A Scheduling Method for On-Demand Delivery of Selective ...... · 1 Introduction Due to the continued popularization of digital broadcasting services, on-demand delivery of selective

A Scheduling Method for On-DemandDelivery of Selective ContentsConsidering Selection Time

Yusuke Gotoh(B) and Ken Ohta

Okayama University, Okayama 7008530, [email protected]

Abstract. Due to the recent popularization of digital broadcasting ser-vices, on-demand delivery of selective contents on the Internet, i.e.,watching contents selected by the users themselves, is attracting muchattention. For example, in a quiz program, a user selects answer andwatches the video content to learn the answer. When the waiting timefor playing the data after the selection is lengthened, the continuity ofplaying them may be interrupted. In this paper, we propose a schedulingmethod to reduce the waiting time for the on-demand delivery of selec-tive contents based on the selection time. In our proposed method, theserver reduces the waiting time by delivering all the candidate contentsduring the selection time for watching subsequent content. Our evalua-tion confirmed that the average waiting time under our proposed methodwas reduced more than that under the conventional method.

Keywords: On-demand delivery · Scheduling · Selection timeSelective contents · Waiting time

1 Introduction

Due to the continued popularization of digital broadcasting services, on-demanddelivery of selective contents is attracting much attention. For example, in newsprograms, after watching a summary of each story, the user selects one of interestand watches it. In quiz programs, the server delivers several potential answersfrom which the user selects answer. If the answer is correct, he watches thevideo content about the correct answer. Otherwise, he watches the content ofthe incorrect answer.

By having access to selective contents, clients can watch programs that meettheir individual preferences. In such on-demand delivery systems as YouTube [1],the server delivers the contents based on the selections of clients. However, sincesome of the network bandwidth is wasted for each delivery, clients have to waitif insufficient bandwidth is provided. Here we assume that the server does notdeliver contents if the available bandwidth is less than the data consumptionrate. Therefore, users can watch the content without interruption after theystart watching it.c© Springer International Publishing AG, part of Springer Nature 2018O. Gervasi et al. (Eds.): ICCSA 2018, LNCS 10960, pp. 611–622, 2018.https://doi.org/10.1007/978-3-319-95162-1_42

http://crossmark.crossref.org/dialog/?doi=10.1007/978-3-319-95162-1_42&domain=pdf

612 Y. Gotoh and K. Ohta

Fig. 1. VoD environment

In this paper, we propose a scheduling method for the on-demand deliveryof selective contents to reduce the waiting times based on the selection times.Our proposed method reduces the waiting time by making a delivery scheduleto distribute all the candidate contents while selecting subsequent content.

The remainder of our paper is organized as follows. We explain deliverysystems for webcasts in Sect. 2 and introduce selective contents in Sect. 3. Thewaiting time for selective contents is explained in Sect. 4. Related works areintroduced in Sect. 5. Our proposed method is explained in Sect. 6 and evaluatedin Sect. 7. Finally, we conclude in Sect. 8.

2 Delivering System for Webcast

In webcasts, there are mainly two mainly types of delivery systems: Video onDemand (VoD) and broadcasting. First, we calculate the waiting times for VoDand broadcasting systems. Next, we explain waiting times for broadcasting datathat are divided into several segments.

The situation that causes waiting time in VoD systems is shown in Fig. 1.In the VoD system, the server starts delivering data sequentially based on clientrequests. The waiting times under VoD systems are roughly equal to the receivingtimes.

On the other hand, the situation that causes waiting times in broadcastingsystems is shown in Fig. 2. When the server repetitively broadcasts continuousmedia data, clients have to wait until the first portion of it is broadcast.

A Scheduling Method for On-Demand Delivery of Selective Contents 613

Fig. 2. Broadcasting environment

3 Selective Contents

3.1 Play-Sequence Graph

In this section, we explain a play-sequence graph [2], which is a state transitiongraph for describing the sequence of selective contents.

In a play-sequence graph, each node represents a state in which the clientplays some content. When the client finishes playing it, the state transits to thenext node. For example, a play-sequence graph for a quiz program is shown inFig. 3. In Fig. 3–A, the user selects an answer from given answers X or Y. NodeS1 is a state where the client plays a video that presents the quiz. S2 is a statewhere the client plays a video that explains answers X and Y. The user selectsan answer from X or Y while playing the video. If she selects answer X, thestate transits to S3. If she selects answer Y, the state transits to S4. In this way,the state transits to the next node based on her selections. When the user doesnot select an answer, the state transits to S2 again or automatically transits tosubsequent nodes S3 or S4. S3 is the state where the user selects answer X, anda video for the correct answer is played. S4 is the state where the user selectsanswer Y, and a video for the incorrect answer is played.

