JP2002007257A - Method for scheduling content delivery - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a content delivery scheduling method which minimizes the mean guarantee wait time, while guaranteeing delivery completion time to a user, when delivery is requested. SOLUTION: All contents are divided into groups according to selection probability and FIFOs are prepared by the groups; and when content delivery is requested, it is decided as to whether requested contents are requested by other users and placed in a delivery wait state. When it is decided that the request is a new request, the request is put in the tail of the FIFO prepared for the group that the contents belong to, and when the delivery of the contents are completed and a line becomes free, contents to be delivered next are circulating among the heads of FIFOs on a round-robin basis.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a content distribution system for directly delivering digital content stored in a server to a receiving device at a user's home in response to a request from a user via a shared bandwidth network. BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a content delivery scheduling method for determining the order in which a plurality of contents are delivered, and relates to a content delivery scheduling method capable of determining a time when content delivery is completed and a user can use the content when requested by the user.

[0002]

2. Description of the Related Art In a service, such as VoD, in which a large number of users select a desired digital content from a plurality of digital contents stored in a server and receive a delivery, a delivery request for the same content is made in a short time. It originates from many users. Therefore, it is efficient to complete delivery by multicast communication to all users requesting the same content at once, instead of delivering the content from the server to the client one-to-one.

[0003] On the other hand, digitization of satellite broadcasting and terrestrial broadcasting is rapidly progressing on a worldwide scale, and data communication using broadcast media is expected to be widely spread in the future. In addition, CA that is effective as an inexpensive high-speed access line
TV-based data communication is also becoming widespread. In these bandwidth sharing networks, multicast communication can be easily performed because of the physical form in which transmitted data reaches all users. Therefore, it is effective to realize content distribution using these broadcast networks.

In order to realize content delivery in a shared bandwidth network, it is necessary to consider how to select the next content to be delivered when the delivery of a certain content is completed and the line becomes free. . An important quality measure of a user in this service is a waiting time (response time) from when a delivery request is issued to when delivery of the content is completed. When M kinds of contents are prepared in the server, the distribution order of these M contents greatly affects the average waiting time.
It is important to consider a content delivery scheduling method. Since the number of contents that can be delivered at the same time is small in a bandwidth sharing network, an increase in waiting time is expected when delivering large-capacity contents. If the waiting time is not determined at the time of a delivery request, there is no indication of how long to wait until the delivery is completed, so the stress of the user increases, and the possibility that the user who once requested the delivery cancels before delivery increases. . Therefore, it is important to determine the waiting time when the delivery is requested.

[0005] The following summarizes some typical systems that have already been studied, and considers two points: (1) whether the waiting time is determined when requested, and (2) the average waiting time.

[FIFO (First In Firsts)
t Out)] This is the simplest method, and the contents are delivered in the order in which the requests are generated. However, this is different from a normal FIFO in that when a content already in the delivery queue is requested by another user, the content is not duplicated in the queue. This is due to the feature of broadcast networks that the delivery is broadcast to all users. If there are M types of content, the maximum length of the queue is M. Since the waiting time is determined at the stage of being inserted into the queue, the waiting time can be determined when the delivery is requested. However, since popular contents and low-content contents are treated in the same manner, popular contents may be waited for a long time due to delivery of a large number of unpopular contents.

[CFA (Cyclic Fixed A)
llocation)] as a delivery method of content, it is conceivable to determine the delivery schedule in advance on the basis of the content selection probability q _m. This is called push-type delivery because the content is delivered in a fixed order regardless of the request status from the user.
Since the next delivery time is clear when the content is requested,
It is possible to guarantee the waiting time. In scheduling, the period L is provided, a delivery number k _m of each content in the L in proportion to √Q _m, further when the k _m-number of distribution interval between L / k _m, to minimize the average waiting time And the lower limit is

[0008]

(Equation 1)

[0009]

[0010] Since the status of the request from the user is not reflected, there is a possibility that the content having no waiting user is actually delivered. For this reason, especially at the time of low load, the performance is deteriorated as compared with other methods.

