Sage Journals: Discover world-class research

Abstract

The peer-to-peer-based video resource dissemination is important for handling extreme conditions such as flash crowds which severely break the balance between supply and demand of video content and bring negative effects for quality of service (QoS). In this paper, we propose a novel interest detection-based video dissemination algorithm under flash crowd in mobile ad hoc networks (IDVD). IDVD classifies the user behavior of requesting video content in terms of the popularities and playback time of video and considers the demand of popular video as the cause of flash crowd. IDVD makes use of the prediction of necessary bandwidth and period time of intensive request to be aware of period time and scale of flash crowd. IDVD employs a resource dissemination algorithm which can fast discover interested nodes and achieve the on-demand dissemination of video resources in neighboring geographical area, in order to handle imbalance between supply and demand. IDVD uses the epidemic model to describe the state transition of nodes and define the convergence condition of resource dissemination algorithm. Extensive tests show how IDVD achieves much better performance results in comparison with other state of the art solutions.

1. Introduction

The mobile ad hoc networks (MANETs) rely on packets hopping between nodes which act as hosts and routers to achieve communication of nodes, so they are well known for flexible architectures and rapid deployment [1, 2]. The multimedia streaming services are more popular applications in Internet and mobile Internet [3–10]. The deployment of video services in MANETs supports ubiquitous access for the mobile users, enhancing user experience [11, 12]. P2P technologies provide the solution for large-scale video sharing in wireless networks such as MANETs and wireless sensor networks (WSNs), so that the mobile users conveniently fetch desired video content from the peers [13–17]. The video with wonderful content always attracts mass peers to seek and download the resources from other peers or the media server. However, the relatively limited network bandwidth and capacities of mobile nodes cannot afford traffic requirement caused by intensive resource request, reducing quality of service (QoS). Therefore, a solution based on green communication requirement, which can understand behavior of pursuing popular video content, fast discover the nodes of cooperatively caching resources, and efficiently spread video content should be considered for video streaming system in MANETs.

The flash crowds lead to serious imbalance between supply and demand for video streaming resources and bring negative influence in the scalability and QoS of video streaming systems. For instance, the blowout of video resource request results in the long delay of response and network congestion. The mass researchers recently focus on the study of flash crowd in order to improve QoS of video streaming system and reduce the cost of system deployment [18–20]. A fluid-based model for P2P live streaming systems was proposed in [18], which describes the relationship between the system capacity, peer startup latency, and system recovery time with and without admission control. A mathematical model proposed in [19] captures the relationship between time and scale in P2P live streaming systems and designs a principle of scale control. A model proposed in [20] predicts the scalability of system with increasing number of nodes and provides enough upload bandwidth for sudden resource request according to the estimation of server bandwidth, so as to ensure high QoS. The aforementioned solutions focus on modeling the living streaming systems to address the influence of flash crowd and are unsuitable for the mobile environment with limited network resources and high mobility of mobile nodes. The high-efficiency video resource sharing in wireless networks also is important solution for handling flash crowd. The management of video resources regulates distribution of resources in terms of dynamic user demand, which fast perceives and responds to the variation of video resource demand. The video resource dissemination strategies can fast spread requested resources in overlay networks, which reduce recovery time of balance between supply and demand. Therefore, an efficient solution based on dynamic balance between supply and demand, which supports fast video resource dissemination and efficiently prevents the degraded QoS should be considered for P2P-based video streaming systems over wireless mobile networks.

In this paper, we propose a novel interest detection-based video dissemination algorithm under flash crowd in mobile ad hoc networks (IDVD). By the analysis of resource request behavior of users and the estimation of the popularities of video resources, IDVD constructs a novel “H”-model to explain that the intensive request of popular video content is the main cause of the flash crowd in multimedia streaming system. IDVD makes use of the historical information (e.g., bandwidth and period time) of large-scale request for popular video content to predict the period time and scale of flash crowd. IDVD designs a novel resource dissemination algorithm which can fast spread resources to meet the demand of upload bandwidth. In order to reduce the cost of resource dissemination, the resource carriers only search and spread the interested mobile nodes in neighboring geographical area and dynamically regulate range of dissemination in terms of predicted necessary bandwidth and capacities of carriers. IDVD uses the epidemic model to describe the dissemination process of resources and state transition of nodes and define convergence condition of spreading process. Simulation results show how IDVD achieves much better performance results in comparison with other state of the art solutions.

2. Related Work

The video system models under flash crowd have attracted increasing research interests from various researchers. For instance, a fluid-based model for P2P live streaming systems was proposed in [18], which studies the relationship between capacity and recovery time of system and peer startup latency with and without admission control for flash crowds. In the systems without admission control, this paper finds that there is an independent relationship between the capacity and initial state of system while power law decreases with the departure rate of stable peers. The paper also shows that the admission control can help the system relieve the large flash crowds in the systems with admission control and proposed the flash crowd handling strategies in order to satisfy the peer startup performance under various circumstances. The mathematical framework in [19] researches the inherent relationship between time and scale of P2P live streaming system during a flash crowd. The population control procedure improves the system scale by trading peer startup delays. This paper also analyzed the effects of partial knowledge of peers and the competition of limited upload bandwidth resources between peers. Moreover, an analytical model in [20] for flash crowds is based on the evolution of the utilization of available bandwidth at peer side in order to investigate impact of the utilization of available bandwidth. The model can predict the system scalability with increasing number of nodes and provide necessary bandwidth for sudden request.

