JP5314605B2 - Content distribution system and method and program - Google Patents

Description

  The present invention relates to a content distribution technique in an IP (Internet Protocol) network such as the Internet, and more particularly to a technique suitable for efficiently distributing a large volume of content such as a VoD (Video On Demand) service. is there.

  In recent years, with the increase in the capacity of access lines, VoD services that distribute video content on demand in response to user distribution requests are remarkable. In the VoD service, a service provider installs a distribution server on the network (NW), and a device (STB: Set Top Box) for temporarily storing and playing the distributed video data is installed in the user's home. It is realized with.

  In addition, the video distribution service is intended for UGC (User Generated Content) such as YouTube, which exchanges small-sized video content created by users between users, and by content production companies such as movies and TV dramas. It can be roughly divided into high-quality video content (rich content).

  The former (UGC delivery) is generally provided by a service provider independent of ISP (Internet Service Provider) using a CDN (Contents Delivery Network) service, while the latter (rich content) is content. Because of its large capacity, a form provided by an ISP that can mainly manage and control the quality of NW is common.

  Many ISPs position the VoD service as a killer service that differentiates their NW services, and it is expected that VoD services targeting rich content will continue to spread. Hereinafter, a description will be given of the VoD service for such large-volume content distribution.

  In general, viewing of moving image content tends to concentrate at night, and fluctuates greatly with the cycle of a day. However, ISPs are required to maintain stable service quality even during peak hours, so ISPs need to design distribution systems in consideration of peak demand.

  As a result, excessive delivery system design occurs during many times other than peak hours, which can reduce ISP revenue.

  Reducing peak server load is a major issue for ISPs in order to improve ISP profits and provide users with inexpensive VoD services. In addition, since rich content has a large capacity, it is also important to suppress the influence of traffic generated by distribution on the NW.

  Various technologies for reducing the server load during peak hours have been studied and put into practical use, but can be classified into three types: technology using a cache server (CDN), multicast distribution, and P2P distribution.

  In the technology using the cache server, the ISP places a copy of the content on a plurality of cache servers arranged on its own NW, and distributes from the cache server close to the user in response to the user's distribution request. According to this technology, an improvement in the response time of the user and an effect of reducing the load on the original server can be expected.

  Multicast distribution is a single distribution session distributed to multiple users who have requested the same content using a multicast distribution tree. Can be expected.

  In addition, the P2P type distribution system realizes content distribution by exchanging content held between users, and is widely used such as BitTorrent (registered trademark). Conventionally, it has been used for the purpose of exchanging UGC, but in recent years, P2P distribution technology is also used as means for distributing rich content.

  In this P2P type distribution system, each peer (user) uploads the content being received and viewed to other peers, so the more popular content increases the number of peers that provide the server function. It is a distribution technology with excellent server load reduction effect.

  In recent years, the price of hard disk drives has been drastically reduced, and it is now possible to purchase a 1T (tera) byte for about 100 USD (US dollars). Therefore, it has been widely studied to operate STB and GW (Gate Way) installed in the user's home as a cache under the management of ISP.

  Such a technique described above is described in Non-Patent Documents 1 to 3.

  However, with distribution technology using caches, the total amount delivered from the server equipment including the cache server provided by the ISP is constant, so it is necessary for the ISP to constantly increase the number of server equipment as the number of users increases. There is no significant reduction in equipment costs.

  On the other hand, multicast distribution is a technique for simultaneously distributing to a plurality of users who request the same content using a single multicast distribution tree with the distribution server as the root and the distribution destination user as the leaf, Multicast distribution is generally used in IPTV, which is effective as a technology for reducing the load on distribution servers and NWs, and provides terrestrial television broadcasting that has been widely spread in recent years on the Internet.

  However, in IPTV, the user adjusts the viewing time to the service provider and does not involve a VCR (Video Cassette Recorder) operation. In a VoD service that requests viewing of arbitrary content at a timing, dealing with inconsistencies in user distribution request times and dealing with VCR operations are issues.

  Various technologies for dealing with such inconsistencies in user distribution requests are being studied, but Pull-type distribution based on user distribution requests and distribution scheduling are designed in advance regardless of user distribution requests. It can be classified into Push type distribution.

  Pull type distribution does not immediately start distribution for a user's distribution request, but provides a waiting period, thereby collecting distribution requests for the same content generated during that time into one multicast distribution. In the simplest technique, an upper limit value D is set for the user response time, and the content is distributed after D from the time when a request for content for which there is no other waiting user exists.

  In addition, those that select distribution content by FIFO (First In First Out), those that select content with the largest number of waiting users as distribution content, those that select distribution content to minimize the average response time, etc. Various technologies are being studied.

  Push distribution, on the other hand, is a technique for statically designing the timing at which each content is distributed in advance by a service provider as a distribution schedule, and broadcast distribution with all users as distribution destinations is usually used.

  In this case, the timing of the user's distribution request is not taken into account, and the user views the content that he / she wants to view according to the distribution schedule. Therefore, the on-demand property is impaired, and the service implementation form is closer to IPTV than VoD. . Various distribution schedule design techniques have been studied, all of which are considered from the viewpoint of reducing response time.

  In Pull type distribution, in order to obtain the load reduction effect of the server and NW, it is necessary to increase the user multiplicity of the distribution tree, and the response time is inevitably deteriorated.