Play-sequence graphs can be simplified by applying the following three oper-ations: abbreviate, merge, and split [3]. By applying them, we can simplify theplay-sequence graph for a quiz program (Fig. 3–C).

3.2 Play-Sequence Graph Considering Selection Time

In this paper, we call the time for selecting one of the candidate contents afterthe user finishes watching the content the selection time. Based on Fig. 3–B, theplay-sequence graph based on the selection time is shown in Fig. 4. Si (i = 1, 2, 3)


Fig. 3. Simplified play-sequence graph

is the video data, and Ti is the state of selecting the next content after watchingSi. In Fig. 4, clients start playing T1 after 60 s of playing S1. When clients selectS2 or S3 during a maximum selection time (20 s), they play it within 60 s.

Fig. 4. Play-sequence graph considering selection time.

4 Mechanism for Waiting Time

In on-demand delivery of selective contents, the server needs to deliver con-tents effectively based on the client’s selection. Many researchers have proposedscheduling methods to reduce waiting times considering the available bandwidth,the amount of contents, and the structure of play-sequence graphs.

In on-demand delivery, we explain the scheduling method to reduce waitingtimes. In the VoD system, when the server makes a delivery schedule called aVoD method, the server distributes contents to the client according to its ownrequirements. For example, the delivery schedule for selective contents is shownin Fig. 5. The server delivers a news program whose play-sequence graph is shown


in Fig. 6. The playing time of S1, · · · , S7 is 60 s, and that of T1, T2, and T3 is 20 s.The consumption rate is 5.0 Mbps, and the available bandwidth is 8.0 Mbps.

In the VoD method, the server delivers contents selected by the clients. There-fore, the server does not deliver the data between finishing the delivery of St andstarting to deliver the next content. In Fig. 5, when the client plays S1, S2, andS4 sequentially and selects the next content within 10 s, the server does not useall of the bandwidth for delivering contents within 40 s.

Fig. 5. Delivery schedule under VoD method

5 Related Works

Several methods have been proposed to reduce waiting times for broadcastingsystems [4–10]. The Harmonic Broadcasting (HB) method [11] divides the datainto several segments of equal sizes and broadcasts them to reduce the waitingtime. By broadcasting divided data repetitively by each channel, the waitingtime is reduced.

In the Optimized Periodic Broadcast (OPB) method [12], each segment ofdata is divided into two parts. After clients completely receive all of the precedingparts of the content, they start to receive the next part. Since clients can get thepreceding parts in advance, the waiting time is reduced. However, the bandwidthincreases as the number of segments increases.

In the Hierarchical Stream Merging (HSM) method [13], after clients havecompletely received all of the data, the server merges the channel that the client


Fig. 6. Play-sequence graph for news program

is using. Since the method reduces the necessary bandwidth, the waiting time isreduced.

In the PolyHarmonic Broadcasting Protocol with Partial Preloading (PHP-PP) method [14], the data are divided into 20 segments. The number of segmentsis derived from the prefetching ratio and the bandwidths of the channels. Thenecessary bandwidth becomes 17.7 Mbps, which is less than the simple repetitionmethod.

In Mayan Temple Broadcasting (MTB) [15], the data are divided into fivesegments. The necessary bandwidth becomes 21.3 Mbps. Although the necessarybandwidth is larger than that under the PHB-PP method, the number of chan-nels needed to broadcast segments is fewer than under the PHB-PP method.Therefore, the implementation is easier than the PHB-PP method.

These schemes use a near-video-on-demand technique, i.e., reducing the wait-ing times by repetitively broadcasting the data. In this paper, we assume thatthe server does not repetitively broadcast the data. In addition, our assumeddata are selective contents.

We previously proposed scheduling schemes to reduce the waiting times forselective contents broadcasting [2,16]. In these schemes, by acquiring channelbandwidth that is identical as the data consumption rate, the waiting time iseffectively reduced. In this paper, we assume that the server creates a deliveryschedule to distribute all the candidate contents while selecting the next content.

6 Proposed Method

6.1 Outline

We propose a scheduling scheme for on-demand delivery called the ContentsCumulated Broadcasting Considering Selection Time (CCB-ST) method, which


reduces waiting times by making a delivery schedule for distributing all candidatecontents while selecting the next content.