Recently, some researchers focus on the resource management strategies in order to optimize resource distribution and make the balance between supply and demand. Kozat et al. proposed a hybrid P2P video-on-demand architecture, which improves transmission efficiency of popular videos [21]. In this architecture, each system member caches a video chunk and makes use of surplus upload bandwidth to serve other nodes. The server schedules the video resources in the system to respond to the request of nodes and provide reliable streaming service. In order to balance the load between server and system members, the architecture considered the caching problem as a utility optimization problem based on supply and demand and used the multiple caching mechanisms to optimize the performance of system. PECAN in [22] proposed a peer cache adaptation strategy, in which each peer dynamic regulates the local storage capacity to improve the scalability of system. PECAN employs a cache replacement algorithm to improve the resource distribution in terms of the popularities of video chunk and show that the storage capacities of peers are corresponding to the request rate of resources. PECAN designed a distributed reputation and monitoring system to discover selfish peers.

Moreover, some file (video) resource dissemination algorithms recently are proposed. For instance, Mokhtarian and Hefeeda show that the problem of allocating the seed server capacity is NP-complete and proposed a seeding capacity allocation algorithm to address the optimal allocation problem [23]. The paper proposed an analytical model to predict the performance of P2P-based video system by using an allocation algorithm, which estimates the long-term network throughput according to video quality and total served bitrate. Venkatramanan and Kumar analyzed the evolution process of the interest in the content under the linear threshold model and made use of an epidemic spread model to control the content copying process [24]. This paper modeled the coevolution process of popularity and delivery of video content according to homogeneous influence linear threshold model. This paper used fluid limit ordinary differential equations to provide the selection of parameters for the control of content suppliers and address optimization problems for content delivery. Altman et al. proposed an extensional epidemic model to characterize file sharing behavior in P2P networks including free-riding peers [25]. This paper modeled P2P network dynamics by a Markov chain, where the state of P2P system evolves from branching process to a supercritical P2P swarm with increasing network size. The paper shows that there are the phase transitions; the small change of parameters causes a large change in the network behavior for two models of epidemic and branching.

3. IDVD Detailed Design

3.1. Media Server

The video resources are stored at the media server in order to provide original video data for all mobile nodes in MANETs; namely, a video file set is defined as $S_{V} = (f_{1}, f_{2}, \dots, f_{n})$ . When the server receives a request message, it assigns a candidate supplier which caches the requested file for the requester and adds the information of requester into local node set $S_{N} = {(n_{1}, f_{a}, t_{1}), (n_{2}, f_{b}, t_{2}), \dots, (n_{m}, f_{h}, t_{m})}$ where f denotes the requested video file and t is the timestamp of joining system of nodes. The nodes also contact other system members to obtain desired resources, which is detailed next. A system member which leaves system sends a quit message containing the information of played files and corresponding playback start-stop timestamp (local system time of node) to the server. The server considers this information as playback logs and analyzes these logs to be aware of the condition of request and playback of resources such as the file popularity and the mean playback time of video files. The popularities of items in $S_{V}$ follow a Zipf distribution [26, 27] and the popularity value of each file $f_{i}$ can be obtained according to $\begin{matrix} {pop}_{i} = \frac{γ_{i}}{\sum_{c = 1}^{n} ‍ γ_{c}}, \end{matrix}$ (1)where $γ_{i}$ is the access frequency of $f_{i}$ and $\sum_{c = 1}^{n} ‍ γ_{c}$ is the total access frequency of all files. The mean playback time ratio of video file $f_{i}$ is defined as $\begin{matrix} {\bar{l}}_{i} = \frac{\sum_{c = 1}^{N_{u}} ‍ l_{c}}{N_{u} \times L_{i}}, \end{matrix}$ (2)where $l_{c}$ is playback time of cth user, $L_{i}$ is the length of video $f_{i}$ , and $N_{u}$ is the total number of users which have watched $f_{i}$ . We use the mean playback time ratio of $f_{i}$ as weight value of popularities, so the weighted popularities of $f_{i}$ can be defined as ${pop}_{i}^{w} = {\bar{l}}_{i} \times {pop}_{i}$ . The files whose weighted popularities are larger/less than $(1 / n) \sum_{c = 1}^{n} ‍ {pop}_{c}^{w}$ (average value of weighted popularities of n files) are popular/unpopular. In terms of the fetched content, the user playback behavior can be classified as $(1)$ the users which are watching popular files request popular files; $(2)$ the users which are watching unpopular files request popular files; $(3)$ the users which are watching unpopular files request unpopular files; $(4)$ the users which are watching popular files request unpopular files. As Figure 1 shows, the popular/unpopular files and channels between them form an “H”-model”. The first and second kind behaviors generate the traffic to the popular files, namely, the channel $C_{p}$ ; the third and fourth kind behaviors form the traffic of channel $C_{u}$ . The popular content always can attract more users, namely, the traffic value in $C_{p}$ is far greater than that in $C_{u}$ . The less active system members in $C_{u}$ do not require mass upload bandwidth, but this leads to the shortage of available resources. For instance, if there are less nodes which cached these files and have the asynchronous playback point with the requesters in overlay networks, this leads to fragile logical link between suppliers and requesters due to dynamic resource demand. The requesters need to fetch desired resources from the server. If there are no requested resources in overlay networks, the server needs to provide the initial streaming data for the requesters. Therefore, the shortage of available resources causes too much intervention of server for scheduling resources, which increases the load of server. We make use of distributed hash table (DHT) to group these system members in $C_{u}$ [28] and optimize the resource distribution according to cooperative cache strategies [29, 30], which can improve the efficiency of sharing resources and reduce the load of server.

Figure 1

“H” model.