  In push-type distribution, the distribution cycle tends to be long except for highly popular contents, and on-demand performance is still sacrificed.

  Originally, in a VoD service that places importance on on-demand, it is desirable to immediately start distribution in response to a user's distribution request, and deterioration of response time is a major problem in multicast distribution.

  In addition, in response to the VCR operation, the user needs to be transferred to another multicast distribution tree close to the playback point after the VCR operation.

  In addition, since there is a high possibility that there is no distribution tree that completely matches the reproduction timing, complicated processing such as unicast distribution of timing error is required.

  In the P2P distribution system, the requested content is distributed from other users who are viewing the content to the distribution requesting user. When there is no other user who views the same content, or when there is a shortage of upload bandwidth even if it exists, it is distributed from the distribution server. The more popular the content, the larger the number of viewing users, so the distribution system is highly scalable.

  In this P2P distribution system, it was originally used for file download that is used after receiving the entire file, but it has been widely applied to video streaming services and is also used as a commercial system. In addition, application of VoD compatible with VCR operation to commercial systems is also seen.

  However, the user is likely to leave the distribution system after the content viewing ends, and even during viewing, there is a possibility that the viewing is interrupted and left.

  It is necessary to continue distribution even if there is a leaving user or movement of the playback point due to VCR operation, and complicated processing such as management of the playback point of each user and switching processing of the distribution source is necessary. It becomes.

  In addition, the performance of the entire system greatly depends on the uplink capacity of each user. The uplink capacity of the user's access line such as ADSL is often smaller than the downlink capacity, and a sufficient effect cannot be obtained when the uplink capacity is small. Also, since many of the server functions depend on users who act autonomously, it is difficult to optimize the entire system.

  In response to such problems of the P2P type distribution system, a VoD system that performs P2P distribution by using STBs and GWs installed in the user's home as a cache device that is managed intensively by an ISP has been studied. Hereinafter, these P2P distribution systems using STBs that are intensively managed by ISPs are called ISP-controlled P2P distributions.

  In this ISP-controlled P2P distribution, since the STB in the user's home is under the management of the ISP, it is possible to always distribute the owned content to other users. This eliminates the need to switch distribution sources due to user detachment, which was a problem with conventional P2P distribution systems. By optimizing each STB's cache content and distribution source STB selection, overall optimization is possible. It becomes possible.

  However, it is necessary for each user to bear the power required to distribute content to other users, and it is necessary to give the user an incentive to install an STB controlled by the ISP.

  In addition, although the load on the distribution server is reduced, traffic is generated in the NW due to P2P distribution between users, so the effect of reducing the load on the NW cannot be obtained.

M. Allen, B. Zhao, and R. Wolski, "Deploying Video-on-Demand Services on Cable Networks," IEEE ICDCD 07. H.Ma and K.Shin, "Multicast Video-on-Demand Services," ACM SIGCOMM CCR, 32 (1), pp.31-43, 2002. B. Cheng, L. Stein, H. Jin, and Z. Zhang, "Towards Cinematic Internet Video-on-Demand," ACM EuroSys 08.

  The problem to be solved is that in the conventional technology, it is impossible to reduce both the load of the server that distributes the content in the IP network such as the Internet and the network load.

  An object of the present invention is to solve these problems of the prior art and enable efficient distribution of large-capacity content such as a VoD service.

  In order to achieve the above object, according to the present invention, in a VoD service or the like, in addition to a distribution mode in which unicast distribution is immediately performed in response to a user's distribution request, a predetermined distribution schedule is provided regardless of the distribution request from the user. As a result, popular content is broadcasted to all users at all times, and pre-distributed to STBs installed in the user's home.

  According to the present invention, it is possible to reduce the load on the distribution server and the load on the network accompanying the distribution of the content by the VoD service or the like, and to efficiently distribute the content.

It is a block diagram which shows the structural example of the content delivery system which concerns on this invention. It is a block diagram which shows the structural example of the delivery schedule design apparatus in FIG. It is a sequence diagram which shows the processing operation example of the content delivery system which concerns on this invention. It is explanatory drawing which shows the example of evaluation of the content delivery system which concerns on this invention. It is explanatory drawing which shows the precision of the TDLT estimation process by the delivery schedule design apparatus in FIG. It is explanatory drawing which shows the efficiency characteristic of the content selection process by the delivery schedule design apparatus in FIG. It is explanatory drawing which shows the time series change of the number of sessions by the content delivery system which concerns on this invention. It is explanatory drawing which shows the efficiency of the delivery process by the content delivery system which concerns on this invention. It is explanatory drawing which shows the effect of the delivery process by the content delivery system which concerns on this invention.

  Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In FIG. 1, 101 is a distribution server for distributing content, 102 is a user terminal device equipped with an STB, 200 is a distribution schedule design device for performing content distribution control according to the present invention, and the distribution server 101 and user terminal device 102 is connected via the Internet, and the distribution server 101 and the distribution schedule design device 200 are connected by a LAN (Local Area Network) or the like.

  Each of the distribution server 101, the user terminal device 102, and the distribution schedule design device 200 includes a computer device including a CPU (Central Processing Unit), a main memory, a display device, an input device, an external storage device, etc. After installing the program and data recorded in a storage medium such as a CD-ROM through an external storage device, the program is read from the external storage device into the main memory and processed by the CPU to execute the functions of each processing unit. To do.

