Cluster server apparatus

[0002] 1. Field of the Invention

[0003] The present invention relates to cluster server apparatuses, and more specifically, to a cluster server apparatus for distributing streaming data such as video data in response to requests from a plurality of terminals connected to the cluster server apparatus through a network.

[0004] 2. Description of the Background Art

[0005] For cluster server apparatuses that distribute streaming data typified by video data, continuous distribution of the streaming data to terminals connected to the cluster server apparatuses through a network is required. Since the volume of the streaming data is generally enormous, a storage device such as a large-capacity hard disk is used for data storage in the cluster server apparatus.

[0006] However, in the cluster server apparatus structured by a server comprising a large-capacity storage device for data storage, the number of terminals for receiving data is restricted by the capability limitations of the server. Therefore, in one conventional method for distribution of streaming data to a plurality of terminals, a plurality of servers are provided in the cluster server apparatus, and requests from the terminals are distributed in rotation among the plurality of servers.

[0007] An example of the structure of a conventional cluster server apparatus using the above-described method of distributing requests in rotation among the plurality of servers is briefly described below.

[0008] FIG. 18 is a block diagram showing an example of the structure of a distribution system using the above-described conventional cluster server apparatus.

[0009] As shown in FIG. 18, m terminals 40₁to 40_m(where m is an integer not less than 2) and a cluster server apparatus 200 are connected together via a network 501 in the conventional distribution system. The cluster server apparatus 200 is structured by n cache servers 20₁to 20_n(where n is an integer not less than 2) and a content server 300. The cache servers 20₁to 20_nare each connected to the content server 300 via an internal network 503.

[0010] The cache servers 20₁to 20_ninclude receivers 21₁to 21_n, caches 22₁to 22_n, and distribution units 23₁to 23_n, respectively. The content server 300 includes a streaming data storage unit 301, a read unit 302, a cache 303, and a distribution unit 304. Stream data is stored in the streaming data storage unit 301 as packets, where packets are a unit of the streaming data.

[0011] Now assume that the terminal 40₁requests the content server 300 for certain streaming data. In response to the request, the read unit 302 issues a read request for the streaming data to the streaming data storage unit 301 in the content server 300. The streaming data, in response to the read request, is then read in the read unit 302. The read streaming data is cached in the cache 303. The distribution unit 304 sends the streaming data which is cached in the cache 303 to the cash server 20₁. In the cash server 20₁, the receiver 21₁receives the streaming data sent from the content server 300. The received streaming data is cached in the cache 22₁of the cache server 20₁. The distribution unit 23₁sends the streaming data cached in the cache 22₁to the terminal 40₁.

[0012] Next, when the terminal 40₂requests the content server 300 for certain streaming data, the cluster server apparatus 200 uses the cache server 20₂, and when the terminal 40₃requests the content server 300 for certain streaming data, the cluster server apparatus 20 uses the cache server 20₃. In this way, the cluster server apparatus 200 uses the cache servers in rotation as the cluster server apparatus 200 receives multiple requests from the terminals.

[0013] As stated above, in the conventional cluster server apparatus 200, the cache servers 20₁to 20_nare structured in parallel for cyclical use in handling requests. With this structure, the number of terminals 40₁to 40_mfor receiving streaming data is increased in the entire system.

[0014] In addition to the above, another conventional cluster server apparatus which caries out a data recovery after recovery from a failure which occurred in one server in the cluster server apparatus has been disclosed in Japanese Patent Laid-Open Publication No. 8-263351 (1996-263351). In the conventional cluster server apparatus disclosed in this publication, each cache server caches on updating history of the content server. After failure recovery, the latest updating history cached in the cache server is used for data recovery in the content server. Therefore, even if a failure occurs in the data of the content server, data recovery can be quickly carried out in this conventional cluster server apparatus without requiring system suspension.

[0015] However, in the conventional cluster server apparatus 200, requests from the terminals 40₁to 40_mare provided to the cache servers 20₁to 20_nsimply in rotation. Therefore, if streaming data that is continuous in time such as video data is to be distributed, loads on the cache servers 20₁to 20_nare not properly distributed, and thereby continuous video images may be interrupted.

[0016] Further, if a failure occurs in the cache server during streaming data distribution, another server has to distribute the streaming data from the start of the streaming data instead of the portion of the streaming data which remains to be transmitted when the failure occurred.

[0017] Still further, in the conventional cluster server apparatus disclosed in the above publication, the updating history of the content server is held only by the cache servers. Therefore, if any failure occurs in the cache servers, the service to the terminals is thereby interrupted.

[0018] Therefore, an object of the present invention is to provide a cluster server apparatus for optimally distributing, in response to requests from terminals, loads on cache servers while continuing, even if a failure occurs in any one of the cache servers, data distribution to the terminals without any problem (frame skip) and for recovering from the failure.

[0019] The present invention has the following features to achieve the object above.

[0020] The present invention is directed to a cluster server apparatus for distributing streaming data such as moving-picture data continuously in time in response to requests from a plurality of terminals connected to the cluster server apparatus through a network.

[0021] First, in order to achieve load-distribution among cache servers, the present invention comprises a plurality of cache servers which are each operable to distribute the streaming data corresponding to the requests from the terminals, and a load-distribution server which is operable to receive the request from one of the terminals and to transfer the request to one of the plurality of cache servers based on whether streaming data that corresponds to the request is stored in the cache servers or not and the state of the distribution of streaming data by the cache servers, and a content server with a streaming data storage unit which is operable to store the streaming data, where the content server is operable to send, to the cache server, the streaming data that corresponds to the request from the cache server. Each of the cache servers comprises a streaming data storage unit which is operable to store the streaming data from the content server, and a request transfer unit which is operable to receive the request transferred from the load-distribution server. If the streaming data which corresponds to the request is stored in the streaming data storage unit, the request transfer unit is operable to output an instruction for distributing the stored streaming data. If, however, the streaming data which corresponds to the request is not stored the streaming data storage unit, the request transfer unit is operable to transfer the request to the content server, to store the streaming data from the content server in the steaming data storage unit, and to output an instruction for distributing the stored streaming data. Each of the cache servers also comprise a data distribution unit which is operable to distribute the streaming data stored in the streaming data storage unit to the requesting terminal based on the instruction outputted from the request transfer unit, and a cache control unit which is operable to control the streaming data from the content server to be redundantly stored in the streaming data storage unit of at least one other predetermined cache servers.

[0022] Preferably, each of the cache servers further comprises a session management unit which is operable to manage the state of streaming data distribution by its own cache server, and a streaming data management unit which is operable to manage the streaming data stored in the streaming data storage unit.

[0023] The load-distribution server of the present invention comprises a cluster control unit which is operable to, when receiving the request from the terminal, make an inquiry to all of the session management units and all of the streaming data management units of the plurality of cache servers. The inquiry requests information concerning the distribution state in the cache servers and whether the streaming data that corresponds to the request is stored or not. Based on a response to the inquiry from any one of the cache servers, the load-distribution is operable to transfer the request to any one of the responding cache servers. If any of the cache servers store the corresponding streaming data, the cluster control unit of the load-distribution server transfers the request to a least-loaded cache server among the cache servers that store the corresponding streaming data. If however, none of the cache servers store the corresponding streaming data, the cluster control unit transfers the request to a least-loaded cache server among all of the cache servers.

[0024] Alternatively, it is preferred that each of the cache servers further comprises a session management unit which is operable to manage a state of streaming data distribution by its own cache server, and a streaming data management unit operable to manage the streaming data stored in the streaming data storage unit. It is also preferred that the load-distribution server comprises a load state management unit which is operable to always communicate with the session management units of all of the cache servers and to hold states of the load on each of the cache servers, and a cluster control unit which is operable to, when receiving the request from the terminal, make an inquiry to all of the streaming data management units as to whether the streaming data that corresponds to the request is stored or not. Based on a response to this inquiry, the load-distribution server is operable to transfer the request to any one of the cache servers. Based on information in the load state management unit, if any of the cache servers store the corresponding streaming data, the cluster control unit of the load-distribution system transfers the, request to a least-loaded cache server among the cache servers that store the corresponding streaming data. If, however, none of the cache servers store the corresponding streaming data, the cluster control unit transfers the request to a least-loaded cache server among all of the cache servers.

[0025] Alternatively, it is preferred that each of the cache servers comprises a session management unit which is operable to manage a state of stream distribution by its own cache server, and a streaming data management unit which is operable to manage the streaming data stored in the streaming data storage unit. It is also preferred that the load-distribution server comprises a cluster control unit which is operable to always communicate with the session management units and all of the streaming data management units and, based on the communication, to transfer the request to any one of the cache servers. Upon receiving the request from the terminal, if any of the cache servers store the corresponding streaming data, the cluster control unit transfers the request to a least-loaded cache server among the cache servers that store the corresponding streaming data. If, however, none of the cache servers store the corresponding streaming data, the control unit transfers the request to a least-loaded cache server among all of the cache servers.

