JP5375205B2 - Common resource management system, method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a common resource management system capable of acquiring a resource without concentrating access to a specific node and without depending on a network scale. <P>SOLUTION: A common resource management table 15 manages the using state of a common resource allowed to be used in common by a plurality of nodes in a network. A common resource management table moving means 11 moves the common resource management table 15 from one node to another node. A table pointer 14 stores the identification information of a node holding the common resource management table. When the common resource management table is moved, a table pointer updating means 12 stores the identification information of the movement destination node into the table pointer 14. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a common resource management system, method, and program, and more particularly, to a common resource management system, method, and program for managing resources that can be shared by a plurality of nodes.

  In the field of distributed parallel processing, there are systems in which a plurality of nodes can use common resources. Generally, in such a system, a common resource management table is fixed to one node, and the node performs resource allocation and management. The common resource management table is used by various processes in the parallel processing system. For this reason, access concentrates on the node where the common resource management table is placed, and the access load of the node increases. In addition, since the common resource management table requires exclusive control for its access, processing related to the common resource management table affects the operational efficiency of the system.

  As a technique that can avoid concentration of access to a specific node, there is a technique described in Patent Document 1. A plurality of resource providing entities and a plurality of resource using entities are connected to the network. The resource providing entity and the resource using entity cause the network to circulate an available resource list indicating the availability of resources. When the resource providing entity receives the free resource list, the resource providing entity updates the free resource list based on the availability of resources held by the resource providing entity at that time. After updating the free resource list, the resource providing entity passes the updated free resource list to the next resource providing entity or resource using entity.

  Each resource using entity obtains a free resource list via the network. The resource use subject checks whether there is a resource that it wants to acquire in the free resource list. When there is a resource to be acquired in the free resource list, the resource using entity updates the free resource list to acquire the resource. After updating the free resource list, the resource using entity passes the free resource list to the next resource using entity or resource providing entity. Further, the resource using entity transmits a resource use request to the resource providing entity that holds the resource to be acquired, and acquires the resource from the resource providing entity.

JP-A-10-27109

  In Patent Document 1, since the free resource list circulates the network, it can be avoided that access concentrates on a specific node. However, in Patent Document 1, it is necessary to circulate the free resource list to all nodes of the network. For this reason, depending on the scale, the time required for the free resource list to circulate in the network changes, and as the number of nodes increases, the time required for resource acquisition increases. In addition, there is a problem that the resource using entity can acquire resources only at the timing when the free resource list has been circulated.

  It is an object of the present invention to provide a common resource management system, method, and program capable of accessing common resource management information without concentrating access to a specific node and without depending on the scale of a network. And

In order to achieve the above object, a common resource management system of the present invention is a common resource management system comprising a plurality of nodes connected to each other via a network, wherein at least one node of the plurality of nodes is And at least a part of the plurality of nodes, comprising identification information storage means for storing identification information of a node having a common resource management table for managing a usage status of a common resource that can be commonly used by the plurality of nodes. Each of the nodes includes management table information storage means for storing management table information indicating the identification information of the node comprising the identification information storage means, and the common resource management table from the node specified by the identification information indicated by the management table information. The identification information acquisition means for acquiring the identification information of the node that holds the A common resource management table obtaining means for obtaining the common resource management table from the node identified by the identification information is a common resource management table storage for storing the common resource management table acquired by the common resource management table obtaining means And, in response to the acquisition of the common resource management table, identification information update means for updating the identification information stored in the identification information storage means to the identification information of the node that acquired the common resource management table. It is characterized by providing.

The common resource management method of the present invention is a common resource management method for managing a common resource that can be commonly used by the plurality of nodes by a computer provided in each of the plurality of nodes connected to each other via a network. An identification information storage step in which a computer included in at least one of the nodes stores an identification information of a node having a common resource management table for managing a usage status of the common resource; A management table information storage step in which a computer provided in each of at least some nodes stores management table information indicating identification information of the at least one node; and among the computers provided in each of the at least some nodes, The computer that is trying to obtain the common resource management table Over data obtains the identification information of the node that has the common resource management table from the node identified by the identification information, wherein the management table information indicates, obtaining the common resource management table from the node identified by the identification information The common resource management table is stored in the common resource management table storage means provided in the node including the computer that has acquired the common resource management table, and the identification information stored in the computer included in the at least one node is stored in the common resource management And a common resource management table acquisition step for updating the identification information of a node including the computer that has acquired the table .

The program of the present invention is a program for causing a computer included in each of a plurality of nodes connected to each other via a network to execute processing for managing a common resource that can be commonly used by the plurality of nodes. And a computer included in at least one of the nodes executes an identification information storage process for storing identification information of a node having a common resource management table for managing a usage status of the common resource, Among the computers provided in each of the at least some nodes, the computer provided in each of the at least some nodes is caused to execute management table information storage processing for storing management table information indicating identification information of the at least one node. Trying to get the common resource management table The computer, the acquired identification information from the nodes identified by the identification information indicated by the management table information node that has the common resource management table, and acquires the common resource management table from the node identified by the identification information The common resource management table is stored in a common resource management table storage unit included in a node including the computer that has acquired the common resource management table, and identification information stored in a computer included in the at least one node is stored in the common resource management table. The common resource management table acquisition process for updating the identification information of the node including the computer that has acquired the above is executed.