  For example, as shown in FIG. 2, the delivery schedule design apparatus 200 includes a setting parameter input unit 201, a BC pre-delivery rate and channel number setting unit 202, and a TDLT for each content as processing means for executing programmed computer processing. A measurement unit 203, a TDLT estimation unit 204 for each content of the next day, a distribution content allocation unit 205 for each BC distribution channel, and a distribution schedule unit 206 are provided.

  In FIG. 1, the distribution server 101 is connected to many user terminal devices (not shown) other than the user terminal device 102. Further, even if the distribution server 101 and the distribution schedule design device 200 are configured by individual computer devices, the functions of the distribution server 101 and the distribution schedule design device 200 may be provided in one computer device.

  In such a configuration, in the content distribution system of this example, the distribution server 101 immediately unicasts the content in response to the user's distribution request, and the distribution schedule design apparatus 200 preliminarily distributes the content with many distribution requests. Broadcast distribution (pre-distribution) to all user terminal devices 102 (set top boxes: STBs) in advance according to a predetermined distribution schedule.

  The distribution schedule design apparatus 200 measures the total download time (TDLT: Total DownLoad Time) of each content in the TDLT measuring unit 203 of each content, and can be viewed by the user pre-distributed to the STB in the distribution scheduling unit 206 From the contents other than the contents within the period, the contents having a large TDLT are selected in advance in order of priority, and the selected contents are scheduled to be pre-distributed every day.

  In addition, the delivery schedule design apparatus 200 uses the exponential moving average based on the past measured TDLT value of each content measured by the TDLT measurement unit 203 of each content in the TDLT estimation unit 204 of each content of the next day. The distribution schedule unit 206 selects a content to be distributed in advance based on the TDLT calculated using the exponential moving average.

In addition, the distribution schedule design device 200 uses the distribution schedule unit 206 to perform advance distribution using the downlink capacity A _d (bps) of the access line and the distribution rate R (bps) of unicast distribution that is the same as the content playback rate. the delivery rate R _{b (bps),} for any pre-delivery channel number B, set _{R b = (a d -R)} / B, is scheduled to perform a pre-delivery at the delivery rate set.

  In addition, the delivery schedule design apparatus 200 uses the BC pre-delivery rate and channel number setting unit 202 to sequentially assign the contents whose content length is equal to or less than the free time to the pre-distribution channel having the free delivery time in order of the free delivery time. Assigned in descending order of TDLT guess.

  Then, the distribution schedule design device 200 divides the estimated value of TDLT of each content by the content length from the content allocated to each pre-distribution channel by the BC pre-distribution rate and channel number setting unit 202 in the distribution schedule unit 206. Select values in descending order and schedule to pre-deliver.

  Hereinafter, a content distribution technique using the content distribution system of this example will be described.

  In recent years, the increase in the downlink capacity of access lines in the Internet is remarkable, and it is expected that an environment in which a sufficiently large band can be used compared with the reproduction rate of rich contents is expected.

  For example, while the playback rate of SD (Standard Definition) quality video is 2 Mbps, some environments provide a maximum of 200 Mbps of transmission bandwidth in the downstream direction, although it is shared by multiple users. ing.

  It is unlikely that the user will always use the downlink capacity, and a large transmission band is still unused while receiving content distribution. For this reason, in addition to the normal delivery that is delivered from the delivery server on demand to the user's delivery request, the unused transmission capacity is utilized, and the content is always delivered and stored in the STB regardless of the user's delivery request. It is effective.

  That is, when the requested content is already accumulated in the STB by the prior distribution when the user requests distribution, distribution from the distribution server can be avoided, and the load at the peak of the distribution server and the network can be reduced.

  In general, the transmission resources of the access line are shared by a large number of users, and the throughput decreases when a plurality of users receive different contents at the same time. However, when delivering to the STB regardless of the user's delivery request, the ISP can freely set the delivery content and delivery timing, and by broadcasting the same content to all users, the downlink A decrease in throughput can be avoided.

  Therefore, in this example, in the VoD service, predetermined content is distributed in advance to a predetermined user by broadcasting.

  In such a content distribution technique of this example, unlike the P2P type distribution, the content cached in the STB is viewed only by the user who owns each STB, so it is necessary to give the user an incentive to upload the content to other users In addition, since no traffic is generated in the network when viewing content distributed in advance to the STB, a reduction in the network load can be expected in addition to the server load at the peak time. Furthermore, it can easily handle VCR operations.

  In the configuration of the content distribution system illustrated in FIG. 1, the user of the user terminal device 102 can locally cache content if the content that the user terminal device 102 desires to view is pre-distributed in the STB provided in the user terminal device 102 of the user terminal device 102. Watch by playing.

  If the content to be viewed does not exist in the STB of its own user terminal device 102, the user terminal device 102 requests distribution to the distribution server 101.

  Upon receiving the distribution request from the user terminal device 102, the distribution server 101 immediately distributes the requested content to the requesting user terminal device 102 by unicast.

  The distribution schedule design apparatus 200 measures the total download time (TDLT) for each content distributed from the distribution server 101 in the TDLT measurement unit 203 for each content, and uses the measurement result to estimate the TDLT for each content on the next day. In the unit 204, for example, the TDLT of each content of the next day is estimated at every starting point of the day (midnight).

