sensors Article
Data-Prefetching Scheme Based on Playback Delay and Positioning Satisfaction in Peer-To-Peer Video-On-Demand System Lei Wang 1,2 , Xiaorui Li 1,2 , Yaqiu Liu 1,3 and Guan Gui 1, * 1
2 3
*
ID
National and Local Joint Engineering Laboratory of RF Integration and Micro-Assembly Technology, Nanjing University of Posts and Telecommunications, Nanjing 210003, China;
[email protected] (L.W.);
[email protected] (X.L.);
[email protected] (Y.L.) National Mobile Communications Research Laboratory, Southeast University, Nanjing 210096, China Shanghai Radio Monitoring Station, Shanghai 200031, China Correspondence:
[email protected]; Tel.: +86-25-83492606
Received: 23 February 2018; Accepted: 6 March 2018; Published: 8 March 2018
Abstract: As one of the most important applications in peer-to-peer (P2P) networks, the video-on-demand (VoD) system freely supports video cassette recorder (VCR) operation for users. However, the users may experience significant playback delay after frequent VCR operations in the VoD system, which will affect the quality of experience (QoE) of the users. Hence, selecting an appropriate data-prefetching strategy to support better VCR operation is an important approach to improve the QoE. This paper proposes a data-prefetching strategy (DSA) to determine the most suitable anchor interval by considering the playback delay and positioning satisfaction. According to the DSA, we use the multiple-attribute decision-making (MADM) theory to model the selection of intervals of prefetching data blocks (i.e., anchor interval) and the technique for ordering preference by similarity to an ideal solution (TOPSIS) algorithm to solve the MADM. The simulation results show that the DSA strategy obtains higher positioning satisfaction than the existing schemes, which is approximately 60% higher than the anchor points, popular parts of video, and user interests (API)-based method. Moreover, with the increase in network bandwidth, the DSA strategy can minimize the playback delay after VCR operation using relative few extra bandwidths. Keywords: video-on-demand (VoD); video cassette recorder (VCR); multiple-attribute decision-making (MADM); playback delay; positioning satisfaction
1. Introduction In recent years, with the rapid development of internet and multimedia technology, increasingly more users have gained access to high-speed internet, and the service of online video-on-demand (VoD) [1] has been rapidly developed. Currently, one of the most common methods to support video-streaming services is peer-to-peer (P2P) networks. In P2P-based VoD systems [2], each peer can download and watch videos from the server or other peers, and provide the downloaded video segments to other requesting peers, which can improve the system’s service capabilities and reliability and reduce the server load and network construction costs. Compared with the live video system, a significant feature of the VoD system is that it can support the users to perform video cassette recorder (VCR) operations [3] (e.g., randomly dragging the progress bar for seeking, pause, fast forward, and rewind). In the P2P-based VoD system without data-prefetching technology, the users may experience significant playback delay after frequent video cassette recorder (VCR) operations, which seriously affects the users’ viewing experience. Hence, the most serious challenge in the design of P2P-based VoD systems is to find a suitable segment Sensors 2018, 18, 816; doi:10.3390/s18030816
www.mdpi.com/journal/sensors
Sensors 2018, 18, 816
2 of 13
selection algorithm to reduce the playback delay after the VCR operation. A data-prefetching strategy was proposed, i.e., if residual download bandwidth remains, the peer can prefetch a part of non-urgent data after the urgent data has been downloaded. The urgent data can ensure the video is continuously played in the next several minutes, and the non-urgent data is used to reduce the playback delay after the users perform VCR operations. In this strategy, the peer uses the residual bandwidth to download the data ahead of the current playback point, which can reduce the playback latency and effectively improve the viewing experience of users [4]. In [5], Yu et al. analyzed the VCR operation behaviors of users in large-scale VoD systems, and proposed a strategy to reduce the playback delay after the VCR operation by prefetching videos with high play counts. This strategy assumes that all users in the system have identical preferences, so they attempt to predict the video that is most likely to be played in the next few minutes by analyzing a large volume of user viewing logs. However, the assumption has drawbacks. For example, the viewing preferences of users vary for different times. Hence, inaccurate predictions will cause a low bit rate of prefetched data and waste bandwidth resources. In [6], the authors proposed an anchor-prefetching strategy. The anchor is distributed throughout the video file with a fixed interval (5 min), and each anchor consists of 10 continuous seconds. This traditional anchor-prefetching strategy can effectively reduce the playback delay after the VCR operation and avoid the problem of bandwidth waste caused by inaccurate prediction. The authors of [7] developed a prefetching strategy of API by prefetching both anchor points and popular segments of the video. This approach is complementary to the anchor-prefetching strategy, which can reduce the playback delay and improve the user satisfaction. However, to implement this strategy, it is necessary to detect the video segments that are currently played by each online P2P node, which will consume many extra bandwidth resources. In [8], the authors measured and analyzed the rules of VCR operation when the user watches the video. The forward- and backward-jump lengths have been modeled with a Weibull distribution. The forward and backward distances have different parameter values of Weibull distribution. In addition, for different lengths of video files, the Weibull distribution parameter values also vary. This is the theoretical basis of our study on the data-prefetching strategy. A new strategy to determine the most reasonable anchor interval is proposed in this paper by considering two factors of playback delay and positioning satisfaction. How to minimize the playback delay after the VCR operation is one of the most serious challenges in the design of P2P-based VoD systems, as mentioned above. And as mentioned in [6], in a VoD system with the anchor-prefetching strategy, the playback point is adjusted to the beginning of the closest anchor after the VCR operation. The disparity between the actual playback point and the designated point by the VCR operation may reduce the user’s viewing experience. Therefore, we introduce the concept of positioning satisfaction to reflect the user’s appraisal of the anchor-based VoD system caused by the disparity. Then, the user’s viewing experience is studied by considering the factors of playback delay and positioning satisfaction. The playback delay is a cost factor, and a lower value indicates better viewing experience of the users. The positioning satisfaction is a benefit factor, and a higher value indicates better viewing experience of the users. The problem of selecting the most reasonable anchor interval was modeled using the multiple-attribute decision-making (MADM) theory [9]. Based on the factors of playback delay and positioning satisfaction, the MADM model is established. By solving this model using the technique for ordering preference by similarity to an ideal solution (TOPSIS) method, the optimal anchor interval is selected from 20 optional anchor interval schemes. The remainder of the paper is structured as follows. The model of the P2P-based VoD system is described in Section 2. The problems studied in this paper are described in Section 3. The proposed optimal selection strategy of anchor interval is presented in Section 4. The performance evaluation is presented in Section 5. Section 6 concludes this paper.
