An adaptive harmonic broadcasting scheme with ...

4

Int. J. Multimedia Intelligence and Security, Vol. 2, No. 1, 2011

An adaptive harmonic broadcasting scheme with download and playback synchronisation for video on demand Mohammad Saiedur Rahaman* Department of Computer Science, American International University Bangladesh, Banani, Dhaka – 1213, Bangladesh E-mail: [email protected] *Corresponding author

Ashfaqur Rahman Centre for Intelligent and Networked Systems, Central Queensland University, QLD, Australia E-mail: [email protected]

Sumaira Tasnim Power Engineering Research Group, Central Queensland University, QLD, Australia E-mail: [email protected] Abstract: Harmonic broadcasting is a broadcasting scheme to reduce the bandwidth requirement in video-on-demand service where a video is divided into segments of equal size and each segment is repeatedly transmitted over a number of channels having bandwidth that follows a harmonic series. As the bandwidth of channels differ from each other and a user can join at any time to these multicast channels, a synchronisation problem between download and playback is created. To deal with this issue, some schemes have been proposed, however, at the cost of additional or wastage of bandwidth. In this paper, we propose an adaptive harmonic broadcasting scheme that deals with the synchronisation problem at the same average bandwidth consumption as traditional harmonic broadcasting scheme. We present several analytical results to exhibit the efficiency of the proposed scheme over the existing ones. Keywords: VOD; harmonic broadcasting scheme; download; playback synchronisation. Reference to this paper should be made as follows: Rahaman, M.S., Rahman, A. and Tasnim, S. (2011) ‘An adaptive harmonic broadcasting scheme with download and playback synchronisation for video on demand’, Int. J. Multimedia Intelligence and Security, Vol. 2, No. 1, pp.4–17.

Copyright © 2011 Inderscience Enterprises Ltd.

An adaptive harmonic broadcasting scheme

5

Biographical notes: Mohammad Saiedur Rahaman received his MSc in Computer Science in 2009 and his BSc in Computer Science and Engineering in 2007 from the American International University – Bangladesh (AIUB). Currently, he is working as a Lecturer at the Department of Computer Science in the same university. His research interests include wireless mesh networks, wireless sensor networks, grid computing, mobile and multimedia networks and data mining. Ashfaqur Rahman was with the Department of Computer Science AIUB when this research was conducted. Currently, he is with Centre for Intelligent and Networked Systems at the Central Queensland University, Australia. He has more than 30 peer reviewed journal articles, conference papers and book chapters. Sumaira Tasnim is with the Power Engineering Research Group of Central Queensland University. She received her BSc from the Bangladesh University of Engineering and Technology in 2003. She is now working as a postgraduate research student at the CQUni.

1

Introduction

The concept of a general video on demand system (Juhn and Tseng, 1997, 1998; Yang et al., 1999; Paris et al., 1998; Segarra and Cholvi, 2007; Om and Chand, 2007; Engebretsenl and Sudan, 2006) is like that upon receiving a request for a specific video from clients, the VOD system provides each client with a dedicated server channel and the clients get the service instantly. In this system, a client has special control over the server channel because the server is bound to serve at once after getting a request from client and so this type of service is named as true video on demand (TVOD). Here the load on the server depends on the viewer’s arrival rate. As the viewer arrival rate increases, the load on the server side increases linearly. So, TVOD schemes have serious bottleneck at the server-side when the viewer arrival rate is high. So, it can be said that TVOD systems are not scalable and thus, not suitable for serving a large number of clients and each client is requesting asynchronously for a video of his choice. On the other hand, when a VOD service provider is going to serve a large number of clients from their remote server using dedicated links, they should think about the high expense for maintaining those links. Again, the clients may suffer from congestion, delay and jitter as they are served from a distant place. So, research (Azad and Murshed, 2005) is going on to efficiently reduce the load on server and to eliminate the effects of transmission delay and jitter on the client side while reducing the transmission cost. Near video on demand (NVOD) is another VOD service technique where the main goal is to reduce the server bandwidth requirement by forcing a group of viewers requesting the same video to share a single stream by batching their requests over a specified interval. These techniques introduce a short waiting time on the client side. Since here the viewers are non-interactive they cannot watch the video immediately. This is the reason why it is called NVOD. Harmonic broadcasting (HB) scheme is a NVOD scheme where a video is divided into some equal segments and broadcasted through some channels with channel