Also, in the setting parameter input unit 201, the average length S (seconds) of the content, the downlink capacity A _d (bps) of the access line, and the reproduction rate R (bps) of the content are set in the BC pre-delivery rate and channel number setting unit 202. In the BC pre-delivery rate and channel number setting unit 202, the BC pre-delivery rate and the number of channels are set using the obtained setting parameters.

  Then, in the distribution content allocation unit 205 to each BC distribution channel, the BC pre-distribution rate and channel number setting unit 202 uses the TDLT of each content estimated by the TDLT estimation unit 204 of each content of the next day. Content to be distributed on each BC distribution channel is allocated, and the distribution schedule unit 206 determines the order of the content allocated to each BC distribution channel (broadcast pre-distribution schedule).

  In this way, based on the broadcast pre-distribution schedule determined by the distribution schedule design device 200, the distribution server 101 broadcasts content to each user terminal device (102) registered in advance.

  The procedure of the broadcast pre-distribution processing operation in the VoD service in the content distribution system of this example will be described with reference to FIG.

  3, the service provider 31 corresponds to the distribution schedule design apparatus 200 in FIG. 2, the distribution server 32 corresponds to the distribution server 101 in FIG. 2, and the user 33 corresponds to the user terminal apparatus 102 in FIG.

  The service provider 31 sets the BC pre-distribution rate and the number of channels to the distribution server 32 (step S301), and the service operation starts (step S302).

  On the first day, when the distribution server 32 distributes the content (step S304) in response to the content distribution request from the user 33 (step S303), the total download time of the content is measured (step S305). At the end of the first day, the BC pre-distribution schedule determined by the service provider 31 is notified to the distribution server 32 (step S306).

  On the second day, the distribution server 32 broadcasts the notified content to the corresponding user 33 according to the schedule notified from the service provider 31 on the previous day (step S307), and the first day Similarly to the content distribution request from the user 33 (step S308), the content is distributed (step S309), the total download time of the content is measured (step S310), and at the end of the second day, The BC pre-distribution schedule determined by the service provider 31 is notified to the distribution server 32 (step S311). The above process is repeated every day.

  Details of the broadcast pre-distribution technique in the VoD service by the content distribution system of this example will be described below.

  First, the <overview> will be described. Here, a description will be given assuming that the distribution schedule design device 200 is provided in the distribution server 101 in FIG.

  The content distribution system of this example is composed of a distribution server and a distribution network prepared by an ISP, and a plurality of users (user terminal apparatuses) having moving image viewing apparatuses such as STB and TV.

  The distribution server stores all M contents to be provided, and the distribution server immediately unicasts the requested content in response to a distribution request from the user.

  The user reproduces the distributed content while accumulating it in the STB. Since it is unicast delivery, it can easily handle any VCR operation.

  Further, in addition to such on-demand unicast distribution, the distribution server always distributes the content to the user at all times, and the user always stores the pre-distributed content received at the STB.

  VoD distribution requests have a strong periodicity with a period of one day, so pre-distributed content is determined by the distribution server on a daily basis, regardless of the user's distribution request, and the distribution timing To schedule.

  The product of the number of channels used for pre-delivery and the delivery rate must be less than or equal to the available bandwidth of the downlink capacity of the access line, but the downlink line is shared by multiple users, and different contents are simultaneously sent to multiple users. In the case of distribution, the downlink capacity of each user is reduced.

  However, since the distribution server can freely set which content is distributed at which timing in the pre-distribution, the same content is broadcast (BC) to all users, thereby avoiding a decrease in downlink throughput.

  When the content that the user wants to view exists in his / her STB, the user can view the content without requesting the distribution server.

  When viewing content pre-distributed to STB, no traffic is generated on the network, and any VCR operation can be easily handled.

  Therefore, BC pre-delivery of popular content can greatly reduce the load on the distribution server and network.

  In general, from the viewpoint of copyright protection, since a reproducible period is set for the content stored in the STB, the content that has passed the reproducible period is deleted from the STB.

  Therefore, even for the same content, in order to always cache the STB, it is necessary to repeat BC pre-distribution in the period of the reproducible period.

  Since BC pre-distribution can be performed at all times, including in the early morning hours when there are few distribution requests, it is possible to reduce load during peak hours while effectively utilizing free resources of the distribution server and network.

  Moreover, as a result of smoothing the power consumption in terms of time, surplus power during off-peak hours can be used, and a reduction effect of carbon dioxide emissions can be expected.

  Hereinafter, main differences between the multicast distribution and the P2P distribution, which are existing distribution systems, and the BC pre-distribution in this example will be described.

  First, <Differences from multicast distribution> will be described.

  In multicast distribution, response time increases and immediate distribution is impaired. In BC pre-distribution in this example, viewing can be started immediately for all distribution requests, and on-demand is fully realized. .

  In addition, multicast distribution requires a complicated process of switching the distribution tree to support VCR operations, whereas in BC pre-distribution in this example, the user views the content pre-distributed to STB, or Since the content that is immediately unicast delivered from the delivery server is viewed, VCR operations can be easily handled in either case.

  Next, <Differences from P2P distribution> will be described.

  In the case of conventional P2P distribution, the server function depends on the user's autonomous behavior, so there is a process to continue distribution even when the user leaves or VCR operation, and incentives for the user to upload content to other users. Necessary, and overall optimization is difficult.