The demand from the large number of nodes for popular video is one of the main causes of flash crowd in $C_{p}$ . The transient large-scale request breaks the balance between supply and demand for upload bandwidth, which results in high serving delay and the overload of server. The server needs to be aware of the bandwidth requirement level and period time of intensive request in advance. The huge traffic in $C_{p}$ usually is the main reason for the system bottleneck, so we introduce the traffic prediction scheme for any popular video $f_{i}$ in $C_{p}$ . The length of $f_{i}$ is defined as a period time. We use an average value of traffic during each period time to define the channel traffic set; namely, $S_{f_{i}} = ({\bar{b}}_{t_{1}}, {\bar{b}}_{t_{2}}, \dots, {\bar{b}}_{t_{m}})$ where any average traffic value can be obtained by $\begin{matrix} {\bar{b}}_{t_{c}} = \frac{\sum_{j = 1}^{u} ‍ b_{j}}{t_{c}}, \end{matrix}$ (3)where u is the number of request bandwidth values in $C_{p}$ during $t_{c}$ . $S_{f_{i}}$ is considered as the original series $S_{f_{i}}^{(O)}$ and is processed to an accumulated series $S_{f_{i}}^{(A)} = (b_{t_{1}}^{(A)}, b_{t_{2}}^{(A)}, \dots, b_{t_{m}}^{(A)})$ , according to $\begin{matrix} b_{t_{k}}^{(A)} = \{\sum_{c = 1}^{k} ‍ {\bar{b}}_{t_{c}} | k = 1,2, \dots, m\} . \end{matrix}$ (4)

Because the random traffic values are difficult to form a stable predictable variation trend, we make use of the Grey Forecast Model $GM (b^{(A)})$ [31, 32]; the first-order differential equation of $GM (b^{(A)})$ is defined as $\begin{matrix} \frac{d b^{(1)}}{d t} + r b^{(1)} = h, \end{matrix}$ (5)where t is the time series variable and b is the series variable of traffic accumulation with increasing time interval. r and h denote the grey level of development and control, respectively. We use the ordinary least square method to solve the values of r and h according to $\begin{matrix} \hat{U} = [\begin{bmatrix} \hat{r} \\ \hat{h} \end{bmatrix}] = {(D^{T} D)}^{- 1} D^{T} y, D = [\begin{bmatrix} \frac{1}{2} [b_{t_{2}}^{(A)} + b_{t_{1}}^{(A)}] 1 \\ \dots \\ \frac{1}{2} [b_{t_{m}}^{(A)} + b_{t_{m - 1}}^{(A)}] 1 \end{bmatrix}], \end{matrix}$ (6)where $\hat{r}$ and $\hat{h}$ are the solutions of r and h, respectively. $D^{T}$ is the transposition matrix of D and $y = ({\bar{b}}_{t_{2}}, {\bar{b}}_{t_{3}}, \dots, {\bar{b}}_{t_{m}})^{T}$ . Equation (7) denotes the solution of (5) by using $\hat{r}$ and $\hat{h}$ : $\begin{matrix} \hat{b} (k + 1) = ({\bar{b}}_{t_{1}} - \frac{\hat{h}}{\hat{r}}) e^{- \hat{r} k} + \frac{\hat{h}}{\hat{r}} . \end{matrix}$ (7)

$\hat{b} (k + 1)$ , $k \leq m$ , is the fitting value; $\hat{b} (k + 1)$ , $k > m$ , is the forecast value; namely, we compare $\hat{b} (k + 1)$ , $k > m$ , with the upper limit of server load to make a decision whether to spread $f_{i}$ in networks. We use a posteriori variance ratio $R^{(P)}$ and an occurrence probability $P^{(P)}$ to ensure the confidence level of prediction values. $R^{(P)}$ and $P^{(P)}$ are defined as $\begin{matrix} R^{(P)} = \frac{F}{σ}, P^{(P)} = P \{|C (k) - \bar{C}| < 0.6745 σ\}, \end{matrix}$ (8)where $C (k)$ is the residual value; namely, $C (k) = {\bar{b}}_{t_{k}} - \hat{b} (k)$ , $k = 2,3, \dots, m$ . $\bar{C}$ and F are the residual mean value and variance, respectively, according to $\begin{matrix} \bar{C} = \frac{\sum_{c = 2}^{m} ‍ C (c)}{m - 1}, F = \sqrt{\frac{\sum_{c = 1}^{m} ‍ {(C (c) - \bar{C})}^{2}}{m - 1}} . \end{matrix}$ (9)

σ and $b_{mean}$ are the variance and mean value of items in $S_{f_{i}}^{(O)}$ according to $\begin{matrix} σ = \sqrt{\frac{\sum_{c = 1}^{m} ‍ {({\bar{b}}_{t_{c}} - b_{mean})}^{2}}{m}}, b_{mean} = \frac{\sum_{c = 1}^{m} ‍ {\bar{b}}_{t_{c}}}{m} . \end{matrix}$ (10)

$R^{(P)}$ and $P^{(P)}$ can reflect the confidence level of prediction values of channel traffic, so we make use of two threshold values $T^{(R)}$ and $T^{(P)}$ in terms of the Grey Forecast Model to measure the confidence level. If $R^{(P)} \geq T^{(R)}$ and $P^{(P)} \geq T^{(P)}$ , the prediction value is credible. When the prediction value $\hat{b} (u)$ , $u > m - 1$ , is credible and is larger than the summation of upload bandwidth known by the server (total bandwidth of the server and items in $S_{N}$ ), the server sends a message containing $\hat{b} (u)$ to the nodes which store $f_{i}$ in $S_{N}$ and are considered as the carriers of $f_{i}$ . These carriers return a statistical information about available upload bandwidth to the server by collecting the upload bandwidth from the nodes which are downloading $f_{i}$ from the carriers. If $\hat{b} (u)$ still is larger than the summation of upload bandwidth, the server requires these nodes to fast disseminate $f_{i}$ in order to cope with intensive request.