Sensors FOR PEER REVIEW Sensors 2018, 2018, 18, 18, x816
33 of of 13 13
2. System Model 2. System Model Before stating the research of the data-prefetching strategy, it is necessary to introduce the BeforeVoD stating the research of the data-prefetching is necessary to introduce P2P-based system. This system is an overlay networkstrategy, [10] builtit between the terminals and the P2P-based VoD system. This system is an(CDN) overlay network [10] built between the terminals andand the integration content distribution network [11]. It can automatically distribute the content integration content [11]. It canthe automatically and segment the trafficdistribution flow. Thisnetwork system(CDN) can transmit videos in distribute CDN to the P2Pcontent terminals segment the traffic flow. This system canon transmit videos in CDN to P2P terminals (i.e., client node, (i.e., client node, CN) and share videos the CNthe using specified strategies. CN) As andillustrated share videos the CN strategies. inon Figure 1, using this specified communication system [12] is divided into three layers: As illustrated Figure 1,layer, this communication systemThe [12] management is divided intolayer three is layers: management management layer,in service and terminal layer. deployed on the layer, service layer, and layer. The management layersubsystem is deployed(MS), on the backbone network, backbone network, and terminal mainly includes a P2P management which is responsible andmanaging mainly includes P2P management subsystem which is responsible forinmanaging alllayer P2P for all P2P aregions. It collects the operation(MS), data of the network element the service regions. It collects operation of the web network element in the servicethe layer andonline displays the system and displays the the system statusdata through pages, which includes peer status, video status through web pages,etc. which includes the peerprovide online status, videofunction; distribution information, distribution information, In addition, it can the query each provincial etc. or In addition,network it can provide the query eachInprovincial municipal is divided into aa P2P P2P municipal is divided into afunction; P2P region. each P2Por region, a P2Pnetwork super node (SN) and region. In each P2P (TS) region, P2P super to node (SN) a P2Player. tracker subsystem (TS) are deployed to form tracker subsystem area deployed form theand service The SN is responsible for importing the service Thesystem, SN is responsible for importing into the and managing videos intolayer. the P2P splitting and managingvideos the videos inP2P the system, system,splitting and providing video the videosfor in the system, videoresponsible segments for peers. The TS of is mainly for segments other peers.and Theproviding TS is mainly forother the management CNs inresponsible a P2P region. the management in a P2Pthe region. Meanwhile, TS continues video directory management to Meanwhile, the of TSCNs continues video directory the management tothe support video positioning. The support video positioning. The terminal layer consists of numerous CNs, which can be mobile phones [13], terminal layer consists of numerous CNs, which can be mobile phones [13], tablets, desktops, etc. The tablets, The CNvideos can download and video play videos and for provide segments for other CNs CN can desktops, downloadetc. and play and provide segments othervideo CNs in the same P2P region. in the samethis P2Psystem region.cannot However, this system cannotdata support inter-region data among the CNs. However, support inter-region exchange among theexchange CNs. MS
management layer
P2P Region
P2P Region TS
SN
service layer
TS
SN
...
terminal layer
...
...
CN
CN
... CN
CN
...
...
CN
CN
CN
CN
Figure1. 1.Architecture Architectureof of the the peer-to-peer peer-to-peer (P2P)-based (P2P)-based video-on-demand video-on-demand (VoD) (VoD) system. Figure
When first register with thethe TS.TS. After the the registration is successful, the When aaCN CNgoes goesonline, online,it itmust must first register with After registration is successful, servable video list of this region is sent to this CN by the TS. The users can select the video that the servable video list of this region is sent to this CN by the TS. The users can select the video that interests list. After interests them them from from the the video video list. After receiving receiving the the user’s user’s selection, selection, the the CN CN must must determine determine the the download download order order of of the the video video segments segments according according to to the the segment segment selection selection algorithm. algorithm. Then, Then, the the CN CN requests requests the the servable servable peers peers from from the the TS TS and and sorts sorts the the servable servable nodes. nodes. After After selecting selecting the the most most suitable suitable node as the content provider, the CN can download video segments in order. It is notably node as the content provider, the CN can download video segments in order. It is notably important important to to improve improve the the user’s user’s viewing viewing experience experience by by determining determining aa suitable suitable order order to to download download the the video video segments, which we will introduce in detail in the following chapters. segments, which we will introduce in detail in the following chapters.