  In the case of ISP-controlled P2P distribution, these problems can be solved, but the point of using P2P distribution remains the same, and it is necessary to provide an incentive to bear the power costs caused by content distribution to other users And traffic is generated in the network due to the distribution between users, and therefore the effect of reducing the network load cannot be expected.

  In contrast, the BC pre-distribution in this example is the same in that the STB is used as a cache, but the content cached in each STB is viewed only by the user who owns each STB, so the user is asked for some cooperation. No incentive is required, and no traffic is generated in the network due to viewing of content cached in the STB.

  Next, <prerequisites> when performing processing in the content distribution system of this example will be described. The following (A) to (G) are listed as <preconditions>.

Precondition (A): All the downlink capacities of the access lines are A _d (bps), and the content playback rates are all R (bps).

  Precondition (B): When a unicast is distributed from a distribution server in response to a user's distribution request, distribution is performed at a distribution rate equal to the reproduction rate R.

Precondition (C): The delivery rate of BC delivery is always set to R _b (bps) in each BC delivery channel.

  Precondition (D): After the content is distributed to the STB, it is deleted at midnight when the L day has passed, including the distribution date.

  Precondition (E): The storage capacity of the STB is sufficiently large, and all the contents pre-delivered by BC can be stored in the STB.

  Precondition (F): Log information (content ID and viewing time) when the user views the BC pre-distributed content is notified to the distribution server.

  Precondition (G): Schedule content to be pre-delivered during 24 hours on the nth day at the start time (midnight) on the nth day.

  Next, <delivery control method> in the content delivery system of this example will be described.

  First, <BC pre-delivery rate and channel number setting> in setting parameter input unit 201 and BC pre-delivery rate and channel number setting unit 202 in FIG. 2 will be described.

Prior to the start of operation of the VoD service, the ISP sets the BC pre-distribution distribution rate R _b and the BC pre-distribution channel number B as follows. First, the average length of content is set to S (seconds), and R _b = γR is set for a positive real parameter γ.

In order to secure a transmission band for on-demand unicast distribution for each user, it is necessary to satisfy “B ≦ (A _d −R) / γR”. Therefore, “B = floor ((A _d −R) / γR)” is set using the floor function (floor).

  Further, the average value X of the number of contents that can be distributed from one BC pre-distribution channel per day is “X = 24 × 3600γ / S”.

Since the product of B and X (“B × X”) is the average number of contents that can be BC pre-distributed per day, it is desirable to set γ so that “B × X” is maximized. That is, “B × X” can be maximized by setting “γ = (A _d −R) / BR” for an arbitrary B so that “(A _d −R) / γR” becomes an integer. .

  Next, <BC pre-delivery content selection priority> in the TDLT measurement unit 203 for each content and the TDLT estimation unit 204 for each content of the next day in FIG. 2 will be described.

  In order to increase the effect of BC pre-distribution in this example, it is necessary to increase the ratio of time that each user can view using content cached in the STB by BC pre-distribution to the total viewing time.

In general, the length of content varies, and not all users view the entire content. On the other hand, when the set of distribution requests for the content m generated on the 24th hour on the nth day is Q _{m, n} and the viewing time of the distribution request q is t _q (seconds), the content m on the nth day Total download time (TDLT) D _{m, n} (seconds) is “D _{m, n} = _{Σq∈Qm, nt q} ” _, but content with a large D _{m, n} is given priority on the nth day. The distribution server load is reduced on the nth day as the prior distribution is performed.

According to the delivery scheduling set at the start time of the nth day, the content to be pre-delivered in the 24th hour of the nth day is determined, and these contents are determined from the nth day to the (n + L-1) th day. Present in the STB for L days. Therefore, it is desirable to preferentially select content having a large sum of D _{m, n} over the L days in the distribution scheduling on the nth day.

However, since D _{m, i} of “i ≧ n” cannot be obtained at the start time of the nth day, it is necessary to estimate D _{m, n} by some method.

In this example, D _{m, n} is estimated using an exponential moving average (EMA) widely used in fields such as financial engineering and weather prediction.

The exponential moving average (EMA) uses an average of past actual measurement values weighted exponentially as time passes, and places importance on the latest data and does not completely discard old data. Using the smoothing coefficient α and the measured value D _{m, n-1} on _{the n-1st} day, an estimated value D ′ _{m, n} on the nth day is obtained using the following equation (1).

“D ′ _{m, n} = αD _{m, n−1} + (1−α) D ′ _{m, n−1} ” (Formula 1)

  The smoothing coefficient α takes a value from 0 to 1, and the closer to 1, the more important is the latest data. Generally, “α = 2 / (N + 1)” is set so that the most recent N time-series data has a weight of 86% with respect to an arbitrarily set parameter N that takes an integer value of 2 or more. Set.

  In the distribution server of this example, the latest estimated value D ′ (m, n) is stored for each content m, and D (m, n−1) is actually measured, so that the above-described value is obtained during distribution scheduling. BC pre-distribution is performed on the n-th day with priority from the largest TDLT value D ′ (m, n) estimated using Equation (1).

  However, since the content pre-delivered from the n-L day to the n-1 day is cached in the STB on the n-th day, it is considered to be excluded from the scheduling target on the n-th day.