6

M.S. Rahaman et al.

bandwidth following the harmonic series. But it cannot deliver some segments when necessary and leads to download and playback synchronisation problem. To eliminate this problem two better broadcasting protocols have been proposed in Paris et al. (1998) also known as cautious harmonic broadcasting (CHB) and quasi-harmonic broadcasting (QHB). These two schemes eliminate the synchronisation problem at the cost of additional bandwidth and transmitting some extra segments respectively compared to traditional HB. In this paper, we propose a modified HB scheme that eliminates the download and playback synchronisation problem while keeping the average bandwidth consumption the same as traditional HB scheme. The proposed algorithm adapts itself with the download and playback synchronisation problem and that’s why we call it the adaptive harmonic broadcasting (AHB) scheme. The paper is organised as follows. Section 2 highlights some related works. Section 3 presents the proposed AHB scheme, its features and comparison with other existing methods. Some analytical results are presented in Section 4. Finally, Section 5 concludes the paper.

2

Related works

2.1 HB scheme HB (Juhn and Tseng, 1997, 1998; Yang et al., 1999; Segarra and Cholvi, 2007; Om and Chand, 2007; Engebretsen, and Sudan, 2006) is a pioneer NVOD scheme where a video of size S and length T is partitioned into N segments of equal sizes and periodically broadcasts each segment on a dedicated channel. However, successive channels have decreasing bandwidths following the harmonic series. Assuming a playback rate of b, the first channel broadcast segment one at bandwidth b, the second segment broadcasts segment two at bandwidth b/2, the third channel broadcast segment three at bandwidth b/3 and so on (Figure 1). The client downloads the segments from all the channels concurrently and the segments that are downloaded in advance are stored at a local buffer. The maximum client waiting time in HB scheme is w = Δt =

T N

(1)

and the server bandwidth requirement of HB in a time slot of length Δt is N

BHB = b ×

∑ 1i = b × H ( N )

(2)

i =1

where H(N) is the harmonic number of N. One major flaw in HB scheme is that it cannot deliver some segments on time in some situations (Segarra and Cholvi, 2007; Om and Chand, 2007). For example, in Figure 1, a client who starts at t2 will finish the download and playback of the first segment at t3 at bandwidth b. At that time the client has only the second half of the second segment in its local buffer. If the client starts downloading the first half of the second segment at t3 at bandwidth b/2, it will not be able to play the video at bandwidth b. This is called the download and playback synchronisation problem. A simple display

An adaptive harmonic broadcasting scheme

7

delay equal to the size of the first segment can resolve this problem, however, at the cost of making the client wait in the middle of watching a video which is unacceptable. Figure 1

The HB scheme with video V divided into segments S1, S2, ..., SN and segments are transmitted on different channels at decreasing bandwidth

Note: In a time slot Δt the total server bandwidth requirement is b × H(N).

2.2 Cautious broadcasting scheme In Paris et al. (1998), Carter and Long proposed a new harmonic broadcasting protocol called CHB which resolves the abovementioned synchronisation problem at the cost of some additional bandwidth. Assuming a video of size S divided into N segments, the CHB protocol (Figure 2) broadcasts its two first channels at full bandwidth b. Channel one repeatedly broadcasts segment S1 while channel two broadcasts segments S2 and S3. Channel two broadcasts more than one segment as subsequent segments are broadcast on separate channels at decreasing bandwidths. In particular, segment Si with 4 ≤ i ≤ N is broadcast at channel i – 1 at bandwidth b / (i – 1). This new organisation of video segments guarantees that the customer will either receive a segment at full bandwidth when it is needed, or have the entire segment already in its buffer before it is needed. So, the bandwidth requirements of the CHB protocol are given by: N −1

BCHB = (b + b) +

∑bi i =3

N −1

= 2b +

∑bi −b − b2

(3)

i =1

= b + b × H ( N − 1) 2

where b is the video playback rate, N is the total number of segments and H(N – 1) is the harmonic number of N – 1.

8

M.S. Rahaman et al.