Sensors 2018, 18, x FOR PEER REVIEW
4 of 13
Sensors 2018, 18, 816
4 of 13
3. Detailed Problem Formulation
The segment-downloading sequence affects the viewing experience of the user in the P2P-based 3. Detailed Problem Formulation VoD system. Therefore, many segment selection strategies were proposed. To support the particularity of the VCR operation in the affects P2P-based VoD system, data-prefetching were The segment-downloading sequence the viewing experience of the user instrategies the P2P-based introduced a specificmany segment selection strategy, i.e., were downloading data before the current VoD system.asTherefore, segment selection strategies proposed.the To support the particularity playing point in advance, which is designed to reduce the playback delay after VCR operations. The of the VCR operation in the P2P-based VoD system, data-prefetching strategies were introduced as most mature data-prefetching strategy is the anchor point strategy [6]. When the user watches a a specific segment selection strategy, i.e., downloading the data before the current playing point in video, thewhich CN with the anchor-prefetching strategy delay downloads the anchor data with the residual advance, is designed to reduce the playback after VCR operations. The most mature download bandwidth after downloading the urgent data. When the user jumps to a new point the data-prefetching strategy is the anchor point strategy [6]. When the user watches a video, the CNofwith video using a VCR operation, thedownloads playback point is adjusted thethe beginning the closestbandwidth anchor. If the anchor-prefetching strategy the anchor data to with residualofdownload the anchor has beenthe downloaded, theWhen CN will playing video with this a paper, after downloading urgent data. theresume user jumps to athe new point of no thedelay. video In using VCR we hope to improve the viewing experience of the user after a VCR operation by improving the operation, the playback point is adjusted to the beginning of the closest anchor. If the anchor has been anchor-prefetching strategy. To improve weno consider two aspects: duration and downloaded, the CN will resume playingthis thestrategy, video with delay. In this paper,anchor we hope to improve anchor interval. the viewing experience of the user after a VCR operation by improving the anchor-prefetching strategy. First, we discuss thewe anchor duration. Each anchor be a and blockanchor that can be independently To improve this strategy, consider two aspects: anchormust duration interval. played. Based on the content fragmentation strategy in must the P2P-based in Section 2, a First, we discuss the anchor duration. Each anchor be a blockVoD thatsystem can be independently chunk size is approximately 2 M [14]. An Android native player must download approximately 2 played. Based on the content fragmentation strategy in the P2P-based VoD system in Section 2, a chunk chunks before it begins video, i.e., the player data block has at least 4 M bits that2 chunks can be size is approximately 2 Mplaying [14]. AnaAndroid native must download approximately independently to the the data research PPLive, the4 M video ratebe is independently 381–700 kbps. before it beginsplayed. playingAccording a video, i.e., blockofhas at least bits code that can Therefore, to satisfytothe of the player to play video segments, each anchor duration played. According therequirement research of PPLive, the video codethe rate is 381–700 kbps. Therefore, to satisfy is s. To ensure theplayer normal of video with different bit rates, we set the anchor duration the6–10 requirement of the to playback play the video segments, each anchor duration is 6–10 s. To ensure the to 10 s. normal playback of video with different bit rates, we set the anchor duration to 10 s. Second, anchor interval, interval, we we must must consider consider two two factors: factors: playback Second, to to study study the the anchor playback delay delay and and positioning positioning satisfaction. satisfaction. Playback Delay Delay 3.1. Playback is the the time time the the CN CN takes takes to to resume resume playing playing the the video video after after aa VCR VCR operation, operation, The playback delay is as illustrated in Figure 2. A smaller playback delay indicates a better viewing experience.
0
elapsed time
unelapsed time
30s
time line playback delay VCR operation completed
restart playing Figure Figure 2. 2. Playback Playback delay. delay.
When the anchor duration is fixed, if the anchor interval increases, the anchors that are perfected When the anchor duration is fixed, if the anchor interval increases, the anchors that are perfected with identical residual download bandwidths will have a wider distribution in the video. When the with identical residual download bandwidths will have a wider distribution in the video. When the user user randomly jumps to a certain point using a VCR operation, the probability that the anchor near randomly jumps to a certain point using a VCR operation, the probability that the anchor near the point the point that has been downloaded increases, and the probability that the user experiences a delay that has been downloaded increases, and the probability that the user experiences a delay decreases. decreases. Therefore, the average playback delay decreases. Therefore, the average playback delay decreases. We simulated the VCR behavior of a VoD system. The fast-forward and -backward operations We simulated the VCR behavior of a VoD system. The fast-forward and -backward operations compose a notably small percentage (approximately 1%) of the user’s actual operations. As compose a notably small percentage (approximately 1%) of the user’s actual operations. As mentioned mentioned in [15,16], most of the user interactions are forward search (48%) and pause (51%). Most in [15,16], most of the user interactions are forward search (48%) and pause (51%). Most VCR operations VCR operations can be achieved through the jump process: jump to a new point from the current can be achieved through the jump process: jump to a new point from the current play point and resume play point and resume playing the video. One search or pause must only jump once, whereas fast playing the video. One search or pause must only jump once, whereas fast forwards or backwards forwards or backwards usually consist of a series of jump processes [17]. Therefore, to study the usually consist of a series of jump processes [17]. Therefore, to study the playback delay of the VCR playback delay of the VCR operation, this paper mainly pays attention to the forward-search jump operation, this paper mainly pays attention to the forward-search jump behavior when the users are behavior when the users are watching long videos. watching long videos.
Sensors 2018, 18, 816
5 of 13
Sensors 2018, 18, x FOR PEER REVIEW
5 of 13
The lengthofofeach each jump in the operation complies with Weibull distribution [8], The length jump in the VCRVCR operation complies with Weibull distribution [8], and the and the probability density function is as follows: probability density function is as follows: α x α−1 −(βx()xα ) ( )x e f ( x, , ) β β( ) 1 e f ( x, α, β) =
(1) (1)
Table 1 shows the parameters of the distribution. Table 1 shows the parameters of the distribution. Table 1. Jump length characterization. Table 1. Jump length characterization. Interaction Interaction
Video Type Video Type Short videos Forward jump length Short videos Forward jump length Long videos Long videos Short videos Backward jump length Short videos Backward jump length Long videos Long videos
Distribution Distribution
Parameters
Parameters
Weibull Weibull
β =β0.58631 αα==0.13214, 0.13214, = 0.58631 α = 0.10916, β = 0.54129
Weibull Weibull
β =β0.59850 αα==0.08108, 0.08108, = 0.59850 α = 0.05959, β = 0.59236
α = 0.10916, β = 0.54129 α = 0.05959, β = 0.59236
After the video was played for 1 min, the user randomly performed 20 VCR operations; then, the average playback delay was recorded. As shown in Figure 3, with the increase in anchor interval, the playback delay gradually decreases. When the users watch a video on demand, the delay should be as short From this this perspective, perspective, aa larger anchor interval corresponds to a better short as as possible. possible. From user experience. 3
Average Playback Delay (s)
2.8 2.6 2.4 2.2 2 1.8 1.6 1.4 1.2 1
0
1
2
3
4
5
6
7
8
9
10
Anchor Interval (min) Figure playback delay delay with with different different anchor anchor intervals. intervals. Figure 3. 3. Average Average playback
3.2. Positioning Satisfaction 3.2. Positioning Satisfaction When the user performs a VCR operation, he/she hopes that the player resumes playing the When the user performs a VCR operation, he/she hopes that the player resumes playing the video exactly from the designated point. As mentioned in the anchor-prefetching strategy, the actual video exactly from the designated point. As mentioned in the anchor-prefetching strategy, the actual playback point of the player does not exactly coincide with the designated point by the user. Thus, playback point of the player does not exactly coincide with the designated point by the user. Thus, the concept of positioning satisfaction is generated. the concept of positioning satisfaction is generated. As shown in Figure 4, α is the middle of the anchor interval. Therefore, if the user jumps to point As shown in Figure 4, α is the middle of the anchor interval. Therefore, if the user jumps to α after a VCR operation, the worst case occurs, and the positioning satisfaction is minimal. However, point α after a VCR operation, the worst case occurs, and the positioning satisfaction is minimal. if the user jumps to point γ or β, because the anchors have been perfected, the best case occurs, and However, if the user jumps to point γ or β, because the anchors have been perfected, the best case the positioning satisfaction is 1 (i.e., the maximum value). By closing from α to β (or γ), the occurs, and the positioning satisfaction is 1 (i.e., the maximum value). By closing from α to β (or γ), positioning satisfaction is increased. the positioning satisfaction is increased.