[0026] Here, if the cluster server apparatus comprises first to n-th cache servers (where n is an integer not less than 2), when the streaming data is stored in the streaming data storage unit, the cache control unit of an i-th cache server controls the streaming data to be stored in the streaming data storage units of an (i+1)-th cache server (where I is any number from 1 to n, and if i=n, a first cache server).

[0027] It is also preferred that when the streaming data which corresponds to the request is not stored in the streaming data storage unit, the cache control unit changes a sender address of the request to be transferred to the content server into a predetermined multicast address for transmission. Further, when receiving the streaming data destined for the predetermined multicast address, the cache control unit controls the streaming data to be stored in the streaming data storage unit.

[0028] Further, it is preferred that, in the i-th cache server, when the streaming data which corresponds to the request is not stored in the streaming data storage unit, the cache control unit of the i-th cache server transmits, to the content server, the request for streaming data distribution to addresses of the i-th cache server and the (i+1)-th cache server.

[0029] It is also preferred that the cluster control unit transfers the requests to the cache servers so that the number of distributable streams for each of the cache servers is equal to or less than a value {((n−1)/n)×MAX)} (where MAX is the maximum number of distribution streams that each of the cache servers can distribute), and it is preferred that the total number of distribution streams of the streaming data stored in the streaming data storage units (15_i, 15_i+1) of the i-th cache server and the (i+1)-th cache server is equal to or less than MAX.

[0030] It is also preferred that the cluster control unit transfers the requests to the cache servers so that the total number of distribution streams of the streaming data stored in the streaming data storage units of the i-th cache server and the (i+1)-th cache server is equal to or less than MAX, and it is preferred that the total number of distribution streams of the streaming data stored in the streaming data storage units of the (i−1)-th cache server (if i=1, the n-th cache server) and the i-th cache server is equal to or less than MAX.

[0031] As described above, according to the present invention, one streaming data is stored in the plurality of cache servers. Therefore, the streaming data can be directly distributed from the plurality of cache servers, and the distribution capabilities of the cache servers can thereby be greatly improved. Moreover, equal load-distribution of the cache servers can be achieved according to the volume of streaming data stored in each of the cache servers. Thus, the storage units (caches) can be efficiently used, and data can be distributed to more terminals.

[0032] Further, the present invention preferably has the following features to achieve continuous data distribution by using other cache servers when one of the cache servers becomes unable to carry out distribution.

[0033] The data distribution unit of each of the cache servers notifies the other cache server which redundantly stores the streaming data of distribution information about the streaming data being distributed to the terminal at regular intervals. Each of the cache servers further comprises a distribution-disabled detection unit which is operable to detect, through the distribution information received from the other cache server, that at least one of the other cache servers becomes unable to carry out distribution i.e., at least one of the other cache servers has failed. The distribution-disabled detection of at least one of the non-failing cache servers then provides an instruction to the data distribution unit to distribute, within a predetermined period for avoiding interruption of the streaming data, the streaming data that succeeds the streaming data most recently distributed by the other cache server which has failed and has thus become unable to carry out distribution, instead of the other cache server which has failed.

[0034] It is also preferred that the data distribution unit of the i-th cache server notifies the (i+1)-th cache server and the (i−1)-th cache server which redundantly store the streaming data of distribution information corresponding to the streaming data being distributed to the terminal at regular intervals. It is also preferred that the i-th cache server further comprises a distribution-disabled detection unit which is operable to detect, through the distribution information received from the (i+1)-th cache server and the (i−1)-th cache server, whether either or both of the (i+1)-th cache server or the (i−1)-th cache server have become unable to carry out distribution, and to provide an instruction to the data distribution unit to distribute, within a predetermined period for avoiding interruption of the streaming data, the streaming data that succeeds the streaming data most recently distributed by the either or both of the cache servers that have become unable to carry out distribution instead of the failed cache servers.

[0035] Further, it is preferred that when the data distribution unit distributes the streaming data to the terminal instead of the cache server that becomes unable to carry out distribution by following the instruction provided by the distribution-disabled detection unit, the cache control unit controls new streaming data from being stored in the streaming data storage unit of the cache server which distributes the streaming data instead of the failed cache server.

[0036] As described above, according to the present invention, one streaming data is stored in the plurality of cache servers. Also, the cache servers operate while notifying each other of their states of distribution. Thus, in addition to the effects already mentioned above, the present invention has an effect that continuous data distribution to the terminals can be made even when a failure occurs.

[0037] Still further, the present invention has the following features to achieve data recovery to the cache server which has recovered from a failure.

[0038] When the cache server that becomes unable to carry out distribution is recovered, the recovered cache server outputs a recovery notification to the data distribution unit of the other cache server that redundantly stores the streaming data. When the other cache server which has distributed the streaming data to the terminal instead of the recovery cache server receives the recovery notification, the other cache server copies the streaming data stored in its streaming data storage unit to the streaming data storage unit of the recovered cache server.

[0039] As described above, according to the present invention, one streaming data is stored in the plurality of cache servers. Also, the cache servers operate while notifying each other of their states of distribution. Thus, quick data recovery can be made to the cache server that has been recovered after it has failed.

[0040] Still further, the present invention preferably has the following features to achieve continuous data distribution using other cache servers when one of the cache servers becomes unable to carry out distribution.

[0041] Each of the cache servers further comprises a session management unit which is operable to manage a session, which indicates the execution of distributing the streaming data to the terminal through the network connection, a packet (a unit of streaming data) most recently transmitted by the data distribution unit in each session, and the number of distribution streams of the streaming data. Each of the cache servers further comprises a distribution-disabled detection unit operable to regularly transmit to and receive from the other cache servers that redundantly store the streaming data a notification corresponding to the packets for all sessions managed by the session management unit. When one of the cache servers ceases to send the notification for a predetermined time period, the distribution-disabled detection unit determines that the cache server that ceases to send the notification becomes unable to carry out distribution. Then, the distribution-disabled detection unit provides an instruction to the data distribution unit that the cache server, which redundantly stores the streaming data, to distribute, within a predetermined period for avoiding interruption of the streaming data, a packet that succeeds the packet identified by the notification most recently received for sessions of the redundantly-stored streaming data among sessions managed by the other cache server that becomes unable to carry out distribution instead of the cache server which has become unable to carry out distribution.

[0042] Alternatively, it is preferred that each of the cache servers further comprises a session management unit which is operable to manage a session identifier for identifying a session, which indicates the execution of distributing the streaming data to the terminal through the network connection, a packet identifier for identifying a packet (a unit of streaming data) most recently transmitted by the data distribution unit for each session, a session management table for storing the session identifier and the packet identifier, a table boundary value for dividing the session management table into two based on the session identifier, and the number of distribution streams of the streaming data. It is also preferred that each of the cache servers further comprises a distribution-disabled detection unit which is operable to regularly transmit to and receive from the other cache servers that redundantly store the streaming data a notification corresponding to the session identifiers and the packet identifiers for all sessions in which the streaming data managed by the session management unit are being distributed, and to update, according to the notification received by the other cache servers, the session identifier and the packet identifier stored in the session management table. The session management unit is operable to set the session identifier of the session for the streaming data being distributed by the data distribution unit of its own cache server to a value equal to or smaller than the table boundary value, and to set the session identifier of the session for the streaming data being distributed by the data distribution unit of the other cache server to a value larger than the table boundary value. The data distribution unit is operable to repeatedly distribute the packet to the terminal and to update the packet identifier when the session identifier is equal to or smaller than the table boundary value in all of the sessions. When one of the cache servers ceases to send the notification for a predetermined time period, the distribution-disabled detection unit determines that the cache server that ceases to send the notification becomes unable to carry out distribution. Upon such a determination, the distribution-disabled detection unit sets the table boundary value to a value larger than a maximum value of the session identifier within a predetermined period so as to avoid interruption of the streaming data.

[0043] Alternatively, it is preferred that each of the cache servers further comprises a session management unit which is operable to manage the number of distribution streams of the streaming data distributed from its own cache server, and a distribution-disabled detection unit which is operable to detect the other cache server that becomes unable to carry out distribution and to instruct the data distribution unit to start distributing all streaming data that has been distributed by the other cache server instead. The distribution-disabled detection unit is operable to provide an overflow notification to still another cache server that redundantly stores the streaming data, and to instruct the data distribution unit to stop distributing streaming data that exceeds the maximum number of permissible distribution streams. The distribution-disabled detection unit in the still other cache server that has received the overflow notification is operable to instruct the data distribution unit of its own cache server to transmit streaming data that succeeds the streaming data most recently distributed by the cache server that has provided the overflow notification for all streaming data that exceeds the maximum number of permissible distribution streams that the cache server that has provided the overflow notification stops distributing.