  Next, <BC pre-distribution scheduling> in distribution content allocation section 205 and distribution schedule section 206 for each BC distribution channel in FIG. 2 will be described.

  Conventionally, various delivery scheduling schemes for BC delivery and push multicast delivery have been studied, but all have been considered from the viewpoint of reducing response time.

  On the other hand, in BC pre-distribution in the content distribution system of this example, if the requested content does not exist in the STB for the user's distribution request, it is immediately distributed from the distribution server by unicast, so distribution scheduling It is not necessary to consider response time when designing

  Therefore, in this example, the delivery scheduling is performed in which BC pre-delivery is performed at the start of each day as follows.

Among the contents not cached in the STB, the BC pre-distribution is preferentially performed for the content having a large TDLT estimated value D ′ (m, n). If the length of the content m is S _m (seconds), the number of distribution requests on the nth day of the content m can be estimated by “D ′ (m, n) / S _m ”. By distributing at an earlier timing, it becomes possible to use content distributed in advance for more viewing requests.

  From such consideration, distribution scheduling is performed as follows (a) to (e).

(B) it is initialized to "24 × 3600" seconds free time E _i for each broadcast distribution channel i.

  (B) Priorities are set in descending order of D ′ (m, n) for the contents excluding those pre-delivered by BC on the n-th to n−1 days.

(C) For the distribution channel o with the largest free time E _i , the content with the highest priority among the unallocated content satisfying S _m ≦ E _o is allocated.

  (D) The above allocation process is repeated as long as allocatable content exists.

(E) For each distribution channel, the distribution order of the allocated content is set in descending order of “D ′ (m, n) / S _m ”.

  Next, actual <performance evaluation> of the content distribution system of this example will be described.

  First, the effectiveness of the content distribution system of this example is evaluated using access log data collected in the PowerInfo (trademark) VoD system, which is a commercial VoD service provided by China Telecom.

  The log data used in this evaluation example includes all distribution requests for 20,921,657 that occurred during the seven months from June to December 2004. The total number of contents is 6,735 and the total number of users. Is 37,360.

4 (a), the cumulative Kosuke distribution content length S _m: indicates the (CCD Complementary Cumulative Distribution). The contents having a length of 45 minutes and 90 minutes each occupy about 40% of the total, and the average content length S is 3,510 seconds.

  Further, in FIG. 4B, the average CCD over the entire period of the values obtained by totaling the total download time TDLT per day for each user or content is shown for each of the user set and the content set.

  The difference in TDLT per day for each user and content is large, and it can be confirmed that a small number of users are delivering a large amount of content and that distribution requests are concentrated on specific content. In addition, many users watch content on an average of several tens of minutes to one hour per day, and the total distribution time of many contents per day is about 1 hour to 10 hours.

  The BC pre-distribution channel number B can be set arbitrarily, but since the influence of the division loss increases as B increases, the lower priority content is selected as the BC pre-distribution target. Therefore, B = 1 is set in the evaluation.

  In addition, it is assumed that the user's viewing pattern is not affected by the BC pre-distribution, and a distribution request is generated according to the access log.

And sets the downlink capacity of the access line A _d = 10Mbps, the reproduction rate of the content to R = 2 Mbps. At this time, γ = 4, the BC pre-distribution rate is R _b = 8 Mbps, and the average number of BC pre-distribution contents X per day is X = 98.46.

  Next, <estimated accuracy of total download time> will be described. In this example, the estimated accuracy of TDLT of each content using exponential moving average (EMA) is evaluated.

  In FIG. 5A, for each of the seven lags d (day), the CD of the autocorrelation coefficient of the daily TDLT of each content is plotted. As the d increases and the interval increases, the TDLT autocorrelation of each content decreases. However, if the lag is one day, the TDLT autocorrelation coefficient of about 30% of the content is 0.5 or more, and the correlation is high Is seen.

When the content retention period is set to L = 1 (day), the upper floor (X) = 98 of D ′ (m, n) estimated by the above equation (1) at the start time of each nth day The average value over the entire period of the sum of the TDLTs for the nth day divided by the total of the TDLTs for the top 98 contents in descending order of the actual TDLT on the nth day is the EMA estimation accuracy η Define as _e .

FIG. 5B summarizes η _e when the sequence length N that determines the smoothing parameter α of the EMA is changed. As seen in FIG. 5A, the TDLT autocorrelation of each content decreases as the lag increases, so that the estimation accuracy η _e gradually decreases as N increases. Therefore, N = 2 is set in the subsequent evaluation.

  When the content storage period L is 2 days or more, a part of the content cached in the STB is content distributed in advance based on the TDLT prediction value two days or more ago.

  Therefore, in order to evaluate the selection efficiency of BC pre-distributed content when L ≧ 2, in FIG. 6 (a), among the top x contents of TDLT on the day d days away, The result of having plotted the average value of the ratio of the content which became the whole period with respect to x for the value of seven d is shown.

  As shown in FIG. 6A, in the region where x is small, the matching degree of the top x pieces of content decreases as the interval d increases, but the matching degree increases as x increases. When it is about 1,000 or more, there is no difference in coincidence due to d.

  As the content storage period L increases, the average number of days elapsed from the BC pre-delivery of the content cached in the STB increases. On the other hand, since the number of cache contents increases as L increases, L is the pre-delivery content. The impact on the selection efficiency is expected to be small.