Although CHB solves the synchronisation problem, a major drawback of this scheme is that the first two channels are of full bandwidth and thus the bandwidth consumption is higher than HB as BCHB = b + b × H ( N − 1) 2 = b + b × H (N ) − 1 N 2 b b = b × H (N ) + − N 2 = BHB + b − b N 2

(

(

(

)

) )

(4)

and (b/2 – b/N) > 0 for N > 2. Figure 2

Delivery of segments in CHB

Notes: Considering N segments, the first channel broadcasts segment S1 at playback rate b, the second channel repeatedly broadcasts segments S2 and S3 at bandwidth b whereas the third channel broadcasts segments S4 at bandwidth b/3 and in general segment i(4 ≤ i ≤ N) is served with bandwidth b/(i – 1).

2.3 Quasi-harmonic broadcasting Like other HB protocols, QHB also divides each video into N equal segments and broadcasts the first segment repeatedly on the first channel. But unlike others, the QHB scheme (Figure 3) divides each segment i (where 2 ≤ i ≤ N) into im – 1 fragments for some fragmentation parameter m, and the client receives m fragments from each channel per time slot (a time slot equals the duration of a segment at playback rate b). If each time slot is divided into m equally sized sub slots, then the client receives a single fragment during each sub slot. The most important thing regarding QHB is how the fragments are organised in each channel. Consider any channel i in QHB scheme. The last sub slot of

An adaptive harmonic broadcasting scheme

9

each time slot is used to transmit the first i – 1 fragments of Si in order. The rest of the sub slots transmit the other i × (m – 1) fragments such that the kth sub slot of slot j is used to transmit fragment (ik + j – 1)mod (i(m – 1)) + i (Figure 3). But the problem of this scheme is the repeated transmission of the first fragment of any segment which is a wastage of bandwidth compared to traditional HB. Figure 3

3

An illustration of the first three channels for a video in QHB scheme with m = 4

Proposed scheme and its features

Our proposed modification to the existing HB scheme aims to deal with the synchronisation problem by 1

ensuring timely delivery of segments leading to synchronisation between download and playback so that a client need not wait in the middle of watching a video

2

consuming the same amount of bandwidth as the HB scheme.

As the modified scheme adapts itself to synchronisation problem without increasing the average bandwidth requirement we prefer to name it AHB scheme. In the rest of this section we first present modified HB scheme and then some of its features.

3.1 Adaptive harmonic broadcasting Assuming a video of length T and the playback/consumption rate of the video b (i.e., the size of the video is S = T × b), the proposed AHB scheme (Figure 4) involves the following steps: 1

The video is equally divided into N segments, where N is a positive integer. Suppose Sk is the kth segment of the video where 1 ≤ k ≤ N. So, the concatenation of all the segments, in order of increasing segment numbers, constitutes the video, i.e., S = S1 i S2 i … i S N .

2

The kth segment of the video, Sk is equally divided into k fragment(s) and thus Sk = Sk1 i Sk2 i … i Skk where S ki denotes the ith fragment of the kth segment.

3

The k fragments of Sk are placed on channel Ck which is a variable bit rate channel. The bandwidth allocated to Ck for a normal fragment is b/k, whereas for a time slot that serves a problematic fragment (i.e., fragment with synchronisation problem) the bandwidth allocation for Ck is b (Figure 4) for a duration of Δt/k where Δt denotes

10

M.S. Rahaman et al. the duration of segment S1. In AHB scheme, a problematic fragment i for segment k, S ki starts at kt + (i – 1) where, k > 1, i ∈{1, 2, ..., k – 1}. The proof that problematic segments start at kt + 1 (i – 1) is deferred until the end of this section.

4

Within channel Ck the k fragment(s) of segment Sk will be broadcasted periodically (Figure 4).

Figure 4

Delivery of fragments in the proposed AHB scheme with N = 4 segments

Note: The problematic segments on Channel 2, 3 and 4 are shaded light gray that are served at bandwidth b.

The proposed AHB scheme eliminates the problem regarding download and playback synchronisation in HB. As shown in Figure 4, a client who starts at t2, will finish the playback of the first segment at the beginning of t3. At that time the client has only the second half of the second segment in its local buffer. Using HB scheme (Figure 1), the client starts downloading the first half of the second segment at time slot two at bandwidth b/2 and will not be able to playback at rate b. But using our proposed AHB scheme (Figure 4) the client can start downloading the first half of the second segment at time slot two at bandwidth b and can playback at the same rate and the client will not face any lack of synchronisation between download and playback. So, we can claim that our proposed AHB scheme delivers all data on time. But we need a proof for the claim. To prove the claim we first analyse the problematic fragments of a segment those will be served at full bandwidth. From observations in Figure 4 a relationship between a segment and its problematic fragments are summarised in Table 1.

