JP4887316B2 - Data parallel delivery method, system and program - Google Patents

Description

  The present invention relates to technology for distributing data such as video streaming over a network, and in particular, downloads the data in parallel from a plurality of servers storing the requested data in response to a single distribution request. The present invention relates to a technique suitable for effectively reducing the bias of the link load while suppressing an increase in the average link load in the data parallel distribution mode.

  In recent years, the number of users who use video streaming services such as YouTube is increasing rapidly. There is a strong demand for high-quality video streaming distribution, and it is expected that HDTV (high definition TV) having a bit rate of about 25 Mbps will be available on the Internet in the near future.

  Furthermore, UHDV (Ultra High Definition Video) with improved quality has been proposed by NHK. The uncompressed bit rate of this UHDV image is 24 Gbps, and has a high rate of 180 to 600 Mbps even after compression.

  Such video streaming services with high image quality such as HDTV and UHDV are considered to be one of the important services on the Internet, but the transmission rate of traffic flows generated by these services is very high, This is expected to cause link congestion. Therefore, when providing these services, it is necessary to consider distributing the load of servers and links.

  A CDN (Contents Delivery Network) described in Non-Patent Document 1, for example, is widely known as an efficient technique for distributing server loads. This CDN prepares a plurality of servers on the network and selects one server for each distribution request of the user. As a result, the load is distributed among a plurality of servers.

  However, since this CDN always selects only one server for each distribution request, it is not possible to mitigate the influence of the flow generated by HDTV or UHDV streaming distribution on the network.

  In order to suppress the transfer rate of these flows, it is necessary to use a plurality of servers for one streaming session for one user. For example, by using two servers for one distribution session, the transfer rate of each flow generated by this session can be reduced to half.

  In addition, it is possible to reduce the influence on the high-rate flow network caused by the HDTV or UHDV video streaming service by parallel distribution.

  Such parallel distribution technology is also used in, for example, BitTorrent, which is famous as a P2P file sharing service.

  However, with this BitTorrent, each user needs to upload data segments to other users while downloading data from other users or the original server. Therefore, the performance of BitTorrent greatly depends on the user's behavior and the uplink access line rate.

  In addition, as the number of users being distributed increases, the number of servers that provide content increases. Therefore, the throughput for downloading data increases as the number of users increases.

  Since the transfer bit rate of HDTV and UHDV video distribution is extremely high, it is difficult for a large number of users having a large-capacity uplink access link to expect to use these super-high-quality distribution services.

  Therefore, it is difficult to provide HDTV or UHDV video streaming distribution on a P2P file sharing system such as BitTorrent.

  In order to mitigate the influence of these service flows on the network, it is necessary to perform parallel distribution from a plurality of servers operated by CDN operators, ISPs (Internet Service Providers), network carriers, and the like.

A. Su, D .; Coffnes, A.M. Kuzmanovic, and F.K. Bustamante, “Drafting Behind Akamai (Travelity-Based Detouring),” ACM SIGCOMM 2006.

  The problem to be solved is that the conventional technology cannot efficiently perform parallel distribution of data using a plurality of servers.

  An object of the present invention is to solve these problems of the prior art and to realize a network delivery service for streaming video with high image quality such as HDTV and UHDV.

  In order to achieve the above object, in the present invention, for example, in streaming streaming distribution, for a single distribution request, distribution in which moving image data is downloaded in parallel using a plurality of servers storing the requested moving image content In the embodiment, the server is preferentially arranged in a node having a small average hop length to other nodes so that the average hop length of the distribution flow is shortened. For each distribution request, in addition to the nearest server with the shortest hop length, a server existing within a position within plus B hops from the number of hops is also selected as the distribution server. In order to enhance the effect of parallel distribution, super hub nodes connected to most other nodes are arranged in a plurality of networks, and servers are installed in these super hub nodes.

  ADVANTAGE OF THE INVENTION According to this invention, the bias | inclination of link load can be reduced, suppressing the average hop length of the delivery flow in the parallel delivery of data.