[LWF (Longest Wait F)
rst)] Several methods have been proposed for selecting the next content to be delivered when the delivery of the content is completed and the line becomes free. These are called pull delivery because they select content to be delivered based on a request from the user. In either method, the average waiting time and the throughput are aimed to be improved over the FIFO by using the difference in the popularity of each content. Among these schemes, it is empirically known that a scheme called LWF minimizes the average waiting time. In LWF, the total sum of the waiting time of the waiting user is calculated for each content, and the maximum content is selected and delivered. Although a good waiting time characteristic is exhibited in any of the load regions, there is a problem that the waiting time is not determined because the delivery time is undecided when the content is requested.

[0012]

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and it is possible to complete the delivery of contents to each user at a guaranteed time, and to determine the degree of popularity (selection) of contents. Probabilities) are different, so that all contents are not treated the same, but the average performance can be improved by favoring popular contents, and good waiting time characteristics can be achieved in any load area. It is an object to provide a content delivery scheduling method.

[0013]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention provides a service for broadcasting and delivering digital contents stored in a server via a shared bandwidth network to a delivery request at which delivery is completed. A content delivery scheduling method, wherein all content is divided into a plurality of groups according to a selection probability, a FIFO is prepared for each group, and when content delivery is requested, the requested content is already A step of determining whether or not the content is in a delivery waiting state requested by the user; and, if the determination is a new request, a FIFO prepared in a group to which the content belongs.
And a step of circulating the next content to be delivered in a round-robin manner to the beginning of a plurality of FIFOs when the delivery of the content is completed and the line becomes free. I do.

The present invention also provides the content delivery scheduling method, wherein the content selection probability is known and x ₁ is selected from the most popular ones.
This is characterized by classifying the contents into group 1, the next most popular x ₂ contents into group 2, and the least popular contents x _N into group N in the same manner.

Further, the present invention provides the content delivery scheduling method, wherein N group sizes are set so as to minimize a waiting time guaranteed for a new requesting user from a given content selection probability q _m and the number of groups N. x _n is determined by an integer programming problem.

Further, according to the present invention, in the content delivery scheduling method, when a delivery request for content arrives at a server from a user, if the request is a new request for content without a waiting user, the request is determined in advance. FIFO assigned to given group
Is inserted at the end of

Further, according to the present invention, in the content delivery scheduling method, when the delivery of the content is completed and the line becomes free, the next content to be delivered is selected by circulating through N FIFOs in a round robin manner. And if the candidate FIFO is empty,
The FIFO is skipped.

[0018]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a structural explanatory view showing an embodiment of the present invention. In the figure, 1 is a digital content selection probability information storage unit, 2 is a group size setting unit, 3 is a digital content classification information storage unit, 4 is a user, 5 is a new request determination unit, 6 is an accommodation FIFO determination unit, and 7 is delivery. Waiting digital content storage unit, 8 is a guaranteed waiting time calculating unit, 9
Is a distribution unit, 10 is a request receiving unit, 11 is a video server, 1
Reference numeral 2 denotes a shared bandwidth broadband downlink, 13 denotes a narrowband uplink, and 14 denotes a narrowband downlink. The content is delivered by the bandwidth sharing broadband downlink 12 on the bandwidth sharing network, and the transmission of the request (request) and the waiting time is performed by the bandwidth occupation communication independent of the bandwidth sharing network (for example, ISDN B Channel and the like) on the narrow-band uplink 13 and the narrow-band downlink 14.

Next, the operation of FIG. 1 will be described with reference to FIGS.

FIG. 2 is a flowchart of the preparation.
That is, the selection probabilities qm for the M contents are stored in the digital content selection probability information storage unit 1. The group size setting unit 2 determines the size xn of each group from the selection probability qm and the number of groups N by using a dynamic programming method in accordance with equations (12) and (13) of a group size calculation method described later. According to the calculated group size, each of the M contents is classified into one of the N groups. At this time, in order from the one having the largest selection probability, the size xn determined for the group having the smaller group number
Classify to satisfy. The classification result is stored in the content classification information storage unit 3.

