMWA), which works following the same DMMWAs steps. But in this case the information about the overlay and wireless network is collected into a centralized entity, so that the nodes do not need to exchange information about the overlay and wireless network. Hence, the main difference between the algorithms is the type of information available to join a new node to the overlay. In the case of DMMWA local information is avalaible, instead in the case of MMWA global information is available. VII. P ERFORMANCE E VALUATION MMWA and DMMWA were implemented on a custom simulator, and their performance was compared to an algorithm that connects new nodes to the overlay following a minimum hop scheme, i.e., new nodes are connected to the node in the overlay with minimum number of (wireless) hops. In the following we will refer at this algorithm as Minimum Hop Algorithm (MHpA). We vary the network size between 10 and 100 nodes with random wireless topology. The capacity of each node is selected randomly between 0.3 Mbit/s and 10 Mbit/s. We assume that for a given overlay we have only one multimedia content with required bit rate r = 0.5 Mbit/s. For DMMWA, we assume that the routing protocol chooses the routing path following a minimum-hop metric. We evaluate the fairness of MMWA and DMMWA, and MHpA, by calculating the Jains Fairness Index. Precisely, for given wireless network a set of overlay topologies and the corresponding vector of the workload per node is generated. The same sequence of new nodes for the three algorithms is

used. The results for different sizes of the wireless network are shown in Fig. 5.

Fig. 5.

Average Fairness vs. Network Size

MMWA and DMMWA have higher fairness index than MHpA, since they distribute the workload evenly. This guarantees an increased level of scalability for MMWA and DMMWA. In fact, minimizing the maximum workload for all the wireless nodes and for the nodes in a path with MMWA and DMMWA respectively, it is possible to bound the workload for the nodes. The performance of DMMWA is close to that of MMWA. Figure 6 shows that MMWA and DMMWA distribute the workload among the nodes better than MHpA. The average workload per node for a given size of the wireless network. In this case, for a given wireless network a set of overlay topologies is generated, and the corresponding vector of the workload per node. For each node the average workload is evaluated for all the overlay topologies. These average workload per node values are used to calculate the average workload per node, for given wireless network.

Figures 5 and 6 show that MMWA and DMMWA have the same performance. This happens because the simulation results with both algorithms DMMWA and MMWA, are obtained jointing all the wireless nodes to the overlay. Given that both algorithms follow the steps (1) and (2) explained in the section VI, using all the nodes in the wireless networks at the end they achieve the same overlay tree. Then to understand the performance of both algorithms we need to study the performance of MMWA and DMMWA for different sizes of the overlay. Precisely, for a wireless network with 30 nodes, we create an overlay with 7, 14, 21 and 30 nodes. For each different size, the average workload per node and the average Jains fairness index are then calculated. The numerical results are shown in Figs. 8 and 9. The performance of MMWA is better than the performance of DMMWA. In fact, MMWA can optimize the workload per wireless node, for each achieved overlay link, using more complete network information than DMMWA. However, DMMWA requires little information to obtain the overlay, and it is easy to implement it into an handheld wireless device.

Fig. 7.

Average Number of Failures vs Size Network

Fig. 6.

Average Workload vs Network Size

Figure 6 shows that with MMWA and DMMWA the nodes in the wireless network have lower workload that in the case of MHpA, since with MMWA and DMMWA the nodes can preserve their resources better than MHpA. We calculate reliability through the average number of nodes that can not be joined to the overlay. In Fig. 7 we show that MMWA and DMMWA are more reliable than MHpA. In fact, the average number of nodes that are unable to connect for MMWA and DMMWA is lower than for MHpA. This is a consequence of a better distribution of the workload with MMWA and DMMWA than MHpA. 7

Fig. 8.

Average Workload vs Overlay Size

VIII. C ONCLUSIONS An algorithm was proposed to create a wireless multimedia social network (MSN) to share video over an overlay topology. The algorithm is developed by following a crosslayer approach that takes into account characteristics of the overlay (application layer) and schedulability of ows at the medium access control (MAC) layer. The algorithm connects

Fig. 9.