  The best mode for carrying out the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration example of a data parallel delivery system according to the present invention, and FIG. 2 is an explanatory diagram showing a list of networks constituting a backbone network to which the data parallel delivery system in FIG. 1 is applied. 3 is an explanatory diagram showing an example of classification of each network in FIG. 2, FIG. 4 is an explanatory diagram showing a first topology example of a network to which the present invention is applied, and FIG. 5 is a Star type network to which the present invention is applied. FIG. 6 is an explanatory diagram illustrating an example of characteristics of an H & S network to which the present invention is applied, and FIG. 7 is an explanatory diagram illustrating an example of characteristics of a Ladder network to which the present invention is applied. FIG. 9 is an explanatory diagram showing a second topology example of a network to which the present invention is applied, FIG. 9 is an explanatory diagram showing an operation result of the present invention, and FIG. 10 is a data parallel distribution according to the present invention. It is a flowchart illustrating a processing procedure example of the law.

  The data parallel distribution system in FIG. 1 includes a network topology design device 101, a server arrangement design device 102, a video server 103, and a server selection device 104, each of which includes a CPU (Central Processing Unit), a main memory, a display device, an input device, After having installed a program or data recorded on a storage medium such as a CD-ROM via an optical disk drive or the like into the external storage device, the computer is equipped with an external storage device or the like. Then, the processing according to the present invention is executed by reading in and processing by the CPU.

  That is, when distributing video streaming, in response to one distribution request, a distribution mode is used in which the video data is downloaded in parallel using a plurality of servers in which the requested video content is stored. A server is preferentially allocated to a node having a small average hop length to other nodes so that the average hop length is shortened.

  In this example, as a process execution procedure of a programmed computer, the network topology design apparatus 101 designs a network topology so that the effectiveness of parallel distribution is improved, and the server arrangement design apparatus 102 can suppress the average link load. When the server selection device 104 receives the video streaming distribution request from the user terminal 105, the server selection apparatus 104 uses the server (video server 103) used for distribution so as to reduce the link load bias while suppressing the average link load. ) Is selected.

  Specifically, when selecting a node on which a server function for parallel data distribution is installed in the network, the network topology design device 101 generates topology information indicating the connection relation of each node constituting the network and stores it in the storage device. The server arrangement design device 102 reads out and refers to the topology information from the storage device, executes the procedure for calculating the average hop length between the nodes, and gives priority to the node having the small calculated average hop length. Then, the procedure for selecting the server node for installing the server function is executed.

  The server arrangement design device 102 selects a predetermined number of server nodes on which server functions are installed.

  Further, when the server selection device 104 selects a server node that distributes the data in response to a data distribution request from the user terminal 105, the server selection device 104 executes a procedure for selecting the nearest server node having the shortest hop length. At the same time, a procedure for selecting a server node within a hop length obtained by adding a predetermined number of hops to the number of hops of the nearest server node is also executed.

  Further, the server arrangement design device 102 executes a procedure for extracting a super hub node connected to another node at a predetermined ratio or more and a procedure for installing a server function in the extracted super hub node.

  Further, the network topology design apparatus 101 reads out and refers to the topology information from the storage device, and classifies the network as one of a star type (Star), a hub & spoke type (H & S), and a ladder type (Ladder). If the network is classified into the Star type and the H & S type, the server arrangement design device 102 executes the procedure of selecting the highest-order node among the nodes constituting the network as the server node. To do.

  Further, the server installation design apparatus 102 selects a node having a small average hop length and located at the center of the network as a server node.

At that time, the server installation design device 102, as a node in the middle of the network, and the number N all nodes, the expression "Hi with the number of hops h _ij on the shortest hop path from node i to node j = ( ^{N-1) ÷ (Σ N} j = 1, j ≠ i h ij) "measure represents the center of which is defined by Hi selects the highest node.

  Hereinafter, the details of the parallel distribution art in moving picture streaming distribution by the data parallel distribution system of this example having such a configuration will be described.

  First, a network topology design technique by the network topology design apparatus 101 will be described with reference to FIGS. 2, 3, and 4.

  In this example, the actual topology of a public ISP backbone network is used for evaluation. Although the topology “39” is actually disclosed, in this example, the topology “23” having 20 or more nodes is an evaluation target.

  In FIG. 2, the names of these 23 networks, the number of nodes N, the number of links M, the high-order node ratio Rh, and the maximum node order D are summarized. The high-order node ratio Rh is defined as the ratio of the number of nodes whose node order d exceeds the average order E (d) to the total number of nodes N.

  Further, FIG. 3 shows a distribution state of the high-order node ratio Rh and the maximum node order D for each of these 23 networks. Then, based on FIG. 3, the 23 networks are classified into three groups: Star type, H & S type, and Ladder type.