[0044] It is also preferred that if the other cache server becomes unable to carry out distribution and the response from the session management unit shows that the number of distribution streams exceeds the maximum number of distribution streams that the cache server can distribute, the distribution-disabled detection unit provides the overflow notification to still another cache server that redundantly stores the streaming data, and instructs the data distribution unit to stop distributing streaming data exceeding the maximum number of distribution streams. Further, the distribution-disabled detection unit in the still other cache server that has received the overflow notification instructs the data distribution unit of its own cache server to distribute a packet that succeeds the packet identified by a notification most recently received for streaming data exceeding the maximum number of distribution streams of which the cache server that has provided the overflow notification stops distributing.

[0045] It is also preferred that the cache servers that redundantly store the streaming data each have the session management table in which a maximum number of registrable sessions is MAX. Each of the session management units sets the session identifier to a value smaller than the table boundary value for the session of the streaming data being distributed by the data distribution unit of its own cache server, while setting the session identifier to a value not smaller than the table boundary value (setting the session identifiers I1 to In that are smaller than the table boundary value to values (MAX−I1−1) to (MAX−In−1) in the session management table in the other cache server) for the session of the streaming data being distributed by the data distribution unit of the other cache server. The distribution-disabled unit updates, according to the notification, the values in the session management table to the values of the session identifiers (MAX−I1−1) to (MAX−In−1) and the values of the packet identifiers P1 to Pn, and when the table boundary value is changed to F, the distribution-disabled unit sets the table boundary value to (MAX−F).

[0046] Further, it is preferred that when a streaming data set S1 stored in both a cache server A and a cache server B and a streaming data set S2 in both the cache server A and a cache server C are stored in the streaming data storage unit of the cache server A. It is also preferred that the cache server A has session management tables T0 and T1 and table boundary values F0 and F1 at which the session management tables T0 and T1 are divided into two based on the session identifier. In the session management table T0, the session identifier is set to a value smaller than the table boundary value F0 for the session of the streaming data included in the streaming data set S1 being distributed by the cache server A, while the session identifier is set to a value larger than the table boundary value F1 (the session identifier having one of the values I1 to In that are smaller than the table boundary value F1 are set to one of values (MAX−I1−1) to (MAX−In−1) in the session management table of the cache server B) for the session of the streaming data included in the streaming data set S1 being distributed by the cache server B. In the session management table T1, the session identifier is set to a value smaller than the table boundary value F1 for the session of the streaming data included in the streaming data set S2 being distributed by the cache server A, while the session identifier is set to a value larger than the table boundary value F0 (the session identifier having one of the values J1 to Jn that are smaller than the table boundary value F1 are set to one of values (MAX−J1 1) to (MAX−Jn−1) in the session management table of the cache server C) for the session of the streaming data included in the streaming data set S2 being distributed by the cache server C.

[0047] Further, in the cache server A, the distribution-disabled detection unit is operable to transmit a notification indicating the session identifier and the packet identifier for all sessions of the streaming data being distributed to the cache servers B and C by making an inquiry to the session management unit, and to receive the notification from the cache servers B and C. The distribution-disabled detection unit in the cache server A is also operable to set the table boundary value F0 to a value equal to or larger than MAX when no notification comes from the cache server B, and to set the table boundary value F1 to a value equal to or larger than MAX when no notification comes from the cache server C. When the number of distribution streams exceeds the maximum number of distribution streams, the distribution-disabled detection unit in the cache server A sets the table boundary value F1 to a value smaller than the maximum number of distribution streams if no notification comes from the cache server B, and sets the table boundary value F0 to a value smaller than the maximum number of distribution streams if the notification comes from the cache server C.

[0048] As described above, according to the present invention, one streaming data is stored in the plurality of cache servers. Also, the states of distribution of the cache servers adjacent to each other are stored as information in relation to sessions and packets. The cache servers operate while notifying each other of their states of distribution. Thus, in addition to the effects already mentioned above, the present invention has an effect in which continuous data distribution to the terminals can be made even when one of the cache servers becomes unable to carry out distribution.

[0049] Still further, the present invention preferably has the following features to achieve data recovery to the cache server that has been recovered after it has failed.

[0050] Each of the cache server further comprises a recovery detection unit which is operable to detect that the other cache server that has been unable to carry out distribution has recovered. When the other cache server has recovered, the recovery detection unit of the other recovered cache server is operable to provide a recovery notification to the cache server that takes over distribution of the streaming data to the terminal, and to instruct the data distribution unit to distribute the streaming data that succeeds the streaming data that has been distributed by the cache server. Further, the recovery detection unit of the cache server instructs the data distribution unit to stop the distribution of the streaming data that has been distributed thereby instead of by the recovered cache server.

[0051] It is also preferred that each of the cache servers further comprises a distribution recovery unit operable to detect that the other cache server that has been unable to carry out distribution has recovered. When the other cache server has recovered, the cache control unit of the other recovered cache server is operable to provide a recovery notification to the data distribution unit of the cache server that redundantly stores the streaming data, and to copy the streaming data stored in the streaming data storage unit of the cache server that has distributed the streaming data to the terminal instead of the other recovered server to the streaming data storage unit of the other recovered cache server. After the cache control unit completes the copy of the streaming data to the other recovered server, the recovery detection unit of the other recovered cache server is operable to provide a recovery notification to the cache server, and to instruct the data distribution unit to start distributing a packet that succeeds the packet identified by the packet identifier most recently transmitted in the streaming data that has been distributed instead by the cache server. Further, the recovery detection unit of the cache server is operable to instruct the data distribution unit to stop distribution of the streaming data that has been distributed thereby instead of by the recovered cache server.

[0052] Alternatively, it is preferred that the cache server further comprises a distribution recovery unit which is operable to detect that the other cache server that has been unable to carry out distribution has recovered. When the other cache server has recovered, the cache control unit of the other recovered cache server is operable to provide a recovery notification to the data distribution unit of the cache server that redundantly stores the streaming data, and to copy the streaming data stored in the streaming data storage unit of the cache server that has distributed the streaming data to the terminal instead of the other recovered cache server to the streaming data storage unit of the other recovered cache server. After the cache control unit completes the copy of the streaming data to the other recovered cache server, the recovery detection unit of the other recovered cache server is operable to provide a recovery notification to the cache server, and to set the table boundary value back to a value that existed before the cache server became unable to carry out distribution. Further, the recovery detection unit of the cache server is operable to set the table boundary value back to the value that existed before the cache server became unable to carry out distribution.

[0053] When the cache server carries out distribution of streaming data that exceeds the maximum number of distribution streams because the cache server performed the stream distribution instead of the other cache server that became unable to carry out distribution, it is also preferred that the recovery detection unit of the cache server instructs the data distribution unit to stop distribution of the streaming data that has been distributed thereby instead of by the other recovered cache server, provides a recovery notification to still another cache server, and then instructs the data distribution unit to start distributing streaming data that succeeds the streaming data that has been distributed by the still other cache server. In addition, the recovery detection unit of the still other cache server then instructs the data distribution unit of its own cache server to stop distributing the streaming data that has been distributed thereby instead of by the cache server that provided the recovery notification.

[0054] Alternatively, it is preferred that when the cache server carries out distribution of streaming data that exceeds the maximum number of distribution streams because of stream distribution instead of the other cache server that becomes unable to carry out distribution, the recovery detection unit of the cache server instructs the data distribution unit to stop distribution of the streaming data that has been distributed thereby instead of by the other recovered cache server, provides a recovery notification to still another cache server, and then instructs the data distribution unit to start distributing the packet that succeeds the packet identified by the packet identifier most recently transmitted in the streaming data that have been distributed by the still other cache server. In addition the recovery detection unit of the still other cache server instructs the data distribution unit of its own to stop distributing the streaming data that has been distributed thereby instead of the cache server that provided the recovery notification.

[0055] It is also preferred that each of the cache servers further comprises a distribution recovery unit operable to detect that the other cache server that has been unable to carry out distribution has recovered. When the other cache server has recovered, the cache control unit of the other recovered cache server is operable to provide a recovery notification to the data distribution unit of the cache server A that redundantly stores the streaming data, and to copy the streaming data stored in the streaming data storage unit of the cache server A that has distributed the streaming data to the terminal instead of the other received cache server to the streaming data storage unit of the other recovered cache server. After the cache control unit completes the copy of the streaming data by providing a recovery notification, the recovery detection unit of the other recovered cache server sets the table boundary value back to a value that existed before the cache server became unable to carry out distribution. Further, after the table boundary value of the cache server is set to the value that existed before the cache server failed, the recovery detection unit of the cache server A sets the table boundary values F0 and F1 back to values before the cache server became unable to carry out distribution.