Average Fairness vs Overlay Size

new users to the MSN through a link in the overlay topology that corresponds to a multi-hop path in the wireless network. To obtain a scalable network, the overlay topology is created with the objective to equalize the workload among network nodes. Centralized and distributed versions of the algorithm were presented. Through simulation results, we showed that the performance of the distributed version is close to the performance of the centralized version, and that the proposed algorithm distributes the workload per node more fairly than simpler algorithms. We are currently working to implement the developed protocol on an Iphone smart phone to test its performance in a realistic setting. R EFERENCES
[1] I.F. Akyildiz and T. Melodia and K.R. Chowdhury, Wireless Multimedia Sensor Networks: Applications and Testbeds, Proceedings of the IEEE, vol. 96, no. 10, pp. 15881605, October 2008. [2] S. Pudlewski and T. Melodia, Dmrc: Distortion-minimizing rate control for wireless multimedia sensor networks, in Proc. IEEE Intl. Conf. on Mobile Ad Hoc And Sensor Systems (MASS), Macau, China, 2009. [3] I.F. Akyildiz and T. Melodia and K.R. Chowdhury, A Survey on Wireless Multimedia Sensor Networks, (Elsevier) Computer Networks, vol. 51, no. 4, pp. 921960, March 2007. [4] A. Kansal and S. Nath and J. Liu and F. Zhao, SenseWeb: An Infrastructure for Shared Sensing, IEEE MultiMedia, vol. 14, no. 4, pp. 8 13, October 2007. [5] A. T. Campbell and N. D. Lane and E. Miluzzo and R. Peterson and H. Lu and X. Zheng and M. Musolesi and K. Fodor and S. B. Eisenman and G. S. Ahn , The Rise of People-Centric Sensing, IEEE Internet Computing, vol. 12, no. 4, pp. 12 21, July/August 2008. [6] E. Miluzzo, N. D. Lane, K. Fodor, R. Peterson, H. Lu, M. Musolesi, S. B. Eisenman, X. Zheng, and A. T. Campbell, Sensing meets mobile social networks: The design, implementation and evaluation of the cenceme application, in Proc. ACM Conf. on Embedded Networked Sensor Systems (SenSys), Raleigh, NC, USA, 2008. [7] S. Gaonkar, J. Li, R. R. Choudhury, L. Cox, and A. Schmidt, Microblog: Sharing and querying content through mobile phones and social participation, in Proc. ACM Int. Conf. on Mobile Service, Applications and Systems (MobiSys), Brekenridge, CO, USA, 2008. [8] X. Xin, S. Yuanchun, C. Zhijia, and Z. Baopeng, On constructing high performance overlay for layered streaming in heterogeneous networks, in Proc. IEEE Intl. Conf. on Advanced Information Networking and Applications (AINA Workshop ), Gino-wan City, Okinawa, Japan, 2008, pp. 578 584. [9] J. B. Kwon and H. Y. Yeom, Distributed multimedia streaming over peer-to-peer networks, in Proc. Intl. Conf. Euro-Par Parallel Processing, Klagenfurt, Austria, 2003, pp. 851 858. [10] X. Susu and L. Bo and G.Y. Keung and Z. Xinyan, Coolstreaming: Design, Theory, and Practice, IEEE Transactions on Multimedia, vol. 8, no. 8, pp. 1661 1671, December 2007.