Sensors 2018, 18, x FOR PEER REVIEW
6 of 13
Data of downloaded Sensors FOR PEER REVIEW Sensors2018, 2018,18, 18,x816 progress bar progress bar
Data of downloaded current
L
Data to be downloaded β α
γ
γ
Data to be downloaded L β α anchor interval
L
anchor interval L
L
current
66of of13 13
L
Figure 4. Seeking inanchor a video. anchor interval interval
Satisfaction is a fuzzy concept of people psychology, we use the membership function of fuzzy Figure 4. Seeking in aso video. Figure 4. Seeking in adefine video.that the positioning satisfaction is 1 if mathematics to define the positioning satisfaction. We the time gap of the playback the designated point less than 5 function s, and a of greater Satisfaction is a actual fuzzy concept of point peopleand psychology, so we use theismembership fuzzy Satisfaction is a fuzzy concept of people psychology, so we use the membership function of fuzzy difference implies lower positioningsatisfaction. satisfaction.WeThe positioning satisfactionsatisfaction with timeis gap is mathematics to define the positioning define that the positioning 1 if the mathematics to definetothe positioning satisfaction. We define that the positioning satisfaction is 1 if calculated according Equation (2): time gap of the actual playback point and the designated point is less than 5 s, and a greater difference the time gap of the actual playback point and the designated point is less than 5 s, and a greater implies lower positioning satisfaction. The positioning satisfaction gap is calculated according to x 5swith time 1, The difference implies lower positioning satisfaction. positioning satisfaction with time gap is Equation (2): (2) calculated according to Equation (2): S ( x ) ( x 5 )2 e 1, , xx ≤ 55ss (2) S( x ) = 1,x−5 2 x 5s e−( λ ) , x > 5s where ߣ is a constant strength factor.S ( x ) x 5 2 (2) ( ) We have studied how the anchor interval affects the positioning satisfaction using simulations. , x 5s where λ is a constant strength factor. e SinceWe thehave userstudied more how likelythejumps a VCR operation satisfaction [18,19]. Weusing mainly focus onSince the anchorforward interval for affects the positioning simulations. where ߣ more is a constant strength factor. forward-search behavior. After the for video wasoperation played for[18,19]. 1 min, We the mainly user randomly 20 VCR the user likely jumps forward a VCR focus onperformed the forward-search We have studied how theplayed anchorfor interval affects positioning satisfaction simulations. operations; then, average positioning satisfaction was recorded. As shown inusing Figure 5, withthen, the behavior. After thethe video was 1 min, the userthe randomly performed 20 VCR operations; Since the user more likely jumps forward for a VCR operation [18,19]. We mainly focus thea increase in anchor interval, the positioning satisfaction gradually decreases. From perspective, the average positioning satisfaction was recorded. As shown in Figure 5, with the this increase in on anchor forward-search behavior. Aftercorresponds the video wasdecreases. played forFrom 1 min, theperspective, user randomly performed VCR smaller anchor interval to higher positioning satisfaction and 20 better interval, the positioning satisfaction gradually this a smaller anchor interval operations; then, the average positioning satisfaction recorded. As shown in Figure 5, with the system performance. corresponds to higher positioning satisfaction and betterwas system performance. increase in anchor interval, the positioning satisfaction gradually decreases. From this perspective, a smaller anchor interval1 corresponds to higher positioning satisfaction and better system performance. Average Average Positioning Satisfaction Positioning Satisfaction
0.9 1 0.8
0.9 0.7 0.8 0.6 0.7 0.5 0.6 0.4 0.5 0.3 0.4 0.2 0.3 0.1 0.2
0
1
2
3
4
5
6
7
8
9
10
Anchor Interval (min)
Figure 5. Average satisfaction with different anchor intervals. 0.1 Figure with different intervals. 0 5. Average 1 2 satisfaction 3 4 5 6 7 anchor 8 9 10
Anchor Interval (min) It is not acceptable for the user to experience a long playback delay when they perform a VCR It is not acceptable for the user to experience a long playback delay when they perform a VCR Figure 5. Average satisfaction with different anchor operation, i.e., a shorter playback delay corresponds to a better userintervals. experience. According to the operation, i.e., a shorter playback delay corresponds to a better user experience. According to the analysis in Section 3.1, a larger anchor interval is better. analysis in Section 3.1, afor larger anchor is better. ItThe is not acceptable the user to interval experience a longbetween playbackthe delay when they perform a VCR positioning satisfaction reflects the difference actual playback point and the The positioning satisfaction reflects the difference between the actual playback point and the operation, i.e., a shorter playback delay corresponds to a better user experience. According to the designated point. A smaller difference indicates higher satisfaction of the user. According to the designated point. A smaller difference indicates higher satisfaction of the user. According to the analysis anchor interval is is better. analysis in in Section Section 3.1, 3.2, aa larger smaller anchor interval better. analysis Section 3.2, a smaller anchor is better.between the actual playback point and the The in positioning satisfaction reflectsinterval the difference Therefore, theAplayback and positioning satisfaction are two factors. designated point. smaller delay difference indicates higher satisfaction ofconflicting the user. According to the analysis in Section 3.2, a smaller anchor interval is better.