3.2. Resource Disseminate Model

In order to reduce the load of server, the server only sends a message containing necessary upload bandwidth $\hat{b} (u)$ and a carrier list to each carrier. The carriers are responsible for controlling the dissemination process by the message exchange. We use a token-based message exchange strategy to achieve the synchronization of information between carriers. Each item in the carrier list has a random number and successively sends the message to other carriers according to the value of number. For instance, the number of items in the list is k. The $k - 1$ carriers send collected information (e.g., the number of discovered nodes which are interested in the resources) to kth carrier $n_{s}$ which has the token. After $n_{s}$ handles these received messages, it returns a message containing the calculation results to other carriers. Meanwhile, $n_{s}$ turns the token over to $k - 1$ th carrier. After the carrier with the smallest number return message to other carriers, it returns the token over to $n_{s}$ . The token-based exchange strategy can balance the load between carriers and does not cause high message overhead.

When these carriers receive the request of spreading $f_{i}$ from the server, they start to discover interested nodes (INs) and require INs caching $f_{i}$ . The INs include two types of nodes: the interested mobile nodes (IMNs) and the interested system members (IMs) which are playing other videos. The carriers also are considered as the inquirers due to searching the INs. In order to reduce the cost of spreading $f_{i}$ , we employ a guidance-based dissemination strategy to implement geographic region-based file diffusion.

Each inquirer $n_{j}$ makes use of cross-layer method to add the information of viewing file (current playback state) to one-hop multicast message at the MAC layer. If there are the system members in the one-hop neighbor nodes of $n_{j}$ , these members return the information of current played video file. Moreover, if the one-hop neighbor nodes are interested in $f_{i}$ , they also add an interest mark into the return messages. When $n_{j}$ has exchanged messages with the one-hop nodes, it records the information of played content of one-hop nodes and stores the information of INs. We set a variable period time $U T$ for the above neighbor node discovery process according to our previous work in [33]. The nodes dynamically change their own $U T$ in terms of the variation level of mobility of one-hop nodes.

$n_{j}$ needs to select an IMs $n_{p}$ from an encountered node list $L_{j}$ during recent p update period as a cooperative inquirer. The list is defined as $L_{j} = {(n_{1}, t_{1}, f_{a}), (n_{2}, t_{2}, f_{b}), \dots, (n_{u}, t_{u}, f_{h})}$ where t is the encountered timestamp and f denotes the viewing file in the process of encounter. $n_{p}$ has the nearest encountered timestamp and does not cache $f_{i}$ . $n_{j}$ requires that $n_{p}$ continues to search the INs from its one-hop nodes and select next cooperative inquirer from $L_{p}$ . $n_{j}$ sends a message containing the condition of convergence of iterative search to $n_{p}$ (for instance, the number of iteration is defined as q times). After $n_{p}$ has inquired for its one-hop nodes, $n_{p}$ delivers the collected information of the INs to the selected next cooperative inquirer. After the qth cooperative inquirer has inquired, it directly returns the collected information of INs to $n_{j}$ . Because the mobile nodes may be inquired by multiple inquirers or cooperative inquirers, we define the following rule in order to ensure the accuracy of statistical information in the process of inquiry. $(1)$ If a node has sent the interest mark to an inquirer or a cooperative inquirer, it only return the message containing uninterested mark after the reception of inquiry message of other inquirers or cooperative inquirers; $(2)$ a node cannot become the cooperative inquirer of two carriers in the same process of inquiry; $(3)$ the carriers cannot become the cooperative inquirers of other carriers.

After $n_{j}$ receives and stores the collected information of INs from the qth cooperative inquirer, it keeps the contact with discovered INs and forwards the messages to the carrier $n_{t}$ which has the token. When $n_{t}$ receives the information of discovered INs from all carriers, it estimates needed number of INs. We make use of the epidemic model (SIR) to calculate needed number of INs. All parameters of the epidemic model are listed in the Notations. The necessary and sufficient condition of implementing SIR model is defined as $N_{t} > N_{I}$ , $N_{I} = \hat{b} (u) / b_{p}$ where $N_{I}$ denotes needed number of nodes which store $f_{i}$ and $b_{p}$ is the transmission rate of $f_{i}$ . If $N_{t} < N_{I}$ , $n_{t}$ requires that the carriers continue to search more INs. Once the carriers find the IMs, they transmit the data of $f_{i}$ to the IMs; namely, the IMs can be immediately infected. However, the IMNs are considered as the potential infected nodes so that the carriers do not require them to cache $f_{i}$ at once. The dynamic number of carriers leads to the change of available upload bandwidth. For instance, the carriers which have watched the whole video content remove $f_{i}$ from local buffer. When the inquirers do not discover more IMs to meet the demand of bandwidth due to the limited detection range, they require that the IMNs cache $f_{i}$ . The IN discovery rate $D R_{j}$ of any inquirer $n_{j}$ is defined as $\begin{matrix} D R_{j} = \frac{N_{I N_{j}}}{N_{t}}, \end{matrix}$ (11)where $N_{I N_{j}}$ denotes the number of INs discovered by $n_{j}$ . λ is defined as $λ = N_{IM} / N_{t}$ where $N_{IM}$ is the number of discovered IMs. The members which have watched $f_{i}$ are usually uninterested to view $f_{i}$ again; namely, their state becomes uninterested. Therefore, the value of μ is set to 1. We use a differential equation to denote the SIR model as follows: $\begin{matrix} \frac{d I}{d t} = λ S I - μ I, \\ \frac{d S}{d t} = - λ S I, \\ I (t_{0}) = I_{0}, S (t_{0}) = S_{0} . \end{matrix}$ (12)

We obtain the solution of the above differential equation; namely, $I = S_{0} + I_{0} - S + (μ / λ) \ln (S / S_{0})$ . Further, we can obtain the solution of S according to known value of $S_{0}$ , $I_{0}$ , I, μ, and λ, namely, $\hat{S} = W (- λ e^{λ I - λ S_{0} - λ I_{0} - \ln S_{0}}) / - λ$ where W is the Lambert W Function and $I = N_{I} / N_{t}$ . $\hat{S}$ is the needed number of INs in order to ensure the required scale of nodes cached $f_{i}$ based on current spreading rate. $n_{t}$ reassigns the needed number of INs for the carriers according to the collected number of inquired nodes of carriers during an inquiry period and requires that the carriers continue to find new INs. The more the number of inquired nodes of a carrier is, the higher the probability of discovering INs is. Therefore, the carriers which have more inquired nodes should be assigned more number of IN discovery. The above process is considered as an IN discovery period. $n_{t}$ turns the token to next carrier after a discovery period. If the total number of discovered INs is equal to or greater than $\hat{S}$ , the carrier which has the token requires that the carriers keep the state of equalisation during predicted period time $t_{u}$ . After the system went through $t_{u}$ , the carriers disconnect the contact with all INs.