  As shown in FIG. 4A, the Star type is characterized in that a very small number of hub nodes are connected to almost all other nodes, the high-order node ratio Rh is small, and the maximum node order D is large. . Here, a network of D ≧ 10 and Rh <0.2 is defined as a Star type, and as shown in FIG. 3, five networks are of a Star type.

  In the H & S type, as shown in FIG. 4B, there are several interconnected hub nodes. Since other nodes are connected to one or more hub nodes, this structure is called a hub and spoke type. It is known that an air route network is in this form.

  In the H & S type, other nodes can be reached with a small number of hops via the hub node, so that the hop distance (hop length) between the nodes tends to be small. Here, the network of D ≧ 10 and Rh ≧ 0.2 is defined as the H & S type, and as shown in FIG. 3, six networks have the main shape.

  In the Ladder type, as shown in FIG. 4C, there is no hub node, and the topology structure is a combination of several loops. While the total link length can be suppressed, the inter-node hop distance (hop length) tends to increase. It is known that the highway network is in this form.

  Here, a network of D <10 is defined as a Ladder type, and as shown in FIG. 3, 12 networks are in this embodiment.

  In FIG. 3, the networks shown in these three topology examples all exist in the United States, but it can be confirmed that the shapes are greatly different depending on the network type.

  In the Internet AS (Autonomous System) Routing Protocol (OSPF), the shortest hop path of Dijkstra is selected between each server and user because a path with the lowest cost is set between each node. Assume that.

  Next, server placement technology by the server placement design apparatus 102 will be described. Here, a node in which a server is installed is referred to as a server node, and a node in which no server is installed is referred to as a user node.

  The ratio of the number of server nodes to the total number N of nodes is defined as s, and the number of server nodes is defined as S. That is, the number S of server nodes is a value obtained by rounding up the decimal part of “s × N”.

  For users accommodated in server node i, it is most efficient to distribute the entire video data from the server installed in server node i. The user is not subject to evaluation.

  For a user accommodated in a user node, it is necessary to transfer moving image data from a server installed in another node. At this time, in order to minimize the traffic load applied to the link, it is desirable to transfer the moving image data from the server having the shortest hop distance (hop length) from the user.

  Therefore, it is desirable to install a server at a node located at the center of the network with a small average hop distance (hop length) to other nodes.

Here, the number of hops on the shortest hop route from the node i to the node j (the number of routers or the like passing through) is assumed to be _hij . Further Hi, as a measure representing the center of the node i, is defined as "Hi = (N-1) ÷ (Σ N j = 1, j ≠ i h ij) ", and a measure Hi representing the center of Servers are placed on the largest S nodes.

  When s = 0.1, NW14 (MindSpring), which is a Star type network, and NW3 (Allegiance Telecom), which is an H & S type network, are confirmed. As a result, all servers (NW14, NW3, NW3) 6) are arranged at the highest order node.

  In the Star type network and the H & S type network, the centrality (Hi) of the hub node is high, and if the server is arranged under the above-described conditions, the server is arranged in the hub node.

  When the server placement node was also confirmed with respect to NW7 (CAIS Internet) which is a Ladder type network, it was placed on four nodes indicated by arrows in FIG. 4C.

  As described above, in a Ladder type network, a server is not always arranged in a high-order node.

  Next, as a processing content of the server selection device 104, a server selection technique will be described.

  In order to minimize the average link load, it is desirable to select a server with a small hop distance.

  Therefore, in the conventional single flow distribution technique in CDN, one server with the closest hop distance is selected for each distribution request. When there are a plurality of such servers, one of them is selected at random at the start of the simulation.

  On the other hand, in the case of the data parallel distribution system of this example, in response to a distribution request from the user node i, all servers having a hop distance from the user node i of “bi + B” or less are selected. However, “bi” is a hop distance from the user node i to the server node with the shortest hop distance, and “B” is an arbitrarily given integer parameter.

  As described above, in this example, in addition to the closest node, a server having a long hop number up to “B” hop is also included in the selection target.

  However, in this case, with the increase in “B”, the number of servers used in one video streaming session increases and the parallelism improves, but the average hop length of the flow increases. This increases the average link load.

  Hereinafter, the evaluation result of the process by the system of this example will be described.