FIG. 3 is a flowchart when a delivery request arrives. That is, when the user 4 requests delivery of the content m, the following processing is performed.

(1) The new request determination unit 5 checks whether a delivery request for the content m is already waiting. Specifically, a table indicating whether or not there is a waiting user is prepared for each content, and this table is referred to.

(2) If the request is a new request (there is no user waiting for the content m and there is no request for the content m in the FIFO), the accommodation FIFO determination is performed based on the information in the content classification information storage unit 3. The group to which the content m belongs is checked in the unit 6, and a request for the content m is inserted at the end of the FIFO corresponding to the group (a delivery waiting digital content storage unit 7 (not a content itself but a content request). ). Then, the guaranteed waiting time calculation unit 8 calculates the time at which the delivery of the content m is completed according to equations (1) and (2) of the multi-FIFO configuration described later.

(3) If the request is not a new request, the delivery completion time of the content m is checked. That is, for all the waiting requests, the respective delivery completion times are stored in a table and are referred to.

(4) For the user 4, the content m
Notify when delivery is complete.

FIG. 4 is a flowchart when the content delivery is completed. That is, when the delivery of the content is completed, the content to be delivered next is determined by the following processing. The N FIFOs are circulated in a round robin manner.
If the next candidate FIFO is empty and there is no content to be delivered, that FIFO is skipped and the next FIFO is examined. If all N FIFOs are empty, the process waits until a new delivery request arrives.
If a non-empty FIFO is found, the FIFO
Is distributed by the distribution unit 9 for the request at the top of the list.

Next, a description will be given of a precondition, a multi-FIFO configuration, and a group size calculation method for the content delivery scheduling method according to the embodiment of the present invention.

[Prerequisites] M types of contents are stored in the server.
2,..., M are labeled. Delivery request from the user population generated according Poisson process of fixed rate lambda, the content m (1 <m <M) is selected with probability q _m. It is assumed that q _m is obtained from past data by time series prediction or the like. Although q _m is generally a time variable, it is treated here as a fixed value. The M types of titles are all the same size, and the time required for delivery of each content (the content size divided by the transmission band) is T. One line is used for delivery.

[Multi-FIFO configuration] Delivery at delivery request
Set the time and guarantee the user the wait time
A simple method is FIFO, but as described above,
Cannot reflect the request frequency difference of
Therefore, a plurality (N) of FIFOs are prepared and M kinds of content are prepared.
G₁, ..., G_NDivided into groups, and each group G
_nConsider separating the FIFOs accommodated for each.
Popular and q_mBigger content in less time
As many requests are collected as possible,
It is desirable to be delivered in waiting time. On the other hand, the same FI
All contents contained in the FO have the same average waiting time
Because of the characteristics, the selection probability q_m
It is good to house things that are close to each other. So G_nRhinoceros
X_nAnd contents 1 to x ₁To G₁And x₁+1
~ X₁+ X_TwoTo G_TwoAnd similarly

[0031]

(Equation 2)

Are classified into G _N. In other words, the most popular content x ₁ piece is classified in G _1, then ₂ x from the highest popularity is classified in G _2, the most unpopular content x _N pieces are classified into G _N.

FIG. 5 shows a delivery scheduling method.
The delivery request for the content m arriving at the server is sent by the distributor (Distributor).
Is transferred to the FIFOn corresponding to the group G _n (∋m) to which If the content m does not exist in the FIFOOn, m is inserted at the end of the FIFOOn. If the content m already exists in the FIFOn, it is not newly added to the queue. Although the delivery of the content is completed at each time interval T, the N FIFOs are circulated in a round robin manner, and the leading content is delivered in the next slot. However, if the target FIFO is empty,
The FO is skipped and the next FIFO is set as a delivery target. In this way, the content is always delivered unless the N FIFOs are all empty.