4. Testing and Test Results Analysis

We investigate the performance of the proposed IDVD in comparison with HILT-SI model [24]. We chose a 100-second long video clip $f_{i}$ which is considered as copied content. IDVD was modeled and implemented in NS-2, as described in the previous sections.

4.1. Testing Topology and Scenarios

Table 1 lists some NS-2 simulation parameters of the MANET for the two solutions. We define initial random speed and target location of movement for all mobile nodes. After the mobile nodes arrive at the target location, they continue to move according to newly assigned speed and target location of movement. All mobile nodes follow the above iteration of moving behavior during the whole simulation time. The default distance between server and nodes is set to 6 hops in order to ensure the consistency of cost of accessing to the server for all mobile nodes. The variation of default distance can influence the cost of fetching video content from the server. For instance, the increase/decrease of default distance brings high/low transmission delay of video data and packet loss rate. Initially, there are 200 system members where 20 members play $f_{i}$ and 80 members are uninterested in $f_{i}$ . 50 members which are viewing another video want to watch $f_{i}$ and 50 members request $f_{i}$ per 0.5 s from 80 s to 105 s. Moreover, 50 mobile nodes are interested in $f_{i}$ . The INs cache and play $f_{i}$ after they are discovered (influenced). The members which have watched $f_{i}$ quit the system and remove $f_{i}$ from local buffer. The uninterested nodes do not cache $f_{i}$ during the whole simulation time. In HILT-SI, each IN is independently assigned a random infected threshold θ, $0 < θ < 1$ and the values of Γ and α are set to 0.9 and 0.3, respectively.

Table 1

Simulation parameter setting for MANET.

Parameters	Values
Area	1000 × 1000 m²
Channel	Channel/WirelessChannel
Network interface	Phy/WirelessPhyExt
MAC interface	Mac/802_11
Number of mobile nodes	400
Mobile speed range of nodes	[1, 30] m/s
Simulation time	200 s
Signal range of mobile nodes	200 m
Default distance between server and nodes	6 hops
Transmission protocol	UDP
Wireless routing protocol	DSR
Interface queue	CMUPriQueue
Bandwidth of server	20 Mb/s
Bandwidth of mobile nodes	10 Mb/s
Transmission rate of video data	128 kb/s
Travel direction of mobile nodes	random
Pause time of mobile nodes	0 s
p	2
q	3

4.2. Performance Evaluation

The performance of IDVD is compared with that of HILT-SI in terms of capacity of IN discovery and content spreading, message overhead, average data transmission delay, packet loss rate (PLR), and throughput, respectively.

Capacity of IN Discovery and Content Spreading. The number of discovered INs (the interested members and mobile nodes) and carriers (the nodes carry $f_{i}$ ) denotes the capacities of content dissemination for two solutions.

As Figure 2 shows, HILT-SI's blue curve has slow rise from $t = 100 s$ to $t = 200 s$ after fast increase from $t = 0 s$ to $t = 80 s$ , after IDVD's red curve also has a slow increase from $t = 80 s$ to $t = 200 s$ after fast rise from $t = 0 s$ to $t = 60 s$ . The increment of HILT-SI curve is larger than that of IDVD and HILT-SI has a longer increase period time than that of IDVD; namely, HILT-SI nearly searches all interested nodes. Figure 3 presents the variation of number of carriers in the video system with increasing simulation time. The blue curve corresponding to HILT-SI's results experiences a fast decrease from $t = 120 s$ to $t = 200 s$ after it suddenly reaches the peak value 143 at $t = 100 s$ . IDVD's red curve also has similar trend; namely, it has a slow rise, suddenly arrives at the peak value 116 at $t = 100 s$ , and fast falls from $t = 120 s$ to $t = 200 s$ . HILT-SI's results are both higher values and larger fluctuation than those of IDVD during whole simulation time.

Figure 2

The number of discovered INs against simulation time.

Figure 3

The number of carriers against simulation time.

In HILT-SI, the carriers continually influence the nodes connected with them according to the given threshold. When the nodes do not become interested nodes, the influence values of carriers increase by the accumulation so that the state of influenced nodes finally becomes interested and these new interested nodes help the carriers influence other nodes by making use of its own influence value. Because the server periodically broadcasts the state of all nodes in the whole network, all INs and carriers try to influence other potential INs. The efficiency of IN discovery HILT-SI is higher than that of IDVD. When the potential INs become new INs, they cache and play $f_{i}$ ; namely, these new INs immediately become new carriers after they fetch $f_{i}$ . In HILT-SI, the number of carriers has a slow increase from $t = 20 s$ to $t = 80 s$ . 50 members suddenly join the system and request $f_{i}$ from $t = 80 s$ to $t = 105 s$ , so the number of carriers fast reaches the peak value. With increasing simulation time, the initial carriers have watched whole video content and quit the system, so the number of carriers fast decreases from $t = 120 s$ to $t = 200 s$ . HILT-SI does not include the mechanism of controlling the range of IN discovery and spreading resources, so it can fast disseminate the video content to the whole network. In IDVD, the carriers make use of MAC multicast to invite the mobile nodes joining INs and inquire recent encountered system members. IDVD uses p and q to control detection range; namely, the carriers search INs in their neighbor geographic area. IDVD's capacity of IN discovery is limited by local search strategy, but IDVD can regulate the speed of discovery by changing the values of p and q and dynamically assign the number of INs discovered in terms of the capacities of carriers. Moreover, IDVD predicts the needed bandwidth in the future to control the process of discovery, so that the number of INs slowly increases in terms of the needed bandwidth from $t = 80 s$ to $t = 200 s$ . The variation of number of carriers is based on the number of INs, so the increment and decrement of carriers are less than those of HILT-SI.