6.2 Assumed Environment

Our assumed system environment is listed below:

– The data are selective contents.– The server provides an on-demand delivery system.– The network is a guarantee type.– The server only delivers content selected by the client.– Clients have enough buffer to store the received data.

6.3 Scheduling Process

Let n be the number of states, and let b be the depth of play-sequence graph. Si

(i = 1, · · · , n) is the state of playing the content. Ti (i = 1, · · · , n) is the stateof selecting the next content. The available bandwidth of the server is B, theamount of the stored contents on queue L is l, and the number of selections is e.

In the proposed method, in all the routes from the roots to the leaves, theserver finds the longest route. The scheduling process continues as follows:

1. Insert the order of the early start time. If the start times of several contentsare equivalent, they are inserted in a lower number of n.

2. The server removes the top content from L. We indicate the removed contentby Si.

3. Schedule Si and Ti to Cj .4. Schedule Si+1, · · · , Si+e to C1, · · · , Ce during the selection time.5. The client sends information about its selected content to the server.6. From L, the server removes the other contents at the same depth and the

contents on the routes that the client does not select.7. If L is not empty, repeatedly go to process 2.

6.4 Practical Example

An example of a delivery schedule produced by the proposed method is shownin Fig. 7. We use the play-sequence graph shown in Fig. 6. Clients sequentiallyselect S1, S2, and S4. In our proposed scheduling method, when the availablebandwidth is 8.0 Mbps and the consumption rate is 5.0 Mbps, the selection timeis 10 s. As shown in Subsect. 6.3, while the client selects either S2 or S3 after itfinishes playing S1, the server delivers S2 and S3 by 4.0 Mbps, which takes 20 s.When the client selects S2, the server delivers the rest of it by 8.0 Mbps, whichtakes 27.5 s. While the client selects either S4 or S5 after it finishes playing S2,the server delivers S4 and S5 by 4.0 Mbps, which takes 30 s. When the clientselects S4, the server delivers the rest of it by 8.0 Mbps, which takes 22.5 s.Therefore, the CCB-ST method reduces the delivery time (162.5 s) more thanthe VoD method (177.5 s), and no waiting time occurred.


Fig. 7. Delivery schedule under proposed method

7 Evaluation

7.1 Evaluation Environment

According to delivery programs, play-sequence graphs have many patterns. How-ever, evaluating the performance of our proposed method for all of these patternsis not realistic. Therefore, in this paper, we use the play-sequence graph shownin Fig. 8.

The number of potential answers e is 2, the playing time of Si is 60 s, andthat of Ti is 20 s. The consumption rate is 5.0 Mbps.

7.2 Effect of Arrival Interval

We evaluate the waiting time under several average arrival intervals. The resultis shown in Fig. 9. The horizontal axis is the average arrival time that is set every2 s in the 5 to 25 s range. The vertical axis is the average waiting time dividedby the number of clients.

In this evaluation, we used the play-sequence graph shown in Fig. 8. Thenumber of contents is 7 and the number of clients is 100. The available bandwidthis 100 Mbps, and the consumption rate is set every 0.1 Mbps in the 5.0 to 10 Mbpsrange. The selection time is 10 s after it finishes playing the content. In this paper,we assume that the arrival pattern of the user is independent of other users.Therefore, the simulation model used in the paper follows a Poisson distribution.


Fig. 8. Play-sequence graph for evaluating waiting times

Fig. 9. Average arrival interval and waiting times

In Fig. 9, as the average arrival interval increases, the waiting times underthe proposed and VoD methods are reduced further. When the arrival intervalis lengthened, since the data size of the contents delivered by the server to theclients increases, the waiting time is reduced.


In addition, the waiting time in the CCB-ST method is shorter than theVoD method. In the CCB-ST method, since the delivery time of the program isreduced by distributing the candidate contents while selecting the next content,the waiting time can be reduced.

7.3 Effect of Selection Time

Next we evaluated the waiting time under several selection times. The result isshown in Fig. 10. The horizontal axis is a selection time that is set every sec.in the 0 to 20 s range. The vertical axis is the average waiting time divided bythe number of clients. The number of contents is 7, and the number of clientsis 100. The available bandwidth is 100 Mbps, and the consumption rate is setevery 0.1 Mbps in the 5.0 to 10 Mbps range. The average arrival interval basedon a Poisson distribution is 15 s.