[0056] As described above, according to the present invention, one streaming data is stored in the plurality of cache servers. Also, the states of distribution of the cache servers adjacent to each other are stored as information in relation to sessions and packets. The cache servers operate while notifying each other of their states of distribution. Thus, a quick data recovery can be achieved in the cache server that has recovered after it has failed. This recovery returns the previously failed cache server to the state before it became unable to carry out distribution, and thus, continuous data distribution to the terminals can be achieved even when the same cache server becomes unable to carry out distribution again.

[0057] These and other objects, features, aspects and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings.

[0058] FIG. 1 is a block diagram showing the structure of a distribution system using a cluster server apparatus according to an embodiment of the present invention.

[0059] FIGS. 2 and 3 are flowcharts for explaining the operation of distributing streaming data which is performed by a cluster server apparatus 10 according to the embodiment of the present invention.

[0060] FIG. 4 is a diagram showing an example of information stored by a streaming data management unit 14_x.

[0061] FIG. 5 is a diagram showing an example of distribution information sent by data distribution units 12₁to 12_n.

[0062] FIG. 6 is a diagram showing an example of information for a determination operation performed by a cluster control unit 21.

[0063] FIG. 7 is a diagram showing an example of the structure of streaming data held in a streaming data storage unit 15_x.

[0064] FIG. 8 is a diagram showing an example of a state of streaming data stored in streaming data storage units 15₁to 15₄in cache servers 10₁to 10₄.

[0065] FIG. 9 is a diagram showing an example of a state of streaming data stored in session management units 11₁to 11₄in the cache servers 10₁to 10₄.

[0066] FIG. 10 is a diagram showing session management tables 53₂and 53₃shown in FIG. 9 in detail.

[0067] FIG. 11 is a diagram showing a communication relation among distribution-disabled detection units 13₁to 13₄in the cache servers 10₁to 10₄.

[0068] FIG. 12 is a diagram showing the contents to be communicated between the distribution-disabled detection units 13₁and 13₂shown in FIG. 11.

[0069] FIG. 13 is a diagram showing the operation of the distribution-disabled detection unit 13₂of the cache server 10₂if the cache server 10₁becomes unable to carry out distribution.

[0070] FIG. 14 is a diagram showing the operation of the distribution-disabled detection unit 13₂of the cache server 10₂if part of a stream distribution from the cache server 10₁overflows,

[0071] FIG. 15 is a diagram showing the operation of a recovery detection unit 18₁and the distribution-disabled detection unit 13₁when the cache server 10₁which has not been able to carry out distribution is recovered.

[0072] FIG. 16 is a diagram showing a state of session management tables 53₁to 53₈and table boundary values 54₁to 54₈when four cache servers 10₁to 10₄carry out distribution.

[0073] FIG. 17 is a diagram showing a state of the session management tables 53₁to 53₈and the table boundary values 54₁to 54₈when the cache server 10₁becomes unable to carry out distribution.

[0074] FIG. 18 is a block diagram showing an example of the structure of a distribution system using a conventional cluster server apparatus.

[0075] An embodiment of the present invention is now described with reference to the annexed drawings.

[0076] FIG. 1 is a block diagram showing the structure of a distribution system using a cluster server apparatus according to the embodiment of the present invention. In this distribution system, m terminals 4₁to 4_m(where m is an integer not less than 2) and a cluster server apparatus 10 are connected together via a network 51. The cluster server apparatus 10 includes a load-distribution server 20, n cache servers 10₁to 10_n(where n is an integer not less than 2), and a content server 30. The cache servers 10₁to 10_nare each connected to the load-distribution server 20 via an internal network 52 and to the content server 30 via an internal network 53.

[0077] The cache servers 10₁to 10_ninclude session management units 11₁to 11_ndata distribution units 12₁to 12_n, distribution-disabled detection units 13₁to 13_n, streaming data management units 14₁to 14_n, streaming data storage units 15₁to 15_n, request transfer units 16₁to 16_n, cache control units 17₁to 17_n, and recovery detection units 18₁to 18_n, respectively. The content server 30 includes a streaming data storage unit 31, a read unit 32, a cache 33, and a distribution unit 34. The load-distribution server 20 is structured by a cluster control unit 21.

[0078] Note that, as shown in FIG. 1, the cache servers 10₁to 10_nare structured in the same manner. Therefore, for the purpose of describing a process that may be carried out by any one of the cache servers 10₁to 10_n, it is assumed herein that a cache server 10_x(where x is any one of 1 to n) carries out that process. Components included in the cache server 10_xare also provided with the subscript x.

[0079] The operation of each component in the cluster server apparatus 10 according to the present embodiment is now briefly described.

[0080] Upon receiving requests from the terminals 4₁to 4_mfor the streaming data, the cluster control unit 21 transfers these requests to the cache servers 10₁to 10_nin a manner described later so as to distribute the loads on the cache servers 10₁to 10_nbased on information from the session management units 11₁to 11_nand the streaming data management units 14₁to 14_n. The cluster control unit 21 is also provided with streaming data outputted from the cache servers 10₁to 10_n, and the cluster control unit 21 distributes the requested streaming data to the corresponding terminals 4₁to 4_mwhich requested the streaming data.

[0081] In the cache server 10_x, the session management unit 11_xmanages the state of streaming data distribution performed by the cache server 10_x. The data distribution unit 12_xoutputs the streaming data to the cluster control unit 21 based on an instruction from the request transfer unit 16_x. The distribution-disabled detection unit 13_xdetects whether an adjacent cache server 10_x−1or 10_x+1, which precedes or succeeds the cache server 10_xin system management (but not necessarily in a physical sense), becomes unable to carry out distribution of the streaming data. Note that “n” is substituted for “x−1” when x=1, while “1” is substituted for “x+1” when x=n. Hereinafter, the same applies for variables i, j, and k. The recovery detection unit 18_xdetects that the adjacent cache server 10_x−1 or 10_x+1 is recovered from a distribution-disabled state. The streaming data storage unit 15_xis implemented as a RAM (random access memory) to temporarily store the streaming data sent from the content server 30. The streaming data management unit 14_xholds information corresponding to the streaming data stored in the streaming data storage unit 15_x. The request transfer unit 16_xdetermines whether the streaming data requested through the cluster control unit 21 is stored in the streaming data storage unit 15_xor not. If the streaming data is stored in the streaming data storage unit 15_x, the request transfer unit 16_xinstructs the data distribution unit 12_xto provide the cluster control unit 21 with the streaming data stored in the streaming data storage unit 15_x. If, however, the streaming data is not stored in the streaming data storage unit 15_x, the request transfer unit 16_xinstructs the data distribution 12_xto provide the cluster control unit 21 with the streaming data obtained through the cache control unit 17_xfrom the content server 30. The cache control unit 17_xprovides the content server 30 with the request from the request transfer unit 16_x. The cache control unit 17_xcontrols the streaming data sent from the content server 30 upon request to be stored in the streaming data storage unit 15_x+1of the cache server 10_x+1that succeeds the cache server 10_xin system management.

[0082] In the content server 30, the streaming data storage unit 31 is implemented as a large-capacity hard disk, for example, and stores the streaming data as packets which are units (blocks) of the streaming data. The read unit 32 reads the streaming data from the streaming data storage unit 31 based on an instruction from the distribution unit 34. The cache 33 caches the streaming data read by the read unit 32. The distribution unit 34 controls the read unit 32 based on the requests sent from the cache servers 10₁to 10_n, and sends the streaming data cached in the cache 33 to a cache server among the cache servers 10₁to 10_n.

[0083] With such a structure, the cluster server apparatus 10 according to the present embodiment can distribute streaming data with optimal load-distribution to the cache servers. Further, if a failure occurs in one cache server to disable data distribution to the terminals, the cluster server apparatus 10 can also carry out a fail-safe operation to continue data distribution with another cache server. Still further, the cluster server apparatus 10 can carry out a data recovery operation after recovery of the cache server that has not been able to distribute data to the terminals.

[0084] The above-described operation of the cluster server apparatus 10 according to the present embodiment is now described in more detail.

[0085] The operation of distributing streaming data with optimal load-distribution to the cache servers is now described further with reference to FIGS. 2 and 3.

[0086] FIGS. 2 and 3 are each flowcharts for explaining the operation of distributing streaming data performed by the cluster server apparatus 10 according to the embodiment of the present invention. Note that it is assumed herein that, similar to the system environment shown in FIG. 1, the cluster server apparatus 10 is structured by the n cache servers 10₁to 10_n, and carries out data distribution among the m terminals 4₁to 4_m.