  Consider a single type of moving image content with a transfer rate of c (bps) and a time length of T (seconds). For each of the “N−S” user nodes, a distribution request is generated through a Poisson process with a generation rate of λ. The link load is normalized by c, and λ is defined by the equation “λ = ρ × M ÷ {H × (NS) × T}”.

Note that “ρ” in the above formula is the average of each link in a single flow distribution, and M is the number of links. “H” is an average hop distance to the nearest server, and is defined by an expression “H = Σ _iεZ {bi ÷ (N−S)}”. In this equation, “Z” is a user node set.

  In the subsequent evaluation, ρ = 1 and T = 1200 seconds.

  If there are a plurality of shortest hop routes for the selected server, one route is randomly selected for each distribution request.

  In FIGS. 5A and 5B, the average number of flows in each streaming distribution session (the average value of the number of servers selected for each distribution request) is set to “B” for five Star type networks. It shows.

In FIG. 5A, “s = 0.1” is set, and in FIG. 5B, “s = 0.3” is set. However, the results are shown in a range of “0 ≦ B ≦ B _max ” (“B _max ” is the minimum value of B from which all servers are selected for an arbitrary user node).

  In the Star type network, since the hop distance between the hub node and other nodes is short, all user nodes can reach all the server nodes within two hops.

  As indicated by arrows in FIG. 8A, the order of the three hub nodes is extremely large in NW12 (GoodNet). These hub nodes (super hub nodes) are connected to other 58% to 69% nodes, and many user nodes are connected to two or three super hub nodes. Therefore, in NW12, the average number of parallel flows of each distribution session is large.

  5C and 5D show normalized average link loads for a star network. However, this normalized average link load is defined as the average link load in a parallel flow distribution system (PDS) divided by the average link load in a single flow distribution system (SDS). .

  Since the delivery request generation rate is fixed, the average link load is proportional to the average hop length of the flow. Since the closest server is always selected in SDS, the average link load in SDS is minimal for a given network topology and server location. Therefore, the normalized average link load always takes a value of 1 or more.

  As “B” increases, the average hop length of the flow in the PDS increases, so the average link load also increases.

  Further, a normalized value of the link load CV (a value obtained by dividing the link load CV of the PDS by the CV of the link load of the SDS) is shown in FIGS.

  When this value is smaller than “1”, it means that the CV of the link load is reduced compared to the SDS of the SDS.

  In the four Star type networks excluding the NW 12, there are a small number of hub nodes, and many other nodes are connected to one of these hub nodes.

  As shown in FIG. 5E, when s = 0.1, the server is installed only in these few hub nodes, and when “B” is set to 1 or more, many flows are installed between the hub nodes. As a result of passing through the link, the CV of the link load increases in the PDS.

  On the other hand, in the NW 12, many nodes are connected to two or three superhub nodes, so that concentration of load on a link installed between the superhub nodes is avoided.

  In FIG. 6, three characteristics for six H & S networks are shown in the same way.

  Even in an H & S network, a large number of nodes can reach a server node (hub node) with a short hop distance, so that the number of parallel flows during distribution can be increased even when “B” is set to 2 or less.

  In addition, there are many hub nodes in an H & S network, which tend to be interconnected. Therefore, even when “B” is set to 1 or more, load concentration on a specific link between hub nodes is avoided, and the CV of the link load is reduced.

  The larger the total number of nodes N, the larger the number of parallel flows when “B” is set to 1 or more, and the degree of CV reduction of the link load also increases.

  In FIG. 7, these three characteristics are similarly shown for 12 Ladder type networks.

  Since the average distance between nodes is large in the Ladder type network, it is necessary to set “B” to a large value in order to sufficiently increase the number of parallel flows of each streaming distribution.

  Since the increase degree of the average hop length of the flow of the Ladder type network when “B” is increased is large, the normalized average link load of the Ladder type network is larger than that of the Star type and the H & S type. Even in the Ladder type network, the CV of the link load is reduced by the PDS.

  To summarize the above description, except for the NW 12, when only a small number of servers are installed in the Star type network, there is no merit of using the PDS.

  Link load when a server is installed in about 30% of nodes in a Star type network, or in a H & S type or a Ladder type network, a PDS is used, and a server with a long hop number is also used for distribution in addition to the nearest node. However, the average link load increases.

  Therefore, a clear merit of performing parallel distribution including servers other than the nearest server in the PDS cannot be found.