Now, consider a guaranteed waiting time for a user who has requested delivery of the content m classified into the group Gn. The waiting time until the start time of the next slot is δ,
Assuming that FIFOn ′ is serviced in the next slot, the upper limit of the waiting time ω ₁ until the content at the head of FIFOn is serviced is skipped if there is an empty FIFO.

[0035]

(Equation 3)

Thus, the upper limit can be suppressed. Meanwhile, FI
Let h be the insertion position of the content m counted from the beginning of FOn (if there is no other waiting user and the content is newly inserted into the FIFO, it will be the last position, and there is already another user and there is already a FI.
If it is in the FO, it will be that position. ), FIFOn
From the time the delivery of the content at the beginning of
Assuming that the waiting time until the delivery of the content m is completed is ω ₂ , the upper limit value is

[0037]

(Equation 4)

## EQU4 ## From the above, the waiting time ω guaranteed for the user is

[0039]

(Equation 5)

The upper limit can be further suppressed, and this upper limit is notified to the user at the time of content request as a guaranteed waiting time.

According to the present invention, the number of groups N and the size x _{n of} each group are determined in advance, so that processing upon arrival of a request is only distribution to a predetermined FIFO. Further, the processing at the time of selecting the delivery content is very simple because it only involves circulating a plurality of FIFOs in a round robin manner.

[Group Size Calculation Method] Here, a method of determining the content group size x _n when the number of contents M, the content selection probability q _m , and the number of groups (the number of FIFOs) N are given will be examined.

It is desirable to determine xn so as to minimize the average value of the waiting time ω guaranteed for the user, ω￣. However, it is difficult to derive ω￣ when the delivery time of the content is fixed length. Another metric is to limit the target to new users only (requesting content that has no users waiting, so that the content is inserted at the end of the FIFO), and the average guaranteed latency for them ( W￣).

For the FIFO, q _{m is set} to the parameter 0.
271 (Zipf function will be described later), the system load (ΔT) is set to 10.0, and the average guarantee waiting time (AGW: Aver) for a new user when the number of contents M is changed from 100 to 2000.
age Guardated Waiting ti
FIG. 6 shows a correlation graph of me) and AGW for all users.

As is apparent from the figure, since the two are approximately proportional to each other with a proportionality constant of 0.555, it can be seen that it is sufficient to minimize W￣ instead of minimizing ω￣. Therefore, thereafter, the average guaranteed waiting time for new users W ユ ー ザ
Consider a method of determining x _n so as to minimize.

In order to obtain W￣, first, an average queue length h _n found by a newly requested user arriving at the FIFOn is derived by an approximate analysis. Prior to derivation, the following two assumptions are made.

(1) When selecting the next delivery content,
The probability of the corresponding FIFO being empty and being skipped is sufficiently small.

(2) The difference in the selection probabilities of the contents belonging to the same group is sufficiently small.

If the number N of groups is sufficiently smaller than the number M of contents, the assumption (1) holds. Moreover, because a grouping with a large order of q _m as described above, the content close to that of q _m are accommodated in the same group. Therefore, it can be considered that assumption (2) is also satisfied.

According to the assumption (1), the time interval at which the FIFOn is serviced is NT, but since the contents are independently selected by the probability q _m , the state transition of the FIFOn is not affected by other FIFOs. Therefore, FIFOn
Focus on only, and proceed with the analysis. The content m classified into G _n is in the FIFO On state (one or more users are waiting for delivery) (ON state), and
n (OFF state) is alternately repeated. The average of the length of time each state continues is T _on ^{(m, n)} and T _off
^{(m, n)} . T _on ^{(m, n)} is the average value of the guaranteed waiting time for the new request user of the content m, and the average waiting time NT / 2 from the arrival of the new request to the start of the delivery of the leading content of FIFOOn; since the sum of the average waiting time NTH _n since delivery of the first content is started until the delivery of the content m is started,

[0051]

(Equation 6)

Is as follows. On the other hand, the OFF state indicates that the content m
Is the state from the moment when the delivery starts (the request that arrives during the delivery is passed to the next delivery) until a new request for the content m arrives, and T _off ^{(m, n)} is expressed by the following equation. expressed.