An adaptive harmonic broadcasting scheme Table 1

11

Problematic fragments of different segments (for five segments and their fragments)

Problems with segments

Associated problematic fragments

S1

No problem

S2

S 21

S3

S31 , S32

S4

S 41 , S 42 , S43

S5

S51 , S52 , S53 , S54

In order to serve the problematic fragments at full bandwidth, we also need to know the starting time of these fragments. After analysing Figure 4, we observe a relationship between problematic segments and their starting times as presented in Table 2. A close observation from Table 2 shows that a problematic fragment Ski is needed to be served at bandwidth b at time slots beginning at kt +(i – 1) for a duration Δt/k unit time where k > 1, i ∈{1, 2, ..., k – 1} and t ∈ {1, 2, ..., ∞}. Table 2

Times when we need to serve a fragment at full bandwidth b (for five segments and their fragments)

Problematic fragments

Needs b bandwidth at time

Duration (unit)

S 21

2, 4, 6, 8, …

Δt/2

S31

3, 6, 9, 12, …

Δt/3

S32

4, 7, 10, 13, …

Δt/3

S 41

4, 8, 12, 16, …

Δt/4

S 42

5, 9, 13, 17, …

Δt/4

S 43

6, 10, 14, 18, …

Δt/4

S51

5, 10, 15, 20, …

Δt/5

S52

6, 11, 16, 21, …

Δt/5

S53

7, 12, 17, 22, …

Δt/5

S54

8, 13, 18, 23, …

Δt/5

Note: Here Δt = duration of segment S1

Theorem: AHB delivers all video fragments on time. Proof: From the observation we see that serving a problematic segment S ki at kt + (i – 1)with bandwidth b solves our problem regarding synchronisation. So, it is sufficient to prove that the ith fragment of segment k will be served at kt + (i –1). Let’s consider the following observation for segment 4 from Tables 1 and 2. We know that patch channel of bandwidth b is needed for S 41 at time slots starting at

12

M.S. Rahaman et al.

4, 8, 12, 16,... for a duration of time Δt/4. Again, patch channel of bandwidth b is needed for S 42 at time slots starting at 5, 9, 13, 17,... for a duration of time Δt/4. Again, patch channel of bandwidth b is needed for S 43 at time slots starting at 6, 10, 14, 18,... Up to the server is active and at a duration of time Δt/4. So, in general, we can say that problematic fragment 1 of segment k needs to be served at

τ (k , i) = kt

(5)

where t∈{1,2, ...,∞). Similarly, problematic fragment 2 of segment k needs to be served at k + 1, and problematic fragment 3 of segment k needs to be served at k + 2 and so on. With this observation we can generalise the time for serving the fragment i of segment k at full bandwidth as

τ (k , i ) = τ (k , i − 1) + 1

(6)

where i ∈ {1, 2, ..., k – 1). This is the recurrence relation. It gives a boundary value and an equation for the general value in terms of earlier ones. Now we need a closed form for τ(k, i) for finding the solution. Our intension is to look for the smaller cases. Now observe the followings, at i = 1, τ(k, i) = kt + 0; at i = 2, τ(k, i) = kt + 1; at i = 3 τ(k, i) = kt + 2; ...; and in general

τ (k , i ) = kt + i − 1

(7)

where i ∈ {1, 2, ..., k – 1) and is the closed form for recurrence τ(k, i). We have expressed the service delivery time τ as a function of k and i where t ∈ {1, 2, ..., ∞). Now we prove this recurrence by induction. Basis: At i = 1, τ(k, i) = kt, where t ∈ {1, 2, ..., ∞). From Table 2 it follows for S 21 , S31 , S 41 ,… Induction: Let the induction holds for up to i – 1, i.e., τ (k, i – 1) = kt + i – 2 and we have to prove that it holds for i. We can write

τ ( k , i ) = τ (k , i − 1) + 1 = kt + (i − 2) + 1 = kt + (i − 1)

(8)

Hence, (7) holds for all i as well. Thus, we can claim that the ith fragment of segment k, S ki will be served at kt + (i – 1) and AHB delivers all data segment on time.

3.2 Features of AHB Alongside solving the synchronisation problem AHB has some salient features as presented below in this section. 1

Initial client waiting time: the initial waiting time in the worst case is proportional to the duration of segment S1, i.e., T/N where a video of length T is partitioned into N segments.