Message Overhead. The average bandwidth which is used by the sent messages for the IN discovery is considered as the message overhead.

As Figure 4 shows, the message cost values of two systems have similar changing trend with increasing simulation time. The curve corresponding to HILT-SI's results fast decreases from $t = 120 s$ to $t = 200 s$ after fast increases from $t = 20 s$ to $t = 100 s$ , where the peak value is 4.05 kb/s at $t = 120 s$ . The curve corresponding to IDVD has a slow rise trend from $t = 20 s$ to $t = 100 s$ , reaches the peak value 2.92 kb/s at $t = 140 s$ , and slightly decreases from $t = 160 s$ to $t = 200 s$ . The CSPV results are roughly 30% lower than those of HILT-SI, with increasing simulation time.

Figure 4

Message cost against simulation time.

In HILT-SI, the server periodically broadcasts the state of all nodes in network. When the INs and carriers receive the state information of nodes, they update the state of nodes and continue to influence other potential INs. In order to obtain the capacity of fast IN discovery, the server needs to frequently interact with the INs and carriers. HILT-SI needs to consume the large number of network bandwidth to maintain the process of fast discovery. In IDVD, the carriers only inquire the small number of system members and detect the mobile nodes in one-hop range. Moreover, the token-based message exchange strategy also reduces the message exchange between carriers. Therefore, IDVD's message cost can maintain lower level than that of HILT-SI. By regulating the values of p and q to change the range of IN discovery, IDVD can control the range of resource dissemination and adapt dynamic network environment.

Average Data Transmission Delay. We calculate the transmission delay of received video data at application layer during each time slice according to $\begin{matrix} \bar{d} = \frac{\sum_{c = 1}^{k} ‍ d_{c}}{t_{s}}, \end{matrix}$ (13)where $t_{s}$ is the time slice, k is the number of all received data during a time slice, $d_{c}$ denotes the delay of received cth data, and $\sum_{c = 1}^{k} ‍ d_{c}$ is the sum of delay of all received data during a time slice. In terms of the settings of simulation time and the defined strategies of requesting resources, the value of $t_{s}$ is set to 20 s.

As Figure 5 shows, HILT-SI's blue curve experiences a slight fluctuation from $t = 20 s$ to $t = 80 s$ and fast decreases from $t = 140 s$ to $t = 200 s$ after suddenly reaching the peak value 3.3 s at $t = 120 s$ . The red curve corresponding to IDVD's results fast reaches the peak value 2.89 s at $t = 120 s$ after having a slow increase from $t = 20 s$ to $t = 80 s$ and it falls from $t = 140 s$ to $t = 200 s$ . IDVD's delay is roughly 20% better than the values associated with HILT-SI.

Figure 5

Average data transmission delay against simulation time.

In HILT-SI, the carriers and INs fetch the information of nodes from the broadcast messages. They make use of logical connection with the INs to push the video content. HILT-SI does not consider the geographical location relationship between carriers and INs, so that the average data transmission delay maintains higher level than that of IDVD. Moreover, when the large number of request suddenly arrives, the nodes do not assume huge network traffic so as to result in the network congestion from $t = 100 s$ to $t = 140 s$ . Therefore, HILT-SI's delay is higher than that of IDVD. In IDVD, the members and mobile nodes are aware of resource information by receiving push messages of inquirers and cooperative inquirers. They fetch the video content from neighbor carriers. The local resource dissemination and the small number of video streaming relative to HILT-SI (the number of INs in IDVD is less than that of HILT-SI) do not consume more bandwidth of other relay nodes. Therefore, IDVD's peak value is less than that of HILT-SI and the time of duration of network congestion is shorter than that of HILT-SI.

Packet Loss Rate (PLR). The ratio between the number of packets lost in the process of video data transmission and the total number of packets of video data sent is defined as PLR.

As Figure 6 shows, the curves corresponding to HILT-SI and IDVD show a fall after rise with increasing simulation time. The results of HILT-SI and IDVD maintain low levels from $t = 20 s$ to $t = 80 s$ and fast increase from $t = 100 s$ to $t = 120 s$ and reach the peak values, respectively. The PLR values of HILT-SI and IDVD fast decrease from $t = 140 s$ to $t = 200 s$ . IDVD's PLR values are roughly 15% lower than those of HILT-SI.

Figure 6

PLR against simulation time.

The small number of system members fetching the video content only consumes less bandwidth, so the PLR values of HILT-SI and IDVD show slow increase from $t = 20 s$ to $t = 80 s$ . With sudden arrival of mass resource request, the high requirement of network bandwidth introduces the network congestion, so that HILT-SI and IDVD have high PLR from $t = 100 s$ to $t = 120 s$ . When the carriers constantly quit the system, the decreasing network traffic alleviates the congestion level. The PLR values of HILT-SI and IDVD fast decrease from $t = 140 s$ to $t = 200 s$ . In HILT-SI, the video data transmission relies on the logical link between carriers and INs; namely, the geographical distance of the communicating parties cannot be considered. The long-distance delivery of video data consumes the large number of bandwidth of relay nodes. Moreover, the more number of INs requires much network bandwidth. In IDVD, the carriers disseminate the message containing the information of video resources in neighbor geographical area. When the INs receive the resource information, they can download the video content from their neighbor carriers. The data only is forwarded by less relay nodes. Moreover, the small number of INs fetching video content cannot consume more network bandwidth. The congestion level of IDVD is lower than that of HILT-SI, so the PLR values of IDVD are less than those of HILT-SI.