[0053]

(Equation 7)

Therefore, if the steady probability that the content m is in the ON state at any time is P _on ^{(m, n)} ,

[0055]

(Equation 8)

Is as follows. Here, from assumption (2), the selection probabilities of x _n contents belonging to G _n are calculated as the group average value q ⁽ⁿ⁾ :

[0057]

(Equation 9)

Replace with P_on ^{(m, n)}Suffix m
Abbreviate. However, Λ_nIs the delivery request arrival
Of content classified as group n
The sum of the choice probabilities is Q_nThen (Q_n= Σ_m∈Gnq_m),
Λn = ΛQ_nBecomes A new request for content m arrives
Waiting, k contents are waiting in the FIFOn
(ON state) is determined by the parameter (x_n−
1, p_on ⁽ⁿ⁾) According to the binomial distribution
_nIs h '_n= (X_n-1) p_on ⁽ⁿ⁾Becomes h_nWhen
h '_n Are the same, h_nQuadratic equation for

[0059]

(Equation 10)

Is obtained. Here, ρ _n is a load applied to the FIFOn, and the load ρ = ΛT of the entire system is ρ _n = ρ
The relationship of NQ _n. It should be noted that ρ and ρ _n are different from the actual load and can take values larger than 1.0 due to the characteristic that requests for the same content are put together. h _n has a two real solutions different, the smaller adopts the larger to become a negative value.

[0061]

[Equation 11]

The average queue length h _n found by the newly requested user arriving at the FIFOn is derived in this way, but the h _n differs for each FIFO and is uniform among the contents accommodated in the same group.

Assuming that the average of the guaranteed waiting time for a user who has newly requested delivery in FIFOn is W￣ _n (unit is slot), using h _n ,

[0064]

(Equation 12)

Is calculated. By defining W by the sum of the weighted by occurrence rate Q _n of the delivery request W¯ _n,

[0066]

(Equation 13)

Is obtained. y _n ≡x ₁ + x ₂ + ... x _n-1

[0068]

[Equation 14]

Where Q _n is a function of x _n and y _n .
Therefore denoted the _{Q n Q n (x n,} y n) and. Similarly, W
From the fact that ¯ _n also becomes a function of x _n and y _n, W¯
_n (x _n, y _n) is referred to as.

Consider solving an integer programming problem minimizing W￣ for x _n (1 <n <N) using dynamic programming. F _n (m)

[0071]

(Equation 15)

When defined as

[0073]

(Equation 16)

The following recurrence formula is obtained. W￣ = F _N (M), and the value of i selected when calculating F _N (M) in equation (13) is used to calculate x _N and F _N-1 (M−x _N ). The group size x ₁ , x, which minimizes W￣ by reversing the recurrence formula, such that the value of i selected in x is x _N−1 .
_2, ..., it is possible to obtain the x _N.

Next, the simulation results will be described.

[Simulation Conditions] A delivery request was issued to a single server storing M types of contents in accordance with a Poisson process at a fixed rate Λ. When each request occurs, M types of contents are independently selected with a probability q _m , but a Zipf function q _m = c / m ^1-θ widely used as a selection probability distribution of video contents is used. Was. Here, c is a constant for setting the sum of the probabilities to 1. is a parameter indicating the gradient of the distribution and takes a real value of 1 or less. The smaller the value, the greater the gap between the distributions, and the greater the difference in the frequency of selection between popular content and content that is not. Unless otherwise specified, M = 1000 and θ = 0.271.

As shown in equation (8), the average queue length when a new request arrives is the slot length T if the load ρ is constant.
Becomes irrelevant. Group size x determined by
_{Since n} does not depend on T, T = 1 was set in the following evaluation. The contents are all of the same size, and the load ρ of the entire system is ρ = ΛT from the transmission time T required for the delivery of one piece of content and the required arrival rate Λ, and Λ is determined by giving ρ. Unless otherwise specified, ρ = 10.0.