[0087] Referring to FIG. 2, a request for streaming data from a terminal 4_y(where y is any number from 1 to m) is received by the cluster control unit 21 (step S101). Upon receiving the request for the streaming data, the cluster control unit 21 inquires each of the cache servers 10₁to 10_nfor the number of streams currently being distributed and whether or not the requested streaming data is stored in any of the streaming data storage units 15₁to 15_n(step S102). In response to this inquiry, each of the cache servers 10₁to 10_nprovides the cluster control unit 21 with the number of streams currently being distributed by each of the cache servers 15₁to 15_nbased on the information in the session management units 11₁to 11_nrespectively, and whether or not the streaming data is stored based on the information in the streaming data management units 14₁to 14_n, respectively.

[0088] Each of the streaming data management units 14₁to 14_nholds information as shown in FIG. 4, for example. In FIG. 4, the streaming data management unit 14_xholds a streaming data number or name, a storage location in the streaming data storage unit 15_x, and a previous access time. With such information, the cache server 10_xcan recognize which streaming data is stored in which location of the streaming data storage unit 15_x.

[0089] Next, as shown in FIG. 2, the cluster control unit 21 recognizes, based on the responses from all the cache servers 10₁to 10_n, whether or not the requested streaming data is stored in any of the streaming data storage units 15₁to 15_nof the cache servers 10₁to 10_n(step S103).

[0090] Still referring to FIG. 2, the case in which the cluster control unit 21 recognizes that the requested streaming data is not stored in any of the cache servers 10₁to 10_n(Step S103, No) is described. The operation in steps S104 to S111 described below basically corresponds to an initialization operation in the cluster server apparatus 10.

[0091] If the cluster control unit 21 recognizes that the requested streaming data is not stored in any of the cache servers (No in step S103), the cluster control unit 21 selects a cache server 10_k(where k is any number from 1 to n) with the smallest number of streams currently being distributed from among all of the cache servers 10₁to 10_n(step S104).

[0092] The cluster control unit 21 then determines whether the number of streams currently being distributed by the selected cache server 10_kis less than or equal to a value {((n−1)/n)×MAX} or not (step S105). Note that MAX represents the maximum number of streams that a single cache server can distribute.

[0093] If it is determined in step S105 that the number of streams currently being distributed by the selected cache server 10_kis less than or equal to the value {((n−1)/n)×MAX}, the cluster control unit 21 transfers the request from the terminal 4_yto the selected cache server 10_k(step S106). Note that a load state management unit may be provided for holding states of distribution which are obtained through communication between the session management unit 11_kand all of the cache servers 10₁to 10_nat regular intervals. In this case, the load state management unit selects the cache server 10_kfor transferring the request from the terminal 4_ythereto.

[0094] On the other hand, if No in step S105, i.e., it is determined in step S105 that the number of streams currently being distributed by the selected cache server 10_kexceeds the value {((n−1)/n)×MAX}, the cluster control unit 21 issues a notification of rejecting the request, and ends the distribution operation (step S111).

[0095] When receiving the request from the cluster control unit 21, the selected cache server 10_ksends a request to the content server 30 because the requested streaming data is not stored in the streaming data storage unit 15_k. Here, the cache control unit 17_ksends the request as a request also from the cache control unit 17_k+1of the cache server 10_k+1that succeeds the cache server 10_kin system management (step S107).

[0096] On receiving the request from the cache server 10_k, the content server 30 reads the streaming data from the streaming data storage unit 31, and sends the streaming data to the cache servers 10_kand 10_k+1(step S108).

[0097] When receiving the streaming data, the cache server 10_kstores the streaming data in the streaming data storage unit 15_k, and also updates the information in the streaming data management unit 14_k. Note that, if the cache control unit 17_ksends, in step S107, a request with its sender address changed into a multicast address for one-to-many communications, the cache server 10_kcontrols the streaming data to the multicast address transmitted from the content server 30 to be received by a newly-provided cache data management unit and the streaming data is then stored in the streaming data storage unit 15_k. Then, the data distribution unit 12_kof the cache server 10_kdistributes the streaming data stored in the streaming data storage unit 15_kto the terminal 4_ythrough the cluster control unit 21 (step S109). The cache server 10_kalso notifies the cache server 10_k+1of the data contents distributed to the terminal 4_yat regular intervals (step S110). The information for notifying the cache server 10_k+1of the data contents distributed to the terminal 4_yindicates, as exemplarily shown in FIG. 5, which part of which streaming data is distributed to which of any of the terminals 4₁to 4_m. Note that the cache server 10_kmay provide notification of distribution information every time it distributes data to one or more terminals, or at predetermined intervals for all terminals.

[0098] Upon receiving from the content server 30 the streaming data which is not directly requested, the cache server 10_k+1stores the not-directly-requested streaming data in the streaming data storage unit 15_k+1, and also updates the information in the streaming data management unit 14_k+1(step S109).

[0099] As stated above, through the steps S104 to S110, the streaming data distributed upon a request for the streaming data is stored in two successive cache servers that are managed in conjunction with each other in the system.

[0100] Referring to FIG. 3, the case in which the cluster control unit 21 recognizes that the requested streaming data is stored in any one of the cache servers (step S103, Yes) is now described.

[0101] If Yes in step S103, the cluster control unit 21 selects a cache server 10_i(i is a value that x can take on), which is the smallest sequential number provided in the system management from among the cache servers that store the requested streaming data (step S121). Here, as stated above, the streaming data is always stored in at least a pair of successive cache servers 10_xand 10_x+1(step S109). Therefore, the cluster control unit 21 selects the cache server 10₁, which has the smallest sequential number, from among the pair of the cache servers 10_xand 10_x+1.

[0102] For example, if a pair of cache servers 10₂and 10₃and a pair of cache servers 10₇and 10₈store the streaming data, the cluster control unit 21 selects the cache server 10₂.

[0103] The cluster control unit 21 then determines, as to the streaming data currently being distributed by the selected cache server 10_i, whether the total number of distribution streams of the streaming data stored in the streaming data storage units 15_iand 15_i+1is less than MAX or not (step S122). If it is determined in step S122 that the number of streams is not less or equal to than MAX, the cluster control unit 21 selects a cache server 10_j(j is a value that x can take on), which is the smallest sequential number, from among the remaining cache servers that stores the same streaming data (step S128). Then, the procedure returns to step S122 for a determination as to whether the total number of distribution streams of the streaming data stored in the streaming data storage units 15_jand 15_j+1is less than MAX or not.

[0104] For example, in the above example, if the total number of distribution streams in the cache servers 10₂and 10₃is not less than MAX, the cluster control unit 21 selects the cache server 10₇. If no other cache servers store the same streaming data, the procedure goes to step S104 in FIG. 2 for further processing (step S127).

[0105] On the other hand, if it is determined in step S122 that the number of distribution streams of the streaming data stored in the streaming data storage units 15_iand 15_i+1is less than or equal to MAX, the cluster control unit 21 determines whether the number of streams currently being distributed by the cache server 10_iis less than or equal to a value {((n−1)/n)×MAX} or not (step S123).

[0106] The determination made in steps S122 and S123 by the cluster control unit 21 is based on the information as shown in FIG. 6, for example. In FIG. 6, “cache server” indicates each sequential number assigned to the respective cache servers 10₁to 10_n; “cache” indicates whether or not the requested streaming data is cached or not; “the number of distribution streams 1” indicates the total number of distribution streams of the streaming data stored in the cache servers 10_iand 10_i+1; and “the number of distribution streams 2” indicates the total number of distribution streams of the streaming data stored in the cache servers 10_iand 10_i−1.

[0107] In FIG. 6, the number of distribution streams currently being distributed by the cache server 10₂is 120+1000=1120, and the number of distribution streams of the streaming data stored in both cache servers 10₂and 10₃is 120 for the cache server 10₂and 500 for the cache server 10₃. Therefore, 620 is compared with MAX for the determination in step S122, while 1120 is compared with {((n−1)/n)×MAX} for the determination in step S123.

[0108] Note that, if there is no need to concern the case in which a failure occurs in any one of the cache servers, it is not necessary to determine whether the number of streams currently being distributed by the cache server 10_iis less than or equal to the value {((n−1)/n)×MAX} or not. Instead, a determination may be made for the streaming data currently being distributed by the cache server 10_ias to whether the total number of distribution streams of the streaming data stored in the streaming data storage units 15_iand 15_i+1is less than MAX or not. In this case, each streaming data storage unit 15_xcan be efficiently utilized.

[0109] Referring back to FIG. 3, if Yes in step S123, i.e., it is determined that the number of streams currently being transmitted by the cache server 10_iis less than or equal to the value{((n−1)/n)×MAX}, the cluster control unit 21 transfers the request from the terminal 4_yto the selected cache server 10_i(step S124). When the cache server 10_ireceives the request from the cluster control unit 21, the data distribution unit 12_ireads the streaming data stored in the streaming data storage unit 15_i, and then distributes the streaming data to the terminal 4_ythrough the cluster control unit 21 (step S125). The cache server 10_ialso notifies the cache server 10_i+1of the data contents distributed to the terminal 4_yat regular intervals (step S126).