  However, when “B = 0” is set and only the nearest server is used for parallel delivery, it is possible to reduce the CV of the link load while avoiding an increase in the average link load, and it is clear that the PDS is used. Benefits can be found.

  However, in this case, the effect of PDS depends on how many closest servers exist for each user node. In order to confirm this, FIG. 9 shows the CV normalized value of the link load when “B = 0” is set with respect to the nearest server number ratio η.

However, η is defined by the equation “η = { _ΣiεZ xi (bi)} ÷ {(N−S) × S}”. In this equation, “xi (bi)” is the number of closest servers in the user node i (server nodes whose hop distance from the node i is bi).

  A negative correlation can be confirmed between these two scales. That is, PDS is more effective for networks with a larger η.

  Regarding η, there is no clear difference depending on the network type such as Star type, H & S type, and Ladder type.

  NW12 (GoodNet) has a large η, and even if parallel distribution is performed using only the nearest server, the effect of reducing the CV of the link load is noticeable. As described above, this is because, in the NW 12, many user nodes are connected to two or three superhub nodes.

  NW17 (Savivis Communications) and NW19 (Sprint) also have large η, but the effect of reducing the CV of the link load is limited.

  In NW17 and NW19, the maximum node orders are 10 and 8, respectively, and the hub node is only connected to other nodes of about 40% or less.

  In these networks, there is no super hub node as seen in the NW 12, and the effect of reducing the link load CV is limited. FIGS. 8B and 8C show these two networks (NW17 and NW19).

  Therefore, when there are two or more super hub nodes that connect most nodes, such as NW17, and a server is installed in such a super hub node, a video streaming service is provided by parallel distribution. It is possible to greatly reduce the link load bias while avoiding an increase in the average link load.

  Next, a processing operation example according to the present invention by the data parallel distribution system in FIG. 1 will be described with reference to FIG.

  First, the network topology design device 101 executes a procedure for generating topology information indicating the connection relation of each node constituting the network and storing it in the storage device (step S1001).

  Next, the server arrangement design device 102 reads out and refers to the topology information from the storage device, calculates the average hop length between the nodes, and gives priority to the node having the calculated average hop length, and parallelizes the data. A procedure for selecting a server node to install a server function to be distributed is executed (step S1002). At this time, a predetermined number of server nodes are selected as server nodes for installing the server function.

  Then, when the server selection device 104 selects a server node that distributes the data in response to a data distribution request, a procedure for selecting the closest server node with the shortest hop length and the number of hops of the closest server And a procedure for selecting a server node within a hop length obtained by adding a predetermined number of hops to (step S1003).

  As described above with reference to FIGS. 1 to 10, in this example, when parallel distribution of data in the network is performed, the topology information indicating the connection relationship between the nodes constituting the network is displayed by the network topology design apparatus 101. Generate and calculate the average hop length between each node with reference to this topology information by the server arrangement design device 102, and install a server function to distribute data in parallel, giving priority to the node with the small calculated average hop length To do.

  At that time, the server arrangement design device 102 selects a predetermined number of server nodes on which server functions are installed.

  Further, when the server selection device 104 selects a server node that distributes the data in response to a data distribution request from the user terminal 105 via the network, the closest server node having the shortest hop length is selected. At the same time, a server node within a hop length obtained by adding a predetermined number of hops to the number of hops of the closest server is also selected.

  Further, the server arrangement design device 102 extracts super hub nodes connected to other nodes at a predetermined ratio or more, and installs a server function in the extracted super hub nodes.

  Further, the network topology design apparatus 101 reads out and refers to the topology information from the storage device, classifies the network as one of the Star type, the H & S type, and the Ladder type, and the server arrangement design apparatus 102 includes the Star type and the H & S type. If the network is classified into the network, the node having the highest degree is selected as the server node among the nodes constituting the network.

  The server arrangement design device 102 selects a node having a small average hop length and located at the center of the network as a server node.

The server arrangement design device 102, as a node in the middle of the network, and the number N all nodes, the expression "Hi with the number of hops h _ij on the shortest hop path from node i to node j = (N _{^{-1) ÷ (Σ N j =}} 1, j ≠ i h ij) "measure represents the center of which is defined by Hi selects the highest node.