Sensors 2018, 18, 816
7 of 13
4. Anchor Optimization Selection Strategy As analyzed, the playback delay is a cost factor, and the positioning satisfaction is a benefit factor. Based on the two types of factors, we present the anchor optimization selection strategy DSA. This paper aims at selecting the most suitable scheme from the limited number of optional anchor interval schemes. Hence, it is a typical MADM problem. As one of the most widely used method to solve the MADM problem, the TOPSIS method [20] proposes the concept of ideal solution (S+ ) and negative ideal solution (S− ). The basic principle of TOPSIS is using the distance from each candidate to S+ and S− as the criterion to rank the candidates. According to the ranking, we can select the most suitable candidate. The detailed steps of the TOPSIS method are summarized as follows: Step 1.
Construct the normalized decision matrix A
First, it is important to construct the original decision matrix before the normalization, and the decision matrix V with m candidates and n factors is shown in Equation (3): In this paper, the decision matrix V consists of 20 optional schemes and 2 factors. V=
x11 x21 .. . xi1 .. . xm1
x12 x22 .. . ··· .. . xm2
··· ··· .. . xij .. . ···
x1n x2n .. . ··· .. . xmn
(3)
where xij is the j-th factor value of the i-th scheme. To eliminate the difference of these two factors, the decision matrix must be normalized. For the benefit factor and cost factor, the normalization method is shown in Equations (4) and (5), respectively. aij =
xij − min( xij ) max( xij ) − min( xij )
(4)
aij =
max( xij ) − xij max( xij ) − min( xij )
(5)
Then, the normalized decision matrix A is obtained: A=
Step 2.
a11 a21 .. . ai1 .. . am1
a12 a22 .. . ··· .. . am2
··· ··· .. . aij .. . ···
a1n a2n .. . ··· .. . amn
(6)
Calculate the weight of each factor
The entropy weight is calculated as follows: wj =
1 − ej n
n − ∑ ej j =1
(7)
Sensors 2018, 18, 816
8 of 13
In Equation (7), e j is the entropy of factor j: ej = −
n 1 ( ∑ f ij ln f ij ) (0 ≤ e j ≤ 1) ln n i=1
(8)
In Equation (8), f ij is the characteristic proportion of the factor: f ij =
aij
(9)
m
∑ aij
i =1
Step 3.
Construct the evaluation matrix based on the entropy weight: Z=
Step 4.
a11 a21 .. . ai1 .. . am1
a12 a22 .. . ··· .. . am2
··· ··· .. . aij .. . ···
a1n a2n .. . ··· .. . amn
w1 0 .. . 0 .. . 0
0 w2 .. . ··· .. . 0
··· ··· .. . wj .. . ···
0 0 .. . ··· .. . wn
=
z11 z21 .. . zi1 .. . zm1
z12 z22 .. . ··· .. . zm2
··· ··· .. . zij .. . ···
z1n z2n .. . ··· .. . zmn
(10)
Determine the ideal and negative ideal schemes
This step determines the ideal scheme S+ and negative ideal scheme S− . 0 S+ = [z1+ , z2+ · · · , z+ n ] = {(max zij | j ∈ J ), (min zij | j ∈ J )|
i = 1, . . . , m } j = 1, . . . , n
(11)
0 S− = [z1− , z2− · · · , z− n ] = {(min zij | j ∈ J ), (max zij | j ∈ J )|
i = 1, . . . , m } j = 1, . . . , n
(12)
i
i
i
i
where J is the set of subscripts of the benefit factor; J 0 is the set of subscripts of the cost factor. Step 5.
Calculate the distance of each scheme from the ideal and negative ideal schemes v u n u 2 + di = t ∑ (zij − s+ j ) (i = 1, . . . , m )
(13)
v u n u 2 − di = t ∑ (zij − s− j ) (i = 1, . . . , m )
(14)
j =1
j =1
Step 6.
Calculate the distance of each scheme from the ideal and negative ideal schemes ri =
(di−
di−
+ di+ )
(15)
Sensors 2018, 18, 816
Step 7.
9 of 13
Sort the candidate schemes according to ri .