[0110] On the other hand, if No in step S123, i.e., it is determined that the number of streams currently being transmitted by the cache server 10_iexceeds the value {((n−1)/n)×MAX}, the cluster control unit 21 further determines whether the number of streams currently being distributed by the cache server 10_i+1is less than or equal to the value {((n−1)/n)×MAX} or not (step S129).

[0111] If Yes in step S129, the cluster control unit 21 transfers the request received from the terminal 4_yto the cache server 10_i+1(step S130). When the cache server 10_i+1receives the request from the cluster control unit 21, the data distribution unit 12_i+1reads the streaming data stored in the streaming data storage unit 15_i+1, and then distributes the streaming data to the terminal 4_ythrough the cluster control unit 21 (step S131). The cache server 10_i+1also notifies the cache server 10_iof the data contents distributed to the terminal 4_yat regular intervals (step S132).

[0112] If No in step S129, the procedure goes to step S104 in FIG. 2 for further processing.

[0113] Next, a fail-safe operation is now described. In the fail-safe operation, even when any one of the cache servers in the cluster server apparatus becomes unable to carry out data distribution to the terminals due to a failure that occurred in the any one of the cache servers, another cache server takes over for the failed cache server so as to continue data distribution.

[0114] As stated above, the cache server 10_xdistributes the streaming data to the terminal 4_y, while notifying the cache server 10_x+1or 10_x−1of the information corresponding to the distributed streaming data at regular intervals (step S110, S126, S132). On the other hand, the distribution-disabled detection units 13_x+1and 13_x−1of the cache servers 10_x+1and 10_x−1, respectively, monitor the information received at regular intervals from the cache server 10_x.

[0115] Now consider the case in which the cache server 10_xhas a failure and becomes unable to carry out data distribution to the terminal 4_y. It is assumed herein that the cache servers 10_xand 10_x+1manage certain streaming data in common.

[0116] In the above case, when the distribution information, which is supposed to arrive at regular intervals from the cache server 10_x, does not arrive, the distribution-disabled detection unit 13_x+1of the cache server 10_x+1detects a distribution-disabled (occurrence of failure) state of the cache server 10_x. When the distribution-disabled detection unit 13_x+1detects the distribution-disabled state, the data distribution unit 12_x+1reads streaming data that succeeds the streaming data most recently distributed by the cache server 10_xfrom the streaming data storage unit 15_x+1based on the distribution information previously received from the cache server 10_x(before the failure occurs), and then distributes the read streaming data that succeeds the streaming data most recently distributed before the failure of the cache server 10_xto the terminal 4_y. This distribution and reading operation is carried out within a predetermined time period so as not to interrupt the streaming data. The predetermined time period means a time period during which the streaming data most recently distributed is completely played back in the terminal 4_y.

[0117] If the cache servers 10_x−1and 10_xmanage certain streaming data in common, and the cache server 10_x, which is supposed to carry out data distribution, fails, the above-described fail-safe operation is carried out by the cache sever 10_x−1.

[0118] As such, if the cache server 10_x+1or 10_x+1takes over distribution of the streaming data, the cache control unit 17_x+1or 17_x−1notifies the cluster control unit 21 of the fail-safe state so that new streaming data is not to be stored in the streaming data storage units 15_x+1or 15_x−1until the recovery of the cache server 10_xis detected. Once notified of the fail-safe state, the cluster control unit 21 restrains itself from selecting the cache server 10_x+1or 10_x−1when selecting one of the cache servers 10₁to 10_nfor storing new streaming data (steps S104, S121).

[0119] Next, a recovery operation is now described by assuming that the cache server 10_xwhich has not been able to carry out data distribution to the terminals due to a failure thereof is recovered.

[0120] The data distribution unit 12_xof the recovered cache server 10_xprovides a recovery notification to the data distribution units 12_x+1and 12_x−1of the cache servers 10_x+1and 10_x−1, respectively.

[0121] Upon receiving the recovery notification, the data distribution unit 12_x+1of the cache server 10_x+1selects the streaming data stored in both the cache servers 10_xand 10_x+1from among the streaming data stored in the streaming data storage unit 15_x+1, and then copies the selected streaming data to the streaming data storage unit 15_xof the cache server 10_x. Moreover, upon receiving the recovery notification, the data distribution unit 12_x−1of the cache server 10_x+1selects the streaming data stored in both the cache servers 10_xand 10_x−1from among the streaming data stored in the streaming data storage unit 15_x−1, and then copies the selected streaming data to the streaming data storage unit 15_xof the cache server 10_x.

[0122] Then, after the copying of the streaming data to the cache server 10_xis completed, the cache control units 17_x+1and 17_x−1of the cache servers 10_x+1and 10_x−1each send a notification of clearing the fail-safe state to the cluster control unit 21.

[0123] Specific Example of Operation

[0124] Next, the fail-safe and recovery operations are exemplarily described in further detail using a specific example.

[0125] Here, the case in which the cluster server apparatus 10 includes four cache servers 10₁to 10₄is described as the example.

[0126] FIG. 7 is a diagram exemplarily showing streaming data 601 to 60p (where p is a positive integer greater than 1) stored in the streaming data storage unit 15_xof the cache server 10_x. Each of the streaming data 601 to 60p is successive in time such as video and music data. Each of the streaming data 601 to 60p is structured by a plurality of packets 611 to 61q (where q is a positive integer greater than 1), each of which is a unit of transmission. The packets 611 to 61q have packet identifiers 621 to 62q, respectively, for identifying the order of the packets in the streaming data.

[0127] Precondition 1: State of the Streaming Data Stored in Cache Server

[0128] As stated above, in the cluster server apparatus 10, the streaming data distributed based on the request is stored in two successive cache servers which are managed in common in the system. Therefore, in the present example, as shown in FIG. 8, streaming data 701 is stored in the streaming data storage units 15₁and 15₂of the cache servers 10₁and 10₂. Similarly, streaming data 702 is stored in the streaming data storage units 15₂and 15₃of the cache servers 10₂and 10₃; streaming data 703 is stored in the streaming data storage units 15₃and 14₄of the cache servers 10₃and 10₄; and streaming data 704 is stored in the streaming data storage units 15₄and 15₁of the cache servers 10₄and 10₁.

[0129] Precondition 2: The Number of Distribution Streams

[0130] As stated above, the cluster control unit 21 transfers a request to any one of the cache servers 10₁to 10₄so as to satisfy the following conditions, where the maximum number MAX of distribution streams that a single cache server can distribute is assumed to be 8.

[0131] (Condition 1) The number of streams that can be distributed at one time by each of the cache servers 10₁to 10₄is equal to or less than 6 (=((n−1)/n)×MAX=¾×8).

[0132] (Condition 2) The number of distribution streams stored in each of the two successive cache servers managed in common in the system is equal to or less than 8 (=MAX).

[0133] Precondition 3: Session Management

[0134] A session management method carried out by the session management units 11₁to 11₄of the cache servers 10₁to 10₄is now described with reference to FIGS. 9 and 10.

[0135] The session management units 11₁to 11₄hold the number of distribution streams of the cache servers 10₁to 10₄using session management tables 53₁to 53₈. Here, each of the cache servers 10₁to 10₄holds two session management tables. That is, as shown in FIG. 9, the cache server 10₁uses the session management table 53₂for managing sessions related to the streaming data 701 stored in the cache servers 10₁and 10₂, thereby holding the state of distribution of the cache server 10₂, while also using the session management table 53₁for managing sessions related to the streaming data 704 stored in the cache servers 10₁and 10₄, thereby holding the state of distribution of the cache server 10₄. The cache server 10₂uses the session management table 53₄for managing sessions related to the streaming data 702 stored in the cache servers 10₂and 10₃, thereby holding the state of distribution of the cache server 10₃, while also using the session management table 53₃for managing sessions related to the streaming data 701 stored in the cache servers 10₂and 10₁, thereby holding the state of distribution of the cache server 10₁. The cache server 10₃uses the session management table 53₆for managing sessions related to the streaming data 703 stored in the cache servers 10₃and 10₄, thereby holding the state of distribution of the cache server 10₄, while also using the session management table 53₅for managing sessions related to the streaming data 702 stored in the cache servers 10₃and 10₂, thereby holding the state of distribution of the cache server 10₂. The cache server 10₄uses the session management table 53₈for managing sessions related to the streaming data 704 stored in the cache servers 10₄and 10₁, thereby holding the state of distribution of the cache server 10₁, while also using the session management table 53₇for managing sessions related to the streaming data 703 stored in the cache servers 10₄and 10₃, thereby holding the state of distribution of the cache server 10₃.

[0136] For example, if a stream request arrives at the cache server 10₁or 10₂for the streaming data 701 stored in the cache servers 10₁and 10₂in common, the session management tables 53₂and 53₃in FIG. 9 each register information about that request. The information registered in the session management tables 53₂and 53₃is updated as the streaming data is distributed.