Average Throughput. The total number of packets received in the overlay during a certain time period divided by the length of this time period is defined as the average throughput.

As Figure 7 shows, the average throughput curve of HILT-SI experiences severe fluctuation, which fast increases from $t = 20 s$ to $t = 80 s$ , immediately decreases from $t = 80 s$ to $t = 140 s$ , and finally has a fall after rise from $t = 160 s$ to $t = 200 s$ . The curve corresponding to IDVD results fast increases from $t = 20 s$ to $t = 100 s$ , reaches the peak value at $t = 100 s$ , and slowly decreases from $t = 120 s$ to $t = 200 s$ .

Figure 7

Throughput against simulation time.

The more number of INs introduces the transmission of much video streaming data in network. The throughput of HILT-SI fast increases from $t = 20 s$ to $t = 80 s$ . When the intensively arrival of mass resource request leads to the network congestion, the throughput of HILT-SI fast decreases due to the increase in PLR. With the increase in the number of carriers leaving the system, the decreasing congestion level enables the throughput rise. When the large number of carriers quits the system, the throughput fast falls. The more number of streaming and long-distance delivery result in high congestion level, so the throughput of HILT-SI severely jitters. In IDVD, the small number of data transmission requirement and the neighbor distance between nodes only introduce low-level congestion. When the congestion occurs at $t = 100 s$ , the throughput of IDVD reaches the peak value. The congestion influence of IDVD is lighter than that of HILT-SI.

5. Conclusion

In this paper, we propose a novel interest detection-based video dissemination algorithm under flash crowd in mobile ad hoc networks (IDVD). IDVD prevents the degradation of QoS and network congestion caused by large-scale sudden request for popular video content. IDVD constructs an “H” model to build the categories of user request according to the popularities of video content and predict the amount demanded of upload bandwidth and period time of sudden request. The proposed resource dissemination algorithm formulates the area coverage of interested node discovery and resource dissemination and defines the convergence condition of spreading resources according to the epidemic model. The results show how IDVD obtains better performance than HILT-SI.

Footnotes

Notations

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

This work was supported in part by the National Natural Science Foundation of China (NSFC) under Grant no. 61372112;the Beijing Natural Science Foundation (4142037);the key project of science and technology of Henan province (142102210473);the major projects of science and technology research of education department of Henan province (14B520016).

References

Karia

D. C.

Godbole

V. V.

New approach for routing in mobile ad-hoc networks based on ant colony optimisation with global positioning system

IET Networks 2013 2 3 171 180

10.1049/iet-net.2012.0087

Zhou

R. Q.

Qian

Chen

H.-H.

Energy-spectrum efficiency tradeoff for video streaming over mobile ad hoc networks

IEEE Journal on Selected Areas in Communications 2013 31 5 981 991

10.1109/jsac.2013.130516

2-s2.0-84876761018

Zhou

T. Z. J.

Chiu

D. M.

On replication algorithm in P2P VoD

IEEE/ACM Transactions on Networking 2013 21 1 233 243

10.1109/tnet.2012.2196444

2-s2.0-84873991233

Zhong

Zhang

Muntean

G.-M.

CMT-NC: improving the concurrent multipath transfer performance using network coding in wireless networks

IEEE Transactions on Vehicular Technology 2015 1

10.1109/tvt.2015.2409556

Maia

O. B.

Yehia

H. C.

de Errico

A concise review of the quality of experience assessment for video streaming

Computer Communications 2015 57 1 12

10.1016/j.comcom.2014.11.005

Xiao

Schaar

M. V. D.

Optimal foresighted multi-user wireless video

IEEE Journal of Selected Topics in Signal Processing 2015 9 1 89 101

10.1109/jstsp.2014.2332299

Liu

Guan

Zhang

Muntean

G.-M.

CMT-QA: quality-aware adaptive concurrent multipath data transfer in heterogeneous wireless networks

IEEE Transactions on Mobile Computing 2013 12 11 2193 2205

2-s2.0-84884689328

10.1109/tmc.2012.189

Chang

Kim

Binocular suppression-based stereoscopic video coding by joint rate control with KKT conditions for a hybrid video codec system

IEEE Transactions on Circuits and Systems for Video Technology 2015 25 1 99 111

10.1109/tcsvt.2014.2330658

Zhou

Yang

Wang

Guizani

Impact of execution time on adaptive wireless video scheduling

IEEE Journal on Selected Areas in Communications 2014 32 4 760 772

10.1109/jsac.2014.140408

2-s2.0-84897145105

10.

Bethanabhotla

Caire

Neely

Adaptive video streaming for wireless networks with multiple users and helpers

IEEE Transactions on Communications 2015 63 1 99 111

10.1109/tcomm.2014.2378774

11.

Kuo

J.-L.

Shih

C.-H.

C.-Y.

Chen

Y.-C.

A cross-layer approach for real-time multimedia streaming on wireless peer-to-peer ad hoc network

Ad Hoc Networks 2013 11 1 339 354

10.1016/j.adhoc.2012.06.008

2-s2.0-84870061346

12.

Jia

Zhong

Zhang

Muntean

G.-M.

Ant-inspired mini-community-based solution for video-on-demand services in wireless mobile networks

IEEE Transactions on Broadcasting 2014 60 2 322 335

10.1109/tbc.2014.2314791

2-s2.0-84898697509

13.