[Basic Characteristics of the Present Invention] Consideration on Number of Groups The number of groups (the number of FIFOs) N can take any value of 1 or more and M or less (when N = 1, it corresponds to the FIFO method). The average guaranteed waiting time when N is changed is shown in FIG. In an area where N is small, the characteristics are improved with an increase in N, but in an area where N is large, the characteristics are deteriorated with an increase in N. The higher the load, the greater the improvement effect of the former, and the lower the load, the stronger the latter. When the load is high, the grouping of the contents using the selection probability difference is effective, but when the load is low, the FIF is used.
It can be seen that it is desirable that O be shared by all contents since a large grouping effect can be obtained. In the following evaluations, N was set to 20 unless otherwise specified.

FIG. 8 summarizes the configuration group size x _n , the load ρ _n for each FIFO, and the average guaranteed waiting time (AGW).
Because although smaller the group of the group number n x _n is smaller, which are classified into young group of group number in descending order of high demand frequency selection probability content,
The load ρ _{n of} each group increases as the value of n decreases. All contents contained in the same group have the same waiting time characteristics, but as a result of sharing the FIFO with less contents, users who have requested popular contents belonging to a group with a smaller n realize a smaller guaranteed waiting time. it can.

[Comparison with Other Systems] Here, the present invention (gF
FIFO, C
Comparison with FA and LWF is performed. However, in the CFA, the broadcast period L needs to be determined in advance, but here, the value is set to a value three times the number M of contents.

FIGS. 9 and 10 show the average guaranteed waiting time and the average actual waiting time (the waiting time from when the content is requested to when the actual delivery is completed) when M is changed. However, in the LWF, since the waiting time is not determined at the time of requesting the content, it is excluded in FIG. Even if the load applied to the system is the same, as the number of contents to be provided increases, the contents to be delivered are dispersed, so that the number of users who receive delivery by one broadcast decreases. As a result, M
Both characteristics are increasing with the increase of. In the gFIFO, since the FIFO having no waiting content is skipped, the average actual waiting time is slightly smaller than the average guaranteed waiting time. The present invention has generally shown better properties than the FIFO. Particularly, in a region where M is large, the substantial load increases, so that the characteristics of the FIFO deteriorate and gF
The gap with the IFO is increasing. LWF has the best characteristics because it selects delivery content such that the sum of waiting times in each slot is minimized.
gFIFO achieves performance close to this.

Next, FIGS. 11 and 12 show both characteristics when the Zipf parameter θ is changed. When θ = 1, M (in this case, 1000) contents are selected with exactly equal probability, and as a result, a value half of M (500)
Becomes The smaller θ is, the more requests are concentrated on specific contents, and as a result, the efficiency of the broadcast delivery is improved, so that the waiting time is reduced. In a range of −0.5 < θ < 0.5,
The gFIFO has a great improvement effect compared to the FIFO.

Finally, FIGS. 13 and 14 show both characteristics when the load ρ is changed. Since the CFA statically determines a delivery schedule irrespective of the actual delivery waiting state, it shows a constant characteristic regardless of a change in ρ. For this reason, the efficiency is remarkably deteriorated in a region where ρ is small as compared with other methods. On the other hand, in the FIFO method, characteristics deteriorate in a high load region. The present invention shows good characteristics in any load range, and achieves a performance close to LWF.

FIG. 15 summarizes the advantages and disadvantages of the present invention and the three comparative methods.

[0085]

As described above, according to the present invention, contents are classified into a plurality of groups based on the selection probabilities of the respective contents so as to minimize the average guaranteed waiting time for newly requested users, and new requests are made. Is inserted into the FIFO of a predetermined group, and the FIFO to be serviced is circulated in a round-robin manner, thereby guaranteeing the delivery completion time at the time of a delivery request to the user and minimizing the average guarantee waiting time. is there.

[Brief description of the drawings]

FIG. 1 is a configuration explanatory diagram showing an embodiment of the present invention.