2

Bandwidth consumption in a particular time slot: the bandwidth consumption BAHB (k) in AHB scheme (Figure 4) for channel k in a time slot of length Δt is

An adaptive harmonic broadcasting scheme

13

BAHB (k ) = b / k

(9)

for a regular fragment and BAHB ( k ) =

b × ( Δt / k ) + 0 × ( Δt − Δt / k ) =b/k Δt

(10)

for a problematic segment. In general N

BAHB =

∑

N

BAHB ( k ) =

k =1

∑

N

b/k =b ×

k =1

∑1 / k = b × H ( N ) = B

(11)

HB

k =1

The last equality in the previous equation follows from (2). It can thus be observed that AHB schemes consumes the same average bandwidth as HB scheme while solving the synchronisation problem. Note that CHB consumes more bandwidth than HB whereas QHB wastes some bandwidth thus making AHB superior to them. 3

Worst case bandwidth consumption: in Figure 4, we find that a client joining at time slot four needs to download from all the channels at full bandwidth and the server needs to assign all the channels as full bandwidth channel. But it happens for a very short period of time. However, this is the price we need to pay in order to achieve timely delivery of video segments in HB.

4

Worst case storage requirement: in Figure 4, we observe that a client joining at time slot four needs to download from all the channels at full bandwidth and needs greater storage capability. But it happens for a very short period of time. With storage solutions approaching tera-bytes this is not a big issue.

5

Waiting time for discontinued fragment: as there remains no discontinued frame, a client need not wait in the middle of watching a video eliminating the problem in HB scheme. Hence the waiting time for discontinued fragments is zero.

Table 3

Comparison of AHB scheme with traditional HB, CHB and QHB scheme

Criteria Bandwidth consumption in a particular duration Worst case BW consumption

Initial client waiting time Worst case storage requirement

Waiting time for discontinued fragments

AHB vs. HB scheme

AHB vs. CHB scheme

AHB vs. QHB scheme

Equal

Less in AHB

Less in AHB

AHB consumes more than HB but for a very short amount of time

AHB consumes more than CHB but for a very short amount of time

AHB consumes more than QHB but for a very short amount of time

Equal

Equal

Equal

AHB consumes more than HB but for a very short amount of time

AHB consumes more than CHB but for a very short amount of time

AHB consumes more than CHB but for a very short amount of time

No waiting time in AHB

No waiting time in AHB and CHB

No waiting time in AHB and QHB

14

M.S. Rahaman et al.

3.3 Comparison with existing HB scheme and its variants In this section, we provide a comparison among the proposed AHB scheme, the traditional HB scheme, QHB scheme and CHB scheme considering initial client waiting time (Kim and Park, 2008; Jeong et al., 2008), buffer requirement, etc. From previous section, we can see that the proposed scheme solve the download and playback synchronisation problem without additional average bandwidth as done in CHB and QHB. The following table provides a comparative picture between HB, CHB, QHB and AHB schemes.

4

Discussions

4.1 Bandwidth consumption in a particular time slot From Figure 5, we see that in a specific time slot, the bandwidth consumption of the proposed AHB scheme is equal to the HB scheme whereas less than QHB and CHB scheme. In some cases, our scheme needs to serve all the channels at full bandwidth. But it is for a very short time and our scheme consumes no sever bandwidth in some cases. When the bandwidth consumption is zero, additionally we can allocate the server’s free channels for other purposes. Figure 5

Comparison of bandwidth consumption between HB, QHB, CHB and AHB scheme (see online version for colours)

An adaptive harmonic broadcasting scheme

15

4.2 Waiting time for discontinued fragments Figure 6 shows the waiting time for lack of download and playback synchronisation in traditional HB scheme, QHB scheme, CHB scheme and proposed AHB scheme. It is zero for CHB, QHB and AHB scheme because there is no lack of synchronisation in these schemes. But in traditional HB scheme the waiting time is related to number of segments. As the number of video segments increases in HB, the waiting time for lack of synchronisation decreases. The simulation shows for a 120 minutes video that the video with one segment the client needs to wait for 120 minutes, whereas for the video with two, three, and five segments the waiting time is reduced and comes down to 60 minutes, 40 minutes, 30 minutes and 20 minutes respectively. Figure 6