Zhang

Liu

Hei

Cheng

Unreeling Xunlei Kankan: understanding hybrid CDN-P2P video-on-demand streaming

IEEE Transactions on Multimedia 2015 17 2 229 242

10.1109/tmm.2014.2383617

14.

Yang

Zhou

Chen

T. Z. J.

Chiu

D. M.

Turbocharged video distribution via P2P

IEEE Transactions on Circuits and Systems for Video Technology 2015 25 2 287 299

10.1109/tcsvt.2014.2351093

15.

Liu

Zhu

Evolutionary game theoretic modeling and repetition of media distributed shared in P2P-based VANET

International Journal of Distributed Sensor Networks 2014 2014 14

718639

10.1155/2014/718639

16.

W.-T.

Chang

Y. S.

Sheu

R.-K.

Yang

C.-T.

Juang

T.-Y.

Y.-S.

Implementation and evaluation of large-scale video surveillance system based on P2P architecture and cloud computing

International Journal of Distributed Sensor Networks 2014 2014 11

375871

10.1155/2014/375871

2-s2.0-84899934675

17.

Huang

H.-C.

Wang

Dong

Peer selection strategy using mobile agent and trust in peer-to-peer streaming media system

International Journal of Distributed Sensor Networks 2013 2013 15

791560

10.1155/2013/791560

2-s2.0-84890105245

18.

Chen

Zhang

Chen

Chiu

D. M.

Performance modeling and evaluation of peer-to-peer live streaming systems under flash crowds

IEEE/ACM Transactions on Networking 2014 22 4 1106 1120

10.1109/tnet.2013.2272056

2-s2.0-84880967663

19.

Liu

Zhong

Jin

Liao

Flash crowd in P2P live streaming systems: fundamental characteristics and design implications

IEEE Transactions on Parallel and Distributed Systems 2012 23 7 1227 1239

10.1109/tpds.2011.283

2-s2.0-84861719824

20.

Carbunaru

Teo

Y. M.

Leong

Modeling flash crowd performance in peer-to-peer file distribution

IEEE Transactions on Parallel and Distributed Systems 2014 25 10 2617 2626

10.1109/tpds.2013.220

21.

Kozat

U. C.

Harmanci

Ö.

Kanumuri

Demircin

M. U.

Civanlar

M. R.

Peer assisted video streaming with supply-demand-based cache optimization

IEEE Transactions on Multimedia 2009 11 3 494 508

10.1109/tmm.2009.2012918

2-s2.0-63049112302

22.

Kim

Bahk

PECAN: peer cache adaptation for peer-to-peer video-on-demand streaming

Journal of Communications and Networks 2012 14 3 286 295

10.1109/jcn.2012.6253090

23.

Mokhtarian

Hefeeda

Capacity management of seed servers in peer-to-peer streaming systems with scalable video streams

IEEE Transactions on Multimedia 2013 15 1 181 194

10.1109/tmm.2012.2225042

2-s2.0-84871396437

24.

Venkatramanan

Kumar

Co-evolution of content popularity and delivery in mobile P2P networks

Proceedings of the IEEE INFOCOM

2012

25.

Altman

Nain

Shwartz

Predicting the impact of measures against P2P networks: transient behavior and phase transition

IEEE/ACM Transactions on Networking 2013 21 3 935 949

10.1109/tnet.2012.2217505

2-s2.0-84879136223

26.

Guo

Chen

Zhang

Design and evaluation of a scalable and reliable P2P assisted proxy for on-demand streaming media delivery

IEEE Transactions on Knowledge and Data Engineering 2006 18 5 669 682

10.1109/tkde.2006.79

2-s2.0-33645679814

27.

Choi

Reaz

A. S.

Mukherjee

A survey of user behavior in VoD service and bandwidth-saving multicast streaming schemes

IEEE Communications Surveys and Tutorials 2012 14 1 156 169

10.1109/SURV.2011.030811.00051

2-s2.0-84857504070

28.

Yiu

W.-P. K.

Jin

Chan

S.-H. G.

VMesh: distributed segment storage for peer-to-peer interactive video streaming

IEEE Journal on Selected Areas in Communications 2007 25 9 1717 1731

10.1109/jsac.2007.071210

2-s2.0-36749073538

29.

Zhao

Guan

Zhang

Muntean

G.-M.

QoE-driven user-centric vod services in urban multihomed P2P-based vehicular networks

IEEE Transactions on Vehicular Technology 2013 62 5 2273 2289

10.1109/TVT.2012.2228682

2-s2.0-84876577335

30.

Huang

C.-M.

Collaborative content fetching using MAC layer multicast in wireless mobile networks

IEEE Transactions on Broadcasting 2011 57 3 695 706

10.1109/TBC.2011.2147010

2-s2.0-80052074826

31.

Deng

Grey linear programming

Proceedings of the International Conference Information Processing Management Uncertainty Knowledge-Based System

1986

32.

Deng

J. L.

Introduction to grey system theory

The Journal of Grey System 1989 1 1 1 24

MR1025897

33.

Jia

Guan

Zhang

Muntean

G.-M.

A novel cooperative content fetching-based strategy to increase the quality of video delivery to mobile users in wireless networks

IEEE Transactions on Broadcasting 2014 60 2 370 384

10.1109/tbc.2014.2322772

A Novel Interest Detection-Based Video Dissemination Algorithm under Flash Crowd in Mobile Ad Hoc Networks

Abstract

1. Introduction

2. Related Work

3. IDVD Detailed Design

3.1. Media Server

3.2. Resource Disseminate Model

4. Testing and Test Results Analysis

4.1. Testing Topology and Scenarios

4.2. Performance Evaluation

5. Conclusion

Footnotes

Notations

Conflict of Interests

Acknowledgments

References