[0137] Description of the session management method is now further made below with reference to FIG. 10.

[0138] To distribute a stream to the terminals 4₁to 4_m, the cache servers 10₁to 10₄are connected through the networks 51 and 52 to the terminals 4₁and 4_mand the cluster control unit 21. Now, an execution of distributing a stream among the terminals through network connection is herein called a session.

[0139] FIG. 10 is a diagram showing the session management tables 53₂and 53₃shown in FIG. 9. In each of the session management tables 53₂and 53₃, a session identifier for identifying the session and a packet identifier indicating a packet most recently sent out are registered. Assume herein that the maximum number MAX of registrable sessions in each of the session management tables 53₁to 53₈is 8.

[0140] In the session management tables 53₂and 53₃, sessions are identified in a reverse order (symmetrically in a table field). That is, the session identified by a session identifier I (0≦ I<MAX) in the session management table 53₂is identified by a session identifier (MAX−I−1) in the session management table 53₃. For example, packet identifiers “100” and “510” of the sessions identified by session identifiers “0” and “1” in the session management table 53₂are registered in the session management table 53₃as identified by session identifiers “7” and “6”, respectively.

[0141] Table boundary values 54₂of the session management table 53₂and 54₃session management table 53₃indicate numbers of the session identifiers at which the information registered in the session management tables 53₂and 53₃is divided into two. As described above, in the session management tables 53₂and 53₃, sessions are registered in the reverse order. Therefore, if the table boundary value of one session management table is set to F (0≦ F<MAX), the table boundary value of the other session management table is set to (MAX−F). In FIG. 10, the table boundary value 543 of the session management table 53₃is set to “3”, while the table boundary value 54₃of the session management table 53₃is set to “5”.

[0142] The data distribution units 12₁to 12₄of the cache servers 10₁to 10₄transmit to the terminal only streaming data with respect to a session with the session identifier I equal to or smaller than the table boundary value F. More specifically, the cache servers 10₁to 10₄repeat the following first and second steps.

[0143] First step: Compare the session identifier I with the table boundary value F. If I<F, extract, from the streaming data storage unit, a packet that immediately comes after the packet corresponding to the session identifier I in the session management table, and send the extracted packet to the terminal.

[0144] Second step: Update the value of the packet identifier corresponding to the session identifier I in the session management table to the value of the packet identifier of the packet sent out in the above first step.

[0145] For example, in FIG. 10, for the sessions identified by the session identifiers “0” to “2” in the session management table 53₂of the cache server 10₁(surrounded by a thick line in FIG. 10 on the left), the data distribution unit 121 of the cache server 10₁transmits the relevant streaming data and updates the values of the packet identifiers. Similarly, for the sessions identified by the session identifiers “0” to “4” in the session management table 53₃of the cache server 10₂(surrounded by a thick line in FIG. 10 on the right), the data distribution unit 12₂of the cache server 10₂transmits the relevant streaming data and updates the values of the packet identifiers.

[0146] Precondition 4: Operation of the Distribution-disabled Detection Unit

[0147] Each session management table includes a session with a session identifier I larger than the table boundary value F. Stream data corresponding to such session is currently distributed by the other cache server of one pair. Further, as mentioned above, the distribution-disabled detection units 13_xand 13_x+1of the cache servers 10_xand 10_x+1communicate with each other. The operation of these distribution-disabled detection units are now described with reference to FIGS. 11 and 12. FIG. 11 is a diagram showing an exemplary operation of the distribution-disabled detection units 13₁to 13_nof the cache servers 10₁to 10_n, respectively, where n=4.

[0148] Each of the distribution-disabled detection units 13₁to 13₄regularly communicates with its adjacent cache servers. For example, the distribution-disabled detection unit 13₁regularly communicates with the distribution-disabled detection units 13₂and 13₄. If the cache server 10₂or 10₄becomes unable to distribute data, the distribution-disabled detection unit 13₁instructs the data distribution unit 12₁of the cache server 10₁, instead of the cache server 10₂or 10₄, to start distribution of streaming data without interruption.

[0149] FIG. 12 is a diagram schematically showing the operation of sending and receiving a notification between the distribution-disabled detection units 13₁and 13₂. The session management tables 53₂and 53₃are held by the cache servers 10₁and 10₂, respectively.

[0150] The distribution-disabled detection units 13₁and 13₂regularly notify each other of the table boundary value and the values of the session identifier and packet identifier of the session that has been updated since the previous notification.

[0151] For example, consider the case in which, since the previous notification until the present notification the values of the packet identifiers of the sessions identified by the session identifiers “0” and “2” in the cache server 10₁have been updated from “100” and “200” to “101” and “201”, respectively. In this case, the distribution-disabled detection unit 13₁notifies the distribution-disabled detection unit 13₂of its own table boundary value and the updated packet identifier values. Moreover, in the cache server 10₂, if the packet identifier of the session identified by the session identifier “4” has been updated from “220” to “221” since the previous notification until the present notification, the distribution-disabled detection unit 13₂notifies the distribution-disabled detection unit 13₁of its own table boundary value and the updated packet identifier value. Note that the table boundary value is always notified of every time updating occurs, but is changed only when the value F or (MAX−F) for notification differs from the current value (MAX−F) or F. In the example shown in FIG. 12, the table boundary value is not changed.

[0152] With the present notification, the distribution-disabled detection unit 13₁detects that the packet identifier of the session identifier “4” has been updated to “221”. The packet identifier corresponding to the session identifier I in the session management table 53₃is stored with a session identifier “3” (=MAX−I−1) in the session management table 53₂. Therefore, the distribution-disabled detection unit 13₁updates the packet identifier value of the session identifier “3” to “221”. Similarly, with the present notification, the distribution-disabled detection 13₂detects that the packet identifiers of the session identifiers “0” and “2” have been updated to “101” and “201”, respectively. Then, the distribution-disabled detection unit 13₂updates the packet identifier values of the session identifier “7” and “5” to “10₁” and “201”, respectively.

[0153] Basic Operation 1: When the Adjacent Cache Server has a Failure

[0154] FIG. 13 shows the operation when the cache server 10₁becomes unable to carry out distribution due to a failure that occurred therein.

[0155] When the cache server 10₁becomes unable to carry out data distribution due to a failure thereof, a notification from the distribution-disabled detection unit 13₁does not come to the distribution-disabled detection unit 13₂. If the distribution-disabled detection unit 13₂does not receive notification for a predetermined time period, the distribution-disabled detection unit 13₂changes the table boundary value 54₃to MAX. In FIG. 13, MAX=8, and the table boundary value 54₃is therefore changed from “5” to “8”.

[0156] As a result, the data distribution unit 12₂of the cache server 10₂determines that the session identifiers “5” to “7” are smaller than the table boundary value F=8 in the session management table 53₃. Then, the data distribution unit 12₂starts streaming data distribution from the packet that immediately comes after the packet corresponding to the session identifier for each session.

[0157] Basic Operation 2: Session Overflow (Table Boundary Value Shift)

[0158] FIG. 14 shows the operation for changing the table boundary value (shifting the table boundary) when the number of streams that the cache server 10₁can distribute exceeds the maximum.

[0159] The cache server 10₁changes its table boundary value if any of the following conditions is satisfied:

[0160] (1) Upon detecting a failure in the cache server 10₄, the cache server 10₁starts to distribute streaming data instead of the cache server 10₄, and consequently, the number of streams that the cache server 10₁can distribute exceeds the maximum; and

[0161] (2) Upon detecting that the table boundary value of the cache server 10₄has been changed, the cache server 10₁starts streaming data distribution instead of the cache server 10₄, and consequently, the number of streams that the cache server 10₁can distribute exceeds the maximum.

[0162] In FIG. 14, the table boundary value 54₂is changed from “3” to “1”. As a result, the data distribution unit 12₁of the cache server 10₁determines that the session identifiers “1” and “2” are equal to or larger than the table boundary value F=1 in the session management table 53₂, and then the data distribution unit 12₁stops streaming data distribution. Then, the distribution-disabled detection unit 13₂is notified of the table boundary value F=1 by the distribution-disabled detection unit 13₁. The distribution-disabled detection unit 13₂detects through the notification that the table boundary value 54₂has been changed, and the distribution-disabled detection unit 13₂changes the table boundary value 54₃to (MAX−(received table boundary value 54₂)). In FIG. 14₁MAX=8, and the table boundary value 54₃is therefore changed from “5” to “7” (=8−1).

[0163] Consequently, the data distribution unit 12₂of the cache server 10₂determines that the session identifiers “5” and “6” are smaller than the table boundary value F=7 in the table management table 53₃. The data distribution unit 12₂then starts streaming data distribution from the packet that immediately comes after the packet corresponding to the session identifier for each session.