In this study, we propose 20 optional anchor interval schemes, and the length of the interval is 0.5–10 min with uniform increments. The TOPSIS method is used to comprehensively evaluate the 20 schemes, and the priority order of the candidate schemes is shown in Table 2. Table 2. Ranking results for 20 candidate schemes. Ranking
Scheme Num
Anchor Interval
Ranking
Scheme Num
Anchor Interval
1 2 3 4 5 6 7 8 9 10
Scheme 5 Scheme 6 Scheme 7 Scheme 4 Scheme 8 Scheme 9 Scheme 10 Scheme 11 Scheme 12 Scheme 13
2.5 min 3 min 3.5 min 2 min 4 min 4.5 min 5 min 5.5 min 6 min 6.5 min
11 12 13 14 15 16 17 18 19 20
Scheme 14 Scheme 15 Scheme 16 Scheme 17 Scheme 18 Scheme 19 Scheme 20 Scheme 3 Scheme 2 Scheme 1
7 min 7.5 min 8 min 8.5 min 9 min 9.5 min 10 min 1.5 min 1 min 0.5 min
In the TOPSIS method, each factor value of the ideal solution is the optimal value of all candidates. In other words, the ideal solution has the smallest playback delay and highest positioning satisfaction among all candidate schemes. Therefore, the ideal solution can achieve the best viewing experience. Each factor value of the negative ideal solution is the worst value of all candidates. The negative ideal solution will ideally result in the worst viewing experience. In fact, the ideal solution and negative ideal solution do not exist in the proposed candidates. The best candidate has the farthest distance from the negative ideal solution and the shortest distance from the ideal solution. Therefore, the best candidate has the highest viewing experience among all candidates. Table 2 shows that scheme 5 is optimal, and the anchor interval is 2.5 min. We name this data-prefetching strategy the DSA strategy considering the playback delay and positioning satisfaction. 5. Experiment Evaluation To verify the rationality of the proposed strategy in this paper, we compared it with the most mature anchor strategy, whose anchor interval is 5 min [6], and the API strategy proposed in [7]. The DSA strategy proposed in this paper is an improvement strategy for the anchor strategy. The proposed scheme improved the anchor interval to 2.5 min by the anchor optimization selection strategy formulated in Section 4. These two strategies share the same data prefetching mode. Peers with these two strategies download the anchor data before the current playing point with the surplus download bandwidth after downloading the urgent data. The API strategy is also an improvement strategy for anchor strategy, which aims to increase the user satisfaction and reduce the playback delay. In this strategy, peers downloads both the anchor at intervals of 5 min and the video segments that are popular with users in the current system before the current playing point. We implemented the P2P-based VoD system and data-prefetching algorithm using Java. In the comparative experiment, we simulated the VCR operation process when the user watches a video in a P2P-based VoD system, and the test video bit rate is 600 kbps. After the video was played for 1 min, the user randomly performed 5 times VCR operations every 15 s. 5.1. Playback Delay In a P2P-based VoD system, the data-prefetching strategy is proposed to minimize the latency that the user must wait after the VCR operation. Therefore, the playback delay is one of the most important indicators to evaluate the validity of a data-prefetching strategy. For identical network bandwidths, a smaller delay indicates a better data-prefetching strategy.
5.1. Playback Delay 5.1. Playback Delay In a P2P-based VoD system, the data-prefetching strategy is proposed to minimize the latency In a P2P-based VoD system, the data-prefetching strategy is proposed to minimize the latency that the user must wait after the VCR operation. Therefore, the playback delay is one of the most that the user must wait after the VCR operation. Therefore, the playback delay is one of the most important indicators to evaluate the validity of a data-prefetching strategy. For identical network Sensors 2018, 18, 816 10 of 13 important indicators to evaluate the validity of a data-prefetching strategy. For identical network bandwidths, a smaller delay indicates a better data-prefetching strategy. bandwidths, a smaller delay indicates a better data-prefetching strategy. As shown in Figure 6, when the network bandwidth is small, the DSA strategy requires the As shown in Figure 6, when the network bandwidth is small, the DSA strategy requires the As average shown in Figure the0.5 network is small, the DSA strategy highest delay, but6,itwhen is only s more bandwidth than the other two strategies. With therequires increasethe in highest average delay, but it is only 0.5 s more than the other two strategies. With the increase in highest average delay, but it is only 0.5 s more than the other two strategies. With the increase in network bandwidth, the DSA strategy has the fastest performance improvement. When the network network bandwidth, the DSA strategy has the fastest performance improvement. When the network network bandwidth, DSA strategy has performance improvement. thethan network bandwidth increases the to approximately 6 M,the thefastest average delay of the DSA strategyWhen is lower that bandwidth increases to approximately 6 M, the average delay of the DSA strategy is lower than that bandwidth to approximately 6 M, the delay of the DSA strategy is lower than thatthe of of the APIincreases strategy, and almost equal toaverage the anchor-prefetching strategy, which has of the API strategy, and almost equal to the anchor-prefetching strategy, which has the the API strategy, and almost equal to the anchor-prefetching strategy, which has the minimal value. minimal value. minimal value. 2.4 2.4
API API DSA DSA Anchor Anchor
2.2 2.2
Average Delay Average Delay (s)(s)
2 2 1.8 1.8 1.6 1.6 1.4 1.4 1.2 1.2 1
1 1 1
2
3
2
3
4
5
4 5 Bandwidth (Mbps) Bandwidth (Mbps)
6
7
8
6
7
8
Figure delay of different strategies. Figure 6. 6. Average Average playback playback Figure 6. Average playback delay delay of of different different strategies. strategies.
5.2. Positioning Satisfaction 5.2. Positioning Satisfaction Satisfaction 5.2. Positioning The VoD system enables the users to perform VCR operations and quickly search for interesting The VoD VoD system system enables enables the users to perform VCR operations and quickly search for interesting content. The users hope that the player resumes playing the video exactly from their designated point content. The Theusers usershope hope that player resumes playing the video exactly their designated that thethe player resumes playing the video exactly fromfrom their designated point after the VCR operation. Positioning satisfaction is the indicator to evaluate the proximity of the point afterVCR the VCR operation. Positioning satisfaction is the indicator evaluatethe theproximity proximity of of the after the operation. Positioning satisfaction is the indicator totoevaluate designated point and the actual playback point. designated point and the actual playback point. point. As shown in Figure 7, with the increase in number of VCR operations, the positioning As shown Figure 7, with the increase in number of VCR operations, the positioning satisfaction showninin Figure 7, with the increase in number of VCR operations, the positioning satisfaction of the three strategies basically remains stable. The API strategy prefetches the original of the three strategies basically remains stable. The API strategy prefetches the original anchor points satisfaction of the three strategies basically remains stable. The API strategy prefetches the original anchor points and popular segments. To a certain degree, the API strategy improves the positioning and popular segments. To a certain degree, the API strategy improves the positioning satisfaction. anchor points and popular segments. To a certain degree, the API strategy improves the positioning satisfaction. However, compared with the API strategy, the proposed DSA strategy in this paper However, compared with the APIwith strategy, the proposed DSA strategy DSA in this paper improves the satisfaction. However, compared the API strategy, the proposed strategy in this paper improves the positioning satisfaction by approximately 60%. positioning satisfaction by approximately 60%. improves the positioning satisfaction by approximately 60%. 0.8
Average Positioning Satisfaction Average Positioning Satisfaction
0.8 0.7 0.7
API API DSA DSA Anchor Anchor
0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 20 0.1 20
30
40
80
90
100
30
40 50 of VCR 60 Operations 70 80 Number Number of VCR Operations
50
60
70
90
100
Figure 7. Average positioning satisfaction of different strategies.