Comparison of client waiting time between HB, QHB, CHB and AHB scheme for discontinued segments in a 120 minutes video with number of segments = 5 (see online version for colours)

4.3 Worst case bandwidth consumption From Figure 7, a careful examination for five channels shows high bandwidth transmission at the beginning of a time slot in extreme cases and consequently, very low or even only b bandwidth transmission at the end of the same time interval. Although the average bandwidth remains the same, our proposed AHB scheme requires more bandwidth in extreme cases. For example, at the beginning of time slot four in Figure 4, the AHB requires 4 b bandwidth compared to b × H(4) = 1.81 b for the original HB scheme, b2.58 for CHB scheme and 2.20 for QHB scheme. But after the one third duration of a time slot, the bandwidth requirement is much less in AHB compared to other schemes. However, this is the price that we need to pay in extreme cases in order to achieve timely delivery of video segments in HB.

16

M.S. Rahaman et al.

Figure 7

Comparison of worst case bandwidth consumption between HB, QHB, CHB and AHB scheme when number of channel = 5 (see online version for colours)

Figure 8

Comparison of worst case storage requirement between HB, QHB, CHB and AHB scheme five channels (see online version for colours)

An adaptive harmonic broadcasting scheme

17

4.4 Worst case storage requirement From Figures 4 and 8, a careful examination shows that high storage requirement at the client side is needed at the beginning of extreme time slot where the server is serving all the channels at full bandwidth. Very low or even only b storage requirement is needed at the end of the same time interval which is much less than the traditional HB scheme.

5

Conclusions

In this paper, we have presented an AHB scheme to resolve the problem of intermediate client waiting time by synchronising between video download and playback rates. Compared to traditional HB scheme it consumes the same average bandwidth whereas other modifications consume more bandwidth (e.g., CHB and QHB). The worst case bandwidth requirement of AHB, however, is higher than others. Another synchronisation problem arises when number of channels change as it creates a correspondence problem between segment sizes (Engebretsen and Sudan, 2006). While this problem is dealt with by other VOD broadcasting schemes, it is still not explored for HB scheme. In future we aim to extend the AHB scheme to deal with the channel transition problem.

References Azad, S.A. and Murshed, M.M. (2005) ‘Seamless channel transition for popular video broadcasting’, International Conference on Information Technology: Coding and Computing, 2005, April, Vol. 2, pp.283–288. Engebretsen, L. and Sudan, M. (2006) ‘Harmonic broadcasting is bandwidth-optimal assuming constant bit rate’, Networks, May, Vol. 47, No. 3, pp.172–177. Jeong, H., Joe, S., Park, S., Song, S. and Yun, D. (2008) ‘Interleaving harmonic staggered broadcasting scheme for video-on-demand services’, Tenth IEEE Int. Symp. on Multimedia, pp.202–207. Juhn, L-S. and Tseng, L-M. (1997) ‘Harmonic broadcasting protocols for video-on demand service’, IEEE Transactions on Broadcasting, September, Vol. 43, pp.268–271. Juhn, L-S. and Tseng, L-M. (1998) ‘Enhanced harmonic data broadcasting and receiving scheme for popular video service’, IEEE Transactions on Consumer Electronics, March, Vol. 44, No. 2, pp.343–346. Kim, H-I. and Park, S-K. (2008) ‘A hybrid video-on-demand data broadcasting and receiving scheme of harmonic and staggered schemes’, IEEE Trans. on Broadcasting, December, Vol. 54, No. 4, pp.771–778. Om, H. and Chand, S. (2007) ‘Geometrico-harmonic broadcasting scheme with continuous redundancy’, IEEE Transaction on Multimedia, February, Vol. 9, No. 2, pp.410–419. Paris, J-F., Carter, S.W. and Long, D.D.E. (1998) ‘Efficient broadcasting protocols for video on demand’, Proceedings, Sixth International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems, July, Vol. 19, No. 24, pp.127–132. Segarra, J. and Cholvi, V. (2007) ‘Convergence of periodic broadcasting and video-on-demand’, Comp. Comm., March, Vol. 30, No. 5, pp.1136–1141. Yang, Z-Y., Juhn, L-S. and Tseng, L-M. (1999) ‘On optimal broadcasting scheme for popular video service’, IEEE Transactions on Broadcasting, September, Vol. 45, No. 3, pp.318–322.