[0164] Basic Operation 3: When the Adjacent Cache Server is Recovered

[0165] As stated above, the cache control unit 17_xof the recovered cache server 10_xoutputs a recovery notification to the data distribution units 12_x+1and 12_x−1of the adjacent cache servers 10_x+1and 10_x−1. The recovered cache server 10_xrecovers the streaming data by storing the distributed streaming data in the streaming data storage unit 15_x.

[0166] FIG. 15 shows the operation when the cache server 10₁that has not been able to carry out distribution due to a failure is now recovered for streaming data distribution.

[0167] Before the streaming data of the cache server 10₁is recovered, the table boundary value 54₂is “0” (F=0), and the cache server 10₁therefore does not carry out streaming data distribution. When the recovery is completed, the recovery detection unit 18₁changes (updates) the table boundary value 54₂back to “3”, which is the value before the failure occurred.

[0168] Consequently, the data distribution unit 12₁of the cache server 10₁determines that that the session identifiers “0” to “2” are smaller than the table boundary value F=3 in the session management table 53₂. The data distribution unit 12₁then starts streaming data distribution from the packet that immediately comes after the packet corresponding to the session identifier for each session.

[0169] Next, the distribution-disabled detection unit 13₁notifies the distribution-disabled detection unit 13₂of the updated table boundary value 54₂. The distribution-disabled detection unit 13₂detects through the notification that the table boundary value 54₂is returned to “3”, which is the value before the cache server 10₁failed, and then the distribution-disabled unit 13₂changes the table boundary value 54₃to (MAX—(received table boundary value 54₂)). In FIG. 15₁MAX=8, and therefore, the table boundary value 54₃is returned from “8” to “5”.

[0170] Consequently, the data distribution unit 12₂of the cache server 10₂determines that the session identifiers “5” to “7” are equal to or larger than the table boundary value “5” in the session management table 53₃, and then the data distribution unit 12₂stops streaming data distribution.

[0171] Basic Operation 4: Returning Session (Table Boundary Value Shift)

[0172] Basic operation 4 is similar to Basic operation 2 as described above. The recovery detection unit 18₂sets the table boundary value 54₃back to the value of before the cache server 10₁failed, and also sets the other table boundary value held in the cache server 10₂(the table boundary value at which the session management table for managing the sessions of the streaming data stored in the cache servers 10₂and 10₃is divided into two) back to the original value.

[0173] Example of the Fail-safe Operation

[0174] Based on the above-described Preconditions 1 to 4 and Basic operations 1 to 4, a specific example of the fail-safe operation performed by the cluster server apparatus according to the embodiment of the present invention is shown in FIGS. 16 and 17.

[0175] FIG. 16 is a diagram showing an example of states of the session management tables 53₁to 53₈and the table boundary values 54₁to 54₈managed by the four cache servers 10₁to 10₄each having the maximum number of distribution streams “8” in normal stream distribution. Note that, in FIG. 16₁each of the table boundary values 54₁to 54₈is “4”, and the number of distribution streams (total number of sessions surrounded by a thick line in FIG. 16) that can be distributed at one time by each of the cache servers 10₁to 10₄is “6”. FIG. 17 is a diagram showing an example of states of the session management tables 53₁to 53₈and the table boundary values 54₁to 54₈managed by the cache servers 10₁to 10₄when the cache server 10₁has failed in the states shown in FIG. 16.

[0176] In the states shown in FIG. 16, when the cache server 10₁becomes unable to carry out data distribution due to a failure, the following process is performed. First, the distribution-disabled detection units 13₂and 13₄of the cache servers 10₂and 10₄detect that the cache server 10₁becomes unable to carry out distribution, and then the distribution-disabled detection units 13₂and 13₄set the table boundary values 54₃and 54₈to “8”, respectively. Accordingly, the number of distribution streams in the cache server 10₂becomes “10”. Therefore, in order to have the cache server 10₃distribute the two sessions that exceed the maximum number of distribution streams “8”, the distribution-disabled detection unit 13₂sets the table boundary value 54₄to “2”. Similarly, the table boundary value 54₅of the cache server 10₃is set to “6”, but that does not cause the number of distribution streams of the cache server 10₃to exceed the maximum number of distribution streams “8”, thereby posing no problem. Furthermore, the table boundary value 54₈of the cache server 10₄is set to “8”, but that does not cause the number of distribution streams of the cache server 10₃to exceed the maximum number of distribution streams “8”, thereby also posing no problem.

[0177] With this operation, the cache server 10₂to 10₄carry out stream distribution for each session with full capacity with the maximum number of distribution streams “8”. Therefore, if the cache server 10₁fails, the streaming data of all sessions can be distributed without interruption.

[0178] Example of Recovery Operation

[0179] In the states shown in FIG. 17, when the cache server 10₁is recovered, the operation of recovering the streaming data is performed based on the following first to sixth steps.

[0180] First step: The cache control unit 17₁of the cache server 10₁copies all streaming data stored in the streaming data storage units 15₂and 15₄of the cache servers 10₂and 10₄to the streaming data storage unit 15₁.

[0181] Second step: Based on the information from the distribution-disabled detection units 13₂and 13₄of the cache server 10₂and 10₄, the distribution-disabled detection unit 13₁registers the corresponding session identifiers and packet identifiers in the session management tables 53₁and 53₂.

[0182] Third step: The recovery detection unit 18₁of the cache server 10₁sets the table boundary values 54₁and 54₂to “4”.

[0183] Fourth step: The recovery detection unit 18₂of the cache server 10₂detects, based on the notification from the cache server 10₁, that the table boundary value 54₂has been returned to its original value, and then sets the table boundary values 54₃and 54₄to “4”.

[0184] Fifth step: The recovery detection unit 18₃of the cache server 10₃detects, based on the notification from the cache server 10₂, that the table boundary value 54₄has been returned to its original value, and then sets the table boundary value 54₅to “4”.

[0185] Sixth step: The recovery detection unit 184 of the cache server 10₄detects, based on the notification from the cache server 10₁, that the table boundary value 54₁has been returned to its original value, and then sets the table boundary value 54₈to “4”.

[0186] With the above-described operation, the states of the cache servers 10₁to 10₄are returned to the states as shown in FIG. 16. Therefore, if any one of the cache servers becomes unable to carry out distribution in the future, the fail-safe operation can be performed again.

[0187] As such, in the cluster server apparatus according to the embodiment of the present invention, one piece of streaming data is stored in a plurality of cache servers.

[0188] Thus, it is possible to directly distribute the streaming data from the plurality of cache servers, thereby achieving great improvement in distribution performance.

[0189] Further, the loads on the cache servers are equally distributed based on the streaming data stored therein, thereby allowing efficient management of storage units (caches) and data distribution to more terminals.

[0190] Still further, the streaming data is always stored in the plurality of cache servers, which are operated by communicating with each other with reference to the state of distribution of the plurality of cache servers. Thus, data distribution can be continued without interruption even if a failure of any one of the cache servers occurs.

[0191] Still further, since the streaming data is always stored in the plurality of cache servers, the quick recovery operation can be performed on the cache server that is recovered after it has failed. Also, with this recovery operation, the state of the cache server is returned to its original state before the failure occurred, thereby allowing continuation of distribution to the terminals without interruption even if the cache server again becomes unable to carry out data distribution.

[0192] Note that, in the above embodiment, a method is exemplarily described in which the cache servers 10₁to 10_ninclude the session management units 11₁to 11_nand the streaming data management units 14₁to 14_n, respectively, and return the number of streams currently being distributed and the presence or absence of the data in response to the cluster control unit 21. However, other methods can be implemented. For instance, the cluster server apparatus 10 holds one database, where all cache servers 10₁to 10_nindividually update the database, and the cluster control unit 21 refers to the database to recognize the number of streams currently being distributed and the presence or absence of the data.

[0193] Further, in the above embodiment, the load-distribution server 20 and the cache servers 10₁to 10_nare separately structured. Alternatively, any cache server 10_xmay have the function of the load-distribution server 20. Still further, all cache servers 10₁to 10_nmay alternatively have the function of the load-distribution server 20, and the cache server 10_xthat actually received a request may operate as the load-distribution server.

[0194] Still further, in the above embodiment, the maximum number of distribution streams for one cache server is MAX and, with reference thereto, the number of distribution streams that can be distributed is determined. However, if the maximum number of distribution streams cannot be uniquely determined because of the difference in performance of the cache servers, the maximum number of distributable streams may be determined in accordance with the cache server having the lowest performance or the average performance among the cache servers. Alternatively, the maximum number of distribution streams may be set to be smaller in advance than the actual maximum number of distribution streams.

[0195] While the invention has been described in detail, the foregoing description is in all aspects illustrative and not restrictive. It is understood that numerous other modifications and variations can be devised without departing from the scope of the invention.