Sensors 2018, 18, x FOR PEER REVIEW Sensors 2018, 18, 816
11 of 13 11 of 13
Figure 7. Average positioning satisfaction of different strategies.
Extra Bandwidth Consumption 5.3.5.3. Extra Bandwidth Consumption The peers in P2P-based the P2P-based VoD system information todata, prefetch data, The peers in the VoD system requirerequire auxiliaryauxiliary information to prefetch and the and the bandwidth consumed by the auxiliary information is the extra bandwidth. To prefetch data bandwidth consumed by the auxiliary information is the extra bandwidth. To prefetch data for thefor the using peers using the anchor-prefetching strategy orDSA the DSA strategy, extra fields must added peers the anchor-prefetching strategy or the strategy, twotwo extra fields must be be added to to the index file: the anchor length and the list of starting segment numbers for each anchor. When a the index file: the anchor length and the list of starting segment numbers for each anchor. Whenpeer a begins watching an on-demand video, the index file from the server or peers befirst obtained peer begins watching an on-demand video, the index file from the server or must peers first must be and locally saved. saved. Then, Then, the peer the segments according to thetoindex file and obtained and locally therequests peer requests the segments according the index file the anduser’s the playing request. Therefore, to implement these prefetching strategies, one peer must only user’s playing request. Therefore, to implement these prefetching strategies, one peer musttransmit only two additional fields. fields. transmit two additional However,the theAPI API strategy both anchors and popular segments. To achieve this prefetching However, strategyprefetches prefetches both anchors and popular segments. To achieve this strategy, adding the two fields of the anchor length and starting segment number for each anchor to prefetching strategy, adding the two fields of the anchor length and starting segment number forthe index file is to notthe sufficient, mentioned. acquire thementioned. popular segments of the the current system, each anchor index as filepreviously is not sufficient, as To previously To acquire popular the server must perceive which segment of the video is currently played by each online peer. To satisfy segments of the current system, the server must perceive which segment of the video is currentlythis requirement, information (thethis video ID and currently played videofields segment) be and added played by eachtwo online peer. Tofields satisfy requirement, two information (the should video ID to the heartbeat packet,segment) which is should periodically sent by The heartbeat packet is aiscollection of real-time currently played video be added topeers. the heartbeat packet, which periodically sent basic information of the peers that must be periodically uploaded to the tracker server (TS) by online by peers. The heartbeat packet is a collection of real-time basic information of the peers that must peers. be The server uploaded manages the throughserver the heartbeat information. peer sends a heartbeat packet periodically topeers the tracker (TS) bypacket online peers. TheAserver manages the peers every 30 by default.packet information. A peer sends a heartbeat packet every 30 s by default. through thes heartbeat When online peer watches 60 min video, assume watches the video When an an online peer watches a 60a min video, we we assume thatthat the the useruser watches the video for afor a total 30 min a series of VCR operations. extra bandwidth required three strategies total of 30ofmin overover a series of VCR operations. TheThe extra bandwidth required forfor thethe three strategies is shown in Figure 8. is shown in Figure 8.
Extra bandwidth (Byte)
4000
3700
3500 3000 2500 2000 1500 1000 500
150
100
DSA
Anchor
0
API
Data prefetching strategies Figure 8. Extra bandwidth consumption of different strategies. Figure 8. Extra bandwidth consumption of different strategies.
In the API strategy, the peer’s playing information must be added to the periodically sent In the API which strategy, the peer’s playing must be added the two periodically sent heartbeat packet, consumes 20–30 timesinformation more bandwidth than the to other strategies. heartbeat packet, which consumes 20–30 times more bandwidth than the other two strategies. However, However, the anchor-prefetching strategy and proposed DSA strategy in this paper consume notably the anchor-prefetching strategy and proposed DSA strategy in this paper consume notably little extra little extra bandwidth for each video, which is within the acceptable range. bandwidth forofeach video, of which is within the acceptable range. When tens thousands online peers watch thousands of on-demand videos in the P2P-based When tens of thousands of online peers watch thousands of on-demand videos the P2P-based VoD system, the extra bandwidth consumed by the API sharply increases compared toin the other two VoD system, the extra bandwidth consumed by the API sharply increases compared to the other strategies. In other words, the implementation of the API causes a significant system burden. two strategies. other the implementation of the API causes a significant system burden. Therefore, the APIIn has poorwords, practicality. Therefore, the API has poor practicality. 6. Conclusions In this paper, the optimal data-prefetching scheme is derived using the TOPSIS algorithm based on the MADM theory: the DSA strategy. In order to determine the most suitable data prefetching
Sensors 2018, 18, 816
12 of 13
6. Conclusions In this paper, the optimal data-prefetching scheme is derived using the TOPSIS algorithm based on the MADM theory: the DSA strategy. In order to determine the most suitable data prefetching strategy, we considered factors of different type which can influence user’s viewing experience, such as playback delay and positioning satisfaction. The experiment demonstrates that the proposed DSA strategy in this paper improves the positioning satisfaction by approximately 60% relative to the API strategy. Moreover, the DSA strategy can ensure that the delay performance is relatively good. In addition, the delay performance increases the fastest with the increase in network bandwidth, compared with the other two strategies, and achieving this strategy requires less extra bandwidth. Therefore, the proposed DSA strategy in this paper has the best performance based on the comprehensive evaluation in these three strategies, and it can enable users to obtain a better viewing experience. Also, this paper has presented a model to determine the most suitable prefetching data strategy based on different factors. The model has good extensibility, and can be extended from two factors to multiple factors. This model can also be applied to the segment selection algorithm to determine the downloading order of different types of segments. Due to the limited conditions, the experiments were evaluated by simulations with Java. Hence, some future work is still needed. The performance of the DSA strategy needs to be validated in real experimental environments. Acknowledgments: This work was supported by the National Natural Science Foundation of China (61671253, 61571240); The Major Projects of the Natural Science Foundation of the Jiangsu Higher Education Institutions (16KJA510004); the open research Foundation of National and Local Joint Engineering Laboratory of RF Integration and Micro-Assembly Technology, Nanjing University of Posts and Telecommunications (KFJJ20170305); and the open research fund of National Mobile Communications Research Laboratory, Southeast University (2016D01). Author Contributions: Lei Wang and Xiaorui Li proposed the algorithm. Xiaorui Li conceived and designed the experiments; Yaqiu Liu performed the practical experiments; Lei Wang analyzed the data; Guan Gui and Xiaorui Li wrote the paper. Lei Wang checked this paper. Conflicts of Interest: The authors declare no conflict of interest.