  In FIG. 6B, when x contents can be BC pre-delivered on each day, as a result of selecting pre-delivery contents based on TDLT estimated by EMA, Lx contents existing in STB on each day are selected. The average value over the whole period of the value which remove | divided the sum total of TDLT by the sum total of TDLT with respect to the top Lx content of actual TDLT over three cases is shown with respect to x.

  When x is small, as shown in FIG. 6A, as the L increases, the matching degree of the TDLT upper content greatly decreases. As a result, the pre-delivery content selection efficiency decreases as L increases. However, the degree of decrease in selection efficiency is small. As shown in FIG. 6B, for example, if L is about 7 days, a selection efficiency of about 70% can be obtained.

  In addition, the selection efficiency increases as x increases, and is approximately 85% to 90% in a region where x is about 10 or more. When x ≧ M / S with respect to the number of contents M, the selection efficiency is 1 because all contents are cached in the STB.

  From the above, by selecting BC pre-delivery content based on the estimated value of TDLT using EMA, sufficient selection efficiency can be achieved, and the load reduction effect of the delivery server and NW can be expected.

  Hereinafter, <effect> of the content distribution system of this example using such evaluation data will be described.

The total TDLT of all users for all periods is defined as Y ₁ (seconds), and at the start of the nth day, the top floor (X) = 98 contents with a large TDLT on the nth day are instantly displayed on all STBs. In the ideal cache state delivered to the user, the total of all the TDLT periods of all users who watched the STB cache content is defined as Y ₂ (seconds). Y ₃ (seconds) is defined as the total of all TDLT periods of all users who watched STB cache content in the cache state where all the content scheduled in advance is delivered to all STBs at the start of day n. The total of all the TDLT periods of all users who viewed the content actually delivered in advance by the content delivery system of this example is defined as Y ₄ (seconds). However, the first L days are excluded.

In FIG. 8, the values of “Y ₂ / Y ₁ ”, “Y ₃ / Y ₂ ”, “Y ₄ / Y ₃ ”, and “Y ₄ / Y ₁ ” are combined for the seven values of L. Yes. “Y ₂ / Y ₁ ” represents a performance limit achievable by BC pre-distribution, but the number of contents to be cached increases in proportion to an increase in L, and thus greatly improves as L increases.

“Y ₃ / Y ₂ ” represents the estimation accuracy of TDLT using EMA, but as seen in FIG. 6B, the influence of the value of L is small and high estimation of about 85% to 90%. Accuracy is achieved.

Actually, the content scheduled to be delivered cannot be delivered instantaneously at the start of each day, and the efficiency of the pre-delivery decreases because it is delivered sequentially according to the set schedule. ₄ / Y ₃ ”is as high as about 0.8 when L = 1 and about 0.9 or more when L ≧ 2.

“Y ₄ / Y ₁ ” corresponds to the effect of actual pre-delivery, but “Y ₄ / Y ₁ = Y ₂ / Y ₁ × Y ₃ / Y ₂ × Y ₄ / Y ₃ ”. Although it is the product of these three evaluation values, “Y ₂ / Y ₁ ” is most affected by L, so the efficiency of pre-delivery greatly increases as L increases.

  An ISP needs to design a distribution server based on peak demand, and it is important to suppress the peak load on the distribution server to reduce the total cost.

Therefore, FIG. 7A shows the result of plotting the maximum value of the number of simultaneous delivery sessions from the delivery server for each day over the entire period. However, since the distribution server needs to perform BC distribution at all times, “(A _d −R) R” obtained by converting the load for that into the number of sessions for unicast distribution is added.

  As shown in FIG. 7A, the maximum number of simultaneous sessions is greatly reduced by performing the pre-distribution compared to the case where the pre-distribution is not performed over the entire period (L = 0). The bigger it is, the bigger it becomes.

  Moreover, in FIG.7 (b), the result of having similarly plotted the maximum number of simultaneous sessions for every hour for 7 days of 12/1-12/7 is shown.

  As shown in FIG. 7 (b), the distribution demand shows a strong periodicity on a daily basis, and the weekends (12/4 and 12/5) increase. It can be confirmed that the number of sessions can be reduced by using pre-distribution, and that the effect is higher in times of high demand.

In FIG. 9, the maximum number of simultaneous sessions K _{max over} the entire period and the average value V (seconds) of TDLT distributed by the distribution server on each day, when L is not performed in advance (L = 0) and when L is performed Each of the seven cases is summarized.

As shown in FIG. 9, K _max and V decrease as L increases. For example, when L = 4, it can be reduced to about half compared to the case where pre-delivery is not performed.

  In this way, in the VoD service, by performing BC pre-distribution with the content distribution system of this example, the peak load of the distribution server and the load on the network can be greatly reduced, and the service provision cost of the ISP can be greatly reduced. .

However, on the other hand, a large amount of storage is required for the STB because a large amount of content is delivered by BC regardless of the user's delivery request. In FIG. 9, the utilization rate η of the pre-delivery content by the content delivery system of this example and the maximum required storage capacity Z _max (Tbyte) of the STB are shown together. However, η is defined as an average value over the entire period of the proportion of the content that is actually viewed by each user in the pre-distributed content.