[11] J. Peltotalo, J. Harju, M. Saukko, L. Vaatamoinen, I. Bouazizi, I. Curcio, and J. van Gassel, A real-time peer-to-peer streaming system for mobile networking environment, in Proc. IEEE INFOCOM (Workshop on Mobile Video Delivery), Rio de Janeiro, Brazil, 2009, pp. 1 7. [12] H. Radha, Advancements in cross-layer wireless multimedia communications, in Proc. IEEE Intl. Conf. on Communications, Computers and Applications Mosharaka(MIC-CCA), Amman, Jordan, 2008, pp. XV XVI. [13] D. Triantafyllopoulou, N. Passas, G. Lampropoulos, and A. Kaloxylos, Joint application and physical layer adaptation for improved performance in wireless networks, in Proc. IEEE Intl. Conf. on Wireless Pervasive Computing (ISWPC), Santorini, Greece, May 2008, pp. 256 260. [14] D. Qinghe and Z. Xi, A Cross-Layer Framework for Multi-Layer-Video Multicast with QoS Requirements in Multirate Wireless Networks, IEEE Communication Letters, vol. 13, no. 9, pp. 658 660, September 2009. [15] Z. Li and T. Herfet, MAC Layer Multicast Error Control for IPTV in Wireless LANs, IEEE Transaction on Broadcasting, vol. 17, no. 3, pp. 353 362, June 2009. [16] M. Wu and S. Makharia and H. L. Li and S. Mathur, IPTV Multicast Over Wireless LAN Using Merged Hybrid ARQ With Staggered Adaptive FEC, IEEE Transaction on Broadcasting, vol. 55, no. 2, pp. 363 374, June 2009. [17] X. Y. Li, Multicast Capacity of Wireless Ad Hoc Networks, IEEE/ACM Transaction On Networking, vol. 17, no. 3, pp. 950 961, June 2009. [18] T. Suzuki, T. Kutsuna, and S. Tasaka, Qoe estimation from mac-level 008 in audio-video transmission with ieee 802.11e edca, in Proc. IEEE PIMRC (Workshop), Cannes, France, 2008, pp. 16. [19] B. Wang, X. Wen, S. Yong, and Z. Wei, A new approach measuring users qoe in the iptv, in Proc. IEEE Conf. Pacic-Asia on Circuits,Communications and System (PACCS), Chengdu, China, May 2009, p. 453. [20] A. Detti, C. Loreti, and P. Loreti, Effectiveness of overlay multicasting in mobile ad-hoc network, in Proc. IEEE Intl. Conf. on Communications (ICC), Paris, France, 2004, pp. 3891 3895. [21] A. Marques, F. M. da Silva, and R. Rocha, P2p over mobile ad-hoc networks, in Proc. IEEE Sensor, Mesh and Ad Hoc Communications and Networks (SECON Workshops), Rome, Italy, 2009, pp. 1 3. [22] K. Ponmozhi and R. S. Rajesh, Applying p2p in manets for resource sharing, in Proc. IEEE Intl. Conf. on Control, Automation, Communication and Energy Conservation (INCACEC), Perdurai, Tamilnadu, India, 2009, pp. 1 5. [23] M. Bisignano, G. D. Modica, O. Tomarchio, and V. Lorenzo, P2p over manet: a comparison of cross-layer approaches, in Proc. IEEE Database and Expert Systems Applications (DEXA), Regensburg, Germany, 2007, pp. 814 818. [24] M. Ge, S. V. Krishnamurthy, and M. Faloutsos, Overlay multicasting for ad hoc networks, in Proc. IEEE Third Meditteranean Ad Hoc Networking Workshop (Med-Hoc-Net), Bodrum, Turkey, 2004. [25] H. Cho, S. H. Lee, Y. Choiand, F. Yu, and S. H. Kim, Efcient overlay multicast protocol in mobile ad hoc networks, in Proc. IEEE Vehicular Technology Conference (VTC), Dublin, Ireland, 2007, pp. 51 55. [26] N. M. Tahan, H. Rabiee, F. Saremi, and Z. Iranmanesh, An overlay multicast protocol for multimedia applications in mobile ad hoc networks, in Proc. IEEE Asia-Pacic Services Computing Conference (APSCC), Yilan, Taiwan, 2008, pp. 162 167. [27] K. Kwak, G. H. Canepa, Y. Ko, D. Lee, and S. J. Hyun, An overlaybased resource monitoring scheme for social applications in manet, in Proc. IEEE Intl. Conf. on Computer Software and Applications (COMPSAC), Seattle , Washington, USA, 2009, pp. 517 524. [28] T. Clausen and P. Jacquet, Optimized Link State Routing Protocol (OLSR). http://www.ietf.org/rfc/rfc3626.txt: RFC3626, October, 2003. [29] C. Perkins, E. Belding-Royer, and S. Das, Ad hoc On-Demand Distance Vector (AODV) Routing. http://www.faqs.org/rfcs/rfc3561.html: RFC3561, July, 2003. [30] M. Kodialam and T. Nandagopal, Characterizing Achievable Rates in Multi-hop Wireless Mesh Networks With Orthogonal Chnels, IEEE/ACM Transaction On Networking, vol. 13, no. 4, pp. 868 880, August 2005. [31] R. J. Trudeau, Introduction to Graph Theory. Dover, 1994.

You might also like