References 1.
2. 3.
4.
5. 6.
7.
Sheshjavani, A.G.; Akbari, B.; Ghaeini, H.R. CMPVoD: A cluster mesh-based architecture for VoD streaming over hybrid CDN-P2P networks. In Proceedings of the International Symposium on Telecommunications (IST), Tehran, Iran, 27–28 September 2016; pp. 783–788. Passarella, A. A survey on content-centric technologies for the current Internet: CDN and P2P solutions. Comput. Commun. 2012, 35, 1–32. [CrossRef] Zhong, L.; Xu, C. DLCA: Distributed load balancing and VCR-aware two-tier P2P VoD system. In Proceedings of the IEEE International Conference on Consumer Communications and Networking Conference (CCNC), Las Vegas, NV, USA, 10–13 January 2014; pp. 199–204. Liu, P.; Huang, G.; Zhou, Y.; Qin, D.; Liu, S. Server load based prefetching strategy for P2P VoD streaming. In Proceedings of the International Conference on Computer Science and Network Technology (ICCSNT), Dalian, China, 12–13 October 2013; pp. 721–725. Riad, A.M.; Elmogy, M.; Shehab, A.I. A framework for cloud P2P VoD system based on user’s behavior analysis. Int. J. Comput. Appl. 2013, 76, 20–26. Xu, T.; Ye, B.; Wang, Q.; Li, W.; Lu, S.; Fu, X. APEX: A personalization framework to improve quality of experience for DVD-like functions in P2P VoD applications. In Proceedings of the 18th International Workshop on Quality of Service (IWQoS), Beijing, China, 16–18 June 2010; pp. 1–9. Aliakbari, Z.; Fooladi, M.D.T. Pre-fetching strategy to support VCR operation in P2P VoD systems. In Proceedings of the International Conference on Computer and Knowledge Engineering (ICCKE), Mashhad, Iran, 29–30 October 2014; pp. 619–624.
Sensors 2018, 18, 816
8.
9. 10. 11. 12. 13. 14.
15.
16.
17. 18. 19. 20.
13 of 13
García, R.; Pañeda, X.G.; García, V.; Melendi, D.; Vilas, M. Statistical characterization of a real video on demand service: User behaviour and streaming-media workload analysis. Simul. Model. Pract. Theory 2007, 15, 672–689. [CrossRef] Yoon, K.P.; Hwang, C.-L. Multiple Attribute Decision Making: An Introduction; Sage Publications: New York, NY, USA, 1995; Volume 104. Yang, G.; He, S.; Shi, Z. Leveraging crowdsourcing for efficient malicious users detection in large-scale social networks. IEEE Internet Things J. 2017, 4, 330–339. [CrossRef] Zhou, L.; Wu, D.; Dong, Z.; Li, X. When Collaboration Hugs Intelligence: Content Delivery over Ultra-Dense Networks. IEEE Commun. Mag. 2017, 55, 91–95. [CrossRef] Liu, L.; Wang, Y.N.; Hou, L.; Feng, X.R. Easy encoding and low bit-error-rate chaos communication system based on reverse-time chaotic oscillator. IET Signal Process. 2017, 11, 869–876. [CrossRef] Yang, G.; He, S.; Shi, Z.; Chen, J. Promoting cooperation by the social incentive mechanism in mobile crowdsensing. IEEE Commun. Mag. 2017, 55, 86–92. [CrossRef] Li, X.; Wang, L.; Cui, J.; Zheng, B. A New Fragmentation Strategy for Video of HTTP Live Streaming. In Proceedings of the International Conference on Mobile Ad-Hoc and Sensor Networks (MSN), Hefei, China, 16–18 December 2016; pp. 86–89. Guo, L.; Chen, S.; Xiao, Z.; Zhang, X. DISC: Dynamic interleaved segment caching for interactive streaming. In Proceedings of the 25th IEEE International Conference on Distributed Computing Systems (ICDCS), Columbus, OH, USA, 6–10 June 2005; pp. 763–772. Costa, C.P.; Cunha, I.S.; Borges, A.; Ramos, C.V.; Rocha, M.M.; Almeida, J.M.; Ribeiro-Neto, B. Analyzing client interactivity in streaming media. In Proceedings of the ACM 13th International Conference on World Wide Web, New York, NY, USA, 17–20 May 2004; pp. 534–543. Wang, D.; Liu, J. A dynamic skip list-based overlay for on-demand media streaming with VCR interactions. IEEE Trans. Parallel Distrib. Syst. 2008, 19, 503–514. [CrossRef] Ullah, I.; Doyen, G.; Bonnet, G.; Gaiti, D. A survey and synthesis of user behavior measurements in P2P streaming systems. IEEE Commun. Surv. Tutor. 2012, 14, 734–749. [CrossRef] Choi, J.; Reaz, A.S.; Mukherjee, B. A survey of user behavior in VoD service and bandwidth-saving multicast streaming schemes. IEEE Commun. Surv. Tutor. 2012, 14, 156–169. [CrossRef] Azibi, R.; Vanderpooten, D. Construction of rule-based assignment models. Eur. J. Oper. Res. 2002, 138, 274–293. [CrossRef] © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