  As described above with reference to FIGS. 1 to 9, in the content distribution system of this example, in the VoD service, in addition to the distribution mode in which the user's distribution request is immediately unicasted, the distribution request is Irrelevantly, in accordance with a predetermined distribution schedule, popular content is always broadcasted to all users and pre-distributed to STBs installed in the user's house.

  As a result, it is possible to reduce the distribution server load and the network load at the peak time.

  In addition, when the period during which the user can view the pre-delivered content is L days, the daily total download time (TDLT) of each content is prioritized in descending order except for the content pre-delivered in the past L days. Design a pre-delivery schedule on a day-by-day basis to broadcast pre-delivery.

  In addition, the content to be pre-distributed by broadcast is selected based on the TDLT of each content of the next day, which is estimated from the actual measured value of the past TDLT using the exponential moving average.

Further, when the access line downlink capacity is all A _d (bps), the content reproduction rate is all R (bps), and the distribution request is distributed from the distribution server to the distribution request, the distribution rate is equal to the reproduction rate R. When distributing, the broadcast pre-distribution distribution rate R _b (bps) and the number of broadcast pre-distribution channels B are set to R _b = (A _d for any B so that the number of pre-distribution contents per day is maximized. -R) Set to / B.

  In addition, each pre-distribution is performed by repeating the process of sequentially allocating the content with the largest estimated value of TDLT among the unallocated content whose content length is less than or equal to the free time to the pre-distribution channel with the maximum free distribution time. Allocate content for distribution on the channel.

  In addition, the distribution schedule is set in the descending order of the content assigned to each pre-distribution channel by dividing the estimated value of TDLT by the content length.

  In addition, this invention is not limited to the example demonstrated using FIGS. 1-9, In the range which does not deviate from the summary, various changes are possible. For example, in this example, the distribution server 101 and the distribution schedule design device 200 are configured by individual computer devices, but the functions of the distribution server 101 and the distribution schedule design device 200 are provided in one computer device. Also good.

  Also, the computer configuration of this example may be configured without a keyboard or optical disk drive. Further, the recording medium for recording the program is not limited to the optical disc, and an FD (Flexible Disk) or the like may be used as the recording medium. As for the program installation, the program may be downloaded and installed via a network via a communication device.

  31: Service provider, 32: Distribution server, 33: User, 101: Distribution server device, 102: User terminal device, 200: Distribution schedule design device, 201: Setting parameter input unit, 202: BC pre-distribution rate and number of channels A setting unit, 203: a TDLT measurement unit for each content, 204: a TDLT estimation unit for each content of the next day, 205: a distribution content allocation unit to each BC distribution channel, and 206: a distribution schedule unit.

Claims (6)

A content distribution system for distributing content via a network,
A first distribution means for unicasting content immediately in response to a user's distribution request;
Second distribution means for broadcast-distributing content having a high distribution request to set-top box devices (STBs) of all users at a predetermined timing of distributing the content;
The second delivery means is:
A first means for measuring a daily download time (TDLT) of each unicast-delivered content;
A second means for selecting from the contents excluding the contents within the user viewable period broadcast to the STB in descending order of the TDLT;
All the downlink capacity of the access line Ad (bps), all reproduction rate of the content by the STB R (bps), and the number of channels used for broadcast distribution as B, the distribution of the time the broadcast delivery rate Rb ( a third means for setting (bps) to (Ad-R) / B;
Have
The content distribution system characterized in that the content selected by the second means is broadcast-distributed at the distribution rate set by the third means at the timing of distributing the contents predetermined in the order of selection. The content distribution system according to claim 1,
The second delivery means is:
A fourth means for obtaining an estimated value of TDLT of each content of the next day using an exponential moving average from a past measured value of TDLT of each content measured by the first means;
The content distribution system according to claim 2, wherein the second means selects content to be broadcast-distributed based on the estimated value of TDLT obtained by the fourth means. The content distribution system according to claim 2,
The second delivery means is:
In a channel used for the broadcast distribution, which has an empty distribution time to which no content to be broadcast is allocated, in order of increasing the empty distribution time,
A content distribution system comprising: fifth means for allocating content whose content length is equal to or less than the empty distribution time in descending order of estimated value of TDLT. The content distribution system according to claim 3,
The second delivery means is:
Sixth means for performing broadcast distribution by selecting from the contents allocated to each channel used for the broadcast distribution by the fifth means in descending order of the value obtained by dividing the estimated value of TDLT of each content by the content length. Content distribution system characterized by   The program for functioning a computer as each means in the content delivery system in any one of Claims 1-4. A content delivery how for distributing content over a network by a computer device,
The computer device as means for executing programmed computer processing,
A first distribution unit and a second distribution unit in the content distribution system according to any one of claims 1 to 4,
By the first distribution means,
Content is immediately unicasted according to the user's distribution request,
By the second delivery means,
Measure the daily download time (TDLT) of each unicast-delivered content, select from the content excluding the content within the user viewable period broadcasted to the STB, in order from the largest TDLT , The distribution rate Rb (bps) at the time of the broadcast distribution is expressed by assuming that the downlink capacity is all Ad (bps) , the content playback rate by the STB is all R (bps), and the number of channels used for broadcast distribution is B. Set with Rb = (Ad-R) / B ,
At the timing of delivering a predetermined said content to all users of the set top box device (STB), distribution of content choose the TDLT is large, in the order selected in delivery rate Rb which the above setting (bps) A content distribution method characterized by authenticity.

Images