  Thus, according to the present example, in the streaming mode, in the streaming mode, in the streaming mode in which the moving image data is downloaded in parallel using a plurality of servers in which the requested moving image content is stored in response to one distribution request, The effect of reducing the bias of the link load can be obtained while suppressing the average hop length of the distribution flow. As a result, parallel distribution of data using a plurality of servers can be performed efficiently, and it is possible to realize a network distribution service for streaming video with high image quality such as HDTV and UHDV.

  In addition, this invention is not limited to the example demonstrated using FIGS. 1-10, In the range which does not deviate from the summary, various changes are possible. For example, in this example, the network topology design device 101, the server arrangement design device 102, the video server 103, and the server selection device 104 are individually provided as computer devices that execute the processing procedure according to the present invention. Each function may be provided in one computer apparatus.

  Further, the configuration of the computer device of this example may be a computer configuration without a keyboard or optical disk drive. In this example, an optical disk is used as a recording medium. However, an FD (Flexible Disk) or the like may be used as a recording medium. As for the program installation, the program may be downloaded and installed via a network via a communication device.

It is a block diagram which shows the structural example of the data parallel delivery system which concerns on this invention. It is explanatory drawing which shows the list of each network which comprises the backbone network to which the data parallel delivery system in FIG. 1 is applied. It is explanatory drawing which shows the example of a classification | category of each network in FIG. It is explanatory drawing which shows the 1st topology example of the network to which this invention is applied. It is explanatory drawing which shows the example of a characteristic of a Star type network to which this invention is applied. It is explanatory drawing which shows the example of a characteristic of the H & S type network to which this invention is applied. It is explanatory drawing which shows the example of a characteristic of the Ladder type network to which this invention is applied. It is explanatory drawing which shows the 2nd topology example of the network to which this invention is applied. It is explanatory drawing which shows the effect result of this invention. It is a flowchart which shows the example of a process sequence of the data parallel delivery method which concerns on this invention.

Explanation of symbols

  101: Network topology design device, 102: Server arrangement design device, 103: Video server, 104: Server selection device, 105: User terminal.

Claims (8)

A method of performing parallel distribution of data in a network by programmed computer processing,
As a process execution procedure of the programmed computer,
A procedure for generating topology information indicating the connection relationship of each node constituting the network and storing it in a storage device;
A procedure for reading out and referring to the topology information from the storage device and calculating an average hop length between the nodes,
And a procedure for selecting a server node having a server function for distributing data in parallel by giving priority to a node having a small calculated average hop length. The data parallel delivery method according to claim 1,
As a process execution procedure of the programmed computer,
A data parallel distribution method comprising a procedure of selecting a predetermined number of server nodes for installing the server function. A data parallel distribution method according to claim 1 or 2, wherein:
As a process execution procedure of the programmed computer,
When selecting a server node to distribute the data in response to a data distribution request,
Selecting the nearest server node with the shortest hop length;
And a procedure for selecting a server node within a hop length obtained by adding a predetermined number of hops to the number of hops of the closest server. The data parallel delivery method according to any one of claims 1 to 3,
As a process execution procedure of the programmed computer,
A procedure for extracting a super hub node in which the ratio of the number of connected nodes to the total number of nodes is equal to or greater than a predetermined ratio;
And a procedure for installing the server function in the extracted super hub node. A data parallel delivery method according to any one of claims 1 to 4,
As a process execution procedure of the programmed computer,
A procedure for reading the topology information from the storage device and referring to the network, and classifying the network as one of a Star type, an H & S type, and a Ladder type;
If the network is classified into the Star type and the H & S type, the data parallel delivery method includes a procedure of selecting the highest-order node among the nodes constituting the network as the server node. A data parallel delivery method according to any one of claims 1 to 5,
As a process execution procedure of the programmed computer,
As the server node, and the number N all nodes, the expression "Hi = (N-1) using the number of hops _{h ij} on the shortest hop path from node i to node _{^{j ÷ (Σ N j = 1}} , j ≠ _i h _ij ) ”includes a procedure for preferentially selecting from nodes having a high scale Hi representing the centrality defined by“ _i h _ij ) ”.   The program for making a computer perform each procedure in the data parallel delivery method in any one of Claims 1-6. A system that performs parallel distribution of data in a network by programmed computer processing,
As programmed computer processing execution means
A data parallel delivery system comprising means for executing each procedure in the data parallel delivery method according to any one of claims 1 to 6 .

Images