FIG. 2 is a flowchart showing an operation of preparation in FIG. 1;

FIG. 3 is a flowchart showing an operation when a delivery request arrives in FIG. 1;

FIG. 4 is a flowchart showing an operation at the time of completion of content delivery in FIG. 1;

FIG. 5 is a configuration explanatory diagram showing a scheduling method according to an embodiment of the present invention.

FIG. 6 is a characteristic diagram showing a correlation between a new request AGW and all requests AGW according to the embodiment of the present invention.

FIG. 7 is a characteristic diagram showing a correlation of the AGW with respect to the number of groups according to the embodiment of the present invention.

FIG. 8 is an explanatory diagram illustrating a configuration group size according to the embodiment of the present invention.

FIG. 9 is a characteristic diagram showing a correlation between the number of video titles (the number of contents) and the average guaranteed waiting time according to the embodiment of the present invention.

FIG. 10 is a characteristic diagram showing a correlation between the number of video titles (the number of contents) and the average actual waiting time according to the embodiment of the present invention.

FIG. 11 is a characteristic diagram illustrating a correlation between a Zipf parameter θ and an average guaranteed waiting time according to the embodiment of the present invention.

FIG. 12 is a characteristic diagram illustrating a correlation between a Zipf parameter θ and an average actual waiting time according to the embodiment of the present invention.

FIG. 13 is a characteristic diagram showing a correlation between a load and an average guaranteed waiting time according to the embodiment of the present invention.

FIG. 14 is a characteristic diagram showing a correlation between a load and an average actual waiting time according to the embodiment of the present invention.

FIG. 15 is an explanatory diagram showing features of four systems according to the embodiment of the present invention.

[Explanation of symbols]

 1 Digital Content Selection Probability Information Storage Unit 2 Group Size Setting Unit 3 Digital Content Classification Information Storage Unit 4 User 5 New Request Judgment Unit 6 Accommodating FIFO Judgment Unit 7 Delivery Digital Content Storage Unit 8 Guarantee Waiting Time Calculation Unit 9 Distribution Unit 10 Request Receiving unit 11 Video server 12 Shared bandwidth broadband downlink 13 Narrowband uplink 14 Narrowband downlink

Claims (5)

[Claims] 1. A content delivery scheduling method for determining the time at which delivery is completed at the time of a delivery request in a service for broadcasting and delivering digital content stored in a server through a shared bandwidth network. Dividing the content into a plurality of groups according to the selection probability, preparing a FIFO for each group, and determining whether or not the requested content has already been requested by another user and is waiting for delivery when the content delivery is requested. A step of determining and, if the request is a new request, a FIFO prepared for the group to which the content belongs.
And a step of circulating the next content to be delivered in a round-robin manner to the beginning of a plurality of FIFOs when the delivery of the content is completed and the line becomes free. Content delivery scheduling method. 2. The content delivery scheduling method according to claim 1, wherein x ₁ is selected from the most popular ones on the assumption that the content selection probabilities are known.
A content delivery scheduling method characterized by classifying the contents into group 1 and the next most popular x ₂ into group 2, and similarly classifying the least popular content x _N into group N. . 3. The content delivery scheduling method according to claim 2, wherein the given content selection probability q _m and the number of groups N minimize the waiting time guaranteed for the new requesting user. A content delivery scheduling method, wherein a group size _xn is determined by an integer programming problem. 4. A content delivery scheduling method according to claim 3, wherein when the delivery request for the content arrives at the server from the user, if the request is a new request for the content having no waiting user, FIFO assigned to a predetermined group
A content delivery scheduling method, wherein the content delivery scheduling method is inserted at the end of the content delivery schedule. 5. The content delivery scheduling method according to claim 1, wherein when the delivery of the content is completed and the line becomes free, the content to be delivered next is obtained by circulating through N FIFOs in a round-robin manner. And if the candidate FIFO is empty,
A content delivery scheduling method, wherein the FIFO is skipped.