In Fig. 10, as the average arrival interval increases, the waiting times underthe proposed and VoD methods are lengthened. When the selection time islengthened, the delivery time for each client is also lengthened. In the CCB-ST method, since the time that the server is unable to allocate bandwidth tothe clients is reduced, the waiting time can be reduced.

Fig. 10. Selection and waiting times

8 Conclusion

In this paper, we proposed and evaluated a scheduling method named the CCB-ST method for selective contents that considers selection time. The CCB-ST


method reduces the waiting time by making a delivery schedule to distribute allthe candidate contents while selecting the next content. In our evaluation, weconfirmed that the waiting times are reduced more than the conventional VoDmethod.

A future direction of this study will make a scheduling method to considerthe transition probability of play-sequence graphs.

Acknowledgement. This work was supported by JSPS KAKENHI Grant Num-ber 18K11265 and 16K01065. In addition, this work was partially supported by theTelecommunications Advancement Foundation.

References

1. YouTube. http://www.youtube.com/2. Gotoh, Y., Yoshihisa, T., Kanazawa, M.: A scheduling method considering available

bandwidth in selective contents broadcasting. In: Proceedings of the IEEE WirelessCommunications and Networking Conference, WCNC 2007, pp. 2597–2602 (2007)

3. Yoshihisa, T.: A scheduling method for bandwidth reduction on selective contentsbroadcasting. In: Proceedings of the IPSJ International Conference on Mobile Com-puting and Ubiquitous Networking, ICMU 2006, pp. 60–67 (2006)

4. Janakiraman, R., Waldvogel, M.: Fuzzycast: efficient video-on-demand over multi-cast. In: Proceedings of the IEEE INFOCOM, pp. 920–929 (2002)

5. Jinsuk, B., Jehan, F.P.: A tree-based reliable multicast scheme exploiting the tem-poral locality of transmission errors. In: Proceedings of the IEEE International Per-formance, Computing, and Communications Conference, IPCCC 2005, pp. 275–282(2005)

6. Juhn, L.-S., Tseng, L.M.: Fast data broadcasting and receiving scheme for popularvideo service. IEEE Trans. Broadcast. 44(1), 100–105 (1998)

7. Paris, J.-F.: An interactive broadcasting protocol for video-on-demand. In: Pro-ceedings of the IEEE International Performance, Computing, and CommunicationsConference, IPCCC 2001, pp. 347–353 (2001)

8. Viswanathan, S., Imilelinski, T.: Pyramid broadcasting for video on demand ser-vice. In: Proceedings of the SPIE Multimedia Computing and Networking Confer-ence, MMCN 1995, pp. 66–77 (1995)

9. Tantaoui, M., Hua, K., Do, T.: BroadCatch: a periodic broadcast technique forheterogeneous video-on-demand. IEEE Trans. Broadcast. 50(3), 289–301 (2004)

10. Hua, K.A., Bagouet, O., Oger, D.: Periodic broadcast protocol for heterogeneousreceivers. In: Proceedings of the MMCN, pp. 220–231 (2003)

11. Juhn, L.-S., Tseng, L.: Harmonic broadcasting for video-on-demand service. IEEETrans. Broadcast. 43(3), 268–271 (1997)

12. Carlsson, N., Mahanti, A., Li, Z., Eager, D.: Optimized periodic broadcast of non-linear media. IEEE Trans. Multimed. 10(5), 871–884 (2008)

13. Zhao, Y., Eager, D.L., Vernon, M.K.: Scalable on-demand streaming of non-linearmedia. In: Proceedings of the IEEE INFOCOM, vol. 3, pp. 1522–1533 (2004)

14. Mahanti, A., Eager, D.L., Vernon, M.K., Stukel, D.S.: Scalable on-demand mediastreaming with packet loss recovery. IEEE/ACM Trans. Netw. 11(2), 195–209(2003)

http://www.youtube.com/


15. Paris, J.-F., Long, D.D.E., Mantey, P.E.: Zero-delay broadcasting protocols forvideo-on-demand. In: Proceedings of the ACM International Multimedia Confer-ence, Multimedia 1999, pp. 189–197 (1999)

16. Gotoh, Y., Yoshihisa, T., Kanazawa, M., Takahashi, Y.: A broadcasting scheme forselective contents considering available bandwidth. IEEE Trans. Broadcast. 55(2),460–467 (2009)

A Scheduling Method for On-Demand Delivery of Selective ...... · 1 Introduction Due to the continued popularization of digital broadcasting services, on-demand delivery of selective

Documents