  The common resource management system, method, and program of the present invention can access the management information of the common resource without concentrating access to a specific node and without depending on the scale of the network.

The block diagram which shows the common resource management system of 1st Embodiment of this invention. The flowchart which shows the operation | movement procedure at the time of common resource management table access. The flowchart which shows the operation | movement procedure at the time of a common resource management table reconstruction. The flowchart which shows the operation | movement procedure of the node at the time of an operation start. The block diagram which shows two nodes in operation. The block diagram which shows the state after a common resource management table movement. The block diagram which shows the state which cannot access the primary table pointer. The block diagram which shows the state which cannot access the primary and secondary table pointer. The block diagram which shows the state which cannot access a common resource management table. The block diagram which shows the state before reconstruction of a common resource management table. The block diagram which shows the state after the reconstruction of a common resource management table. The block diagram which shows the state of the node of 1 operation | movement. FIG. 3 is a block diagram showing a state in which two nodes are operating. The block diagram which shows the state in which the node in operation is three. The block diagram which shows the common resource management system of 2nd Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 shows a common resource management system according to the first embodiment of the present invention. The common resource management system has a plurality of nodes (nodes 10 and 20) constituting the distributed processing system. The node 10 and the node 20 are computers, for example. The node 10 and the node 20 are connected to each other via a network. In FIG. 1, the number of nodes is two, but the number of nodes may be three or more. The node 10 and the node 20 can use a common resource (not shown).

  The node 10 includes a common resource management table moving unit 11, a table pointer updating unit 12, a common resource management table restructuring unit 13, and a process 19. Further, the node 10 includes a table pointer 14, a common resource management table 15, and management table information 16 as information storage areas. The common resource management table moving unit 11 includes a common resource management table access unit 17. The function of each means in the node 10 can be realized by executing a predetermined program on the computer.

  The node 20 includes a common resource management table moving unit 21, a table pointer updating unit 22, a common resource management table restructuring unit 23, and a process 29. The node 20 includes a table pointer 24, a common resource management table 25, and management table information 26 as information storage areas. The common resource management table moving unit 21 includes a common resource management table access unit 27. The function of each means in the node 20 can be realized by executing a predetermined program on the computer. Hereinafter, each component of the node 10 will be described on behalf of the node, but the role of each component in the node 20 is the same as the role of each corresponding component in the node 10.

  The common resource management table 15 is a table that stores common resource management information (usage status, availability). The common resource is managed by a unique number in the system. Specific examples of the common resource include a plurality of registers of the switch between nodes. The common resource is not limited to a hardware resource, and may be a software resource (resource). The common resource management table moving unit 11 moves the common resource management table from one node to another node. The common resource management table 15 includes the time when the common resource management table 15 moves to the node 10. The process 19 is a part that performs processing with reference to the common resource management table.

  The table pointer 14 stores the node number (node identification information) of the node that holds the common resource management table being used. The table pointer 14 holds the last movement time of the common resource management table as additional information. The table pointer exists in at least one of a plurality of nodes constituting the network. The management table information 16 holds the node number of the node where the table pointer exists. In FIG. 1, the management table information 16 holds the node number (P) of the node that holds the table pointer used as the primary and the node number (S) of the node that holds the table pointer used as the secondary.

  The common resource management table access unit 17 accesses the common resource management table. When accessing, the common resource management table access unit 17 first refers to the management table information 16 and identifies the node where the table pointer exists. Next, the common resource management table access unit 17 refers to the table pointer held by the identified node, and identifies the node that holds the common resource management table. Thereafter, the common resource management table held by the identified node is accessed.

  The common resource management table moving unit 11 stores the contents of the common resource management table accessed by the common resource management table access unit 17 in the common resource management table 15 in the own node, and moves the common resource management table to the own node. Let At that time, the common resource management table moving unit 11 compares the movement time of the common resource management table with the current time (time when the common resource management table is accessed). The common resource management table moving unit 11 moves the common resource management table when the time difference exceeds a predetermined time. When the difference is shorter than the predetermined time, the common resource management table is not moved. The reason for doing this is to avoid frequent movement of the common resource management table and increase in network load.

  When the common resource management table moving unit 11 moves the common resource management table to its own node, the table pointer updating unit 12 stores the node number of the destination node, that is, the node number of the node 10 in the table pointer. The table pointer update unit 12 stores the last movement time of the common resource management table in the table pointer. At this time, the table pointer updating unit 12 stores the node number of the node 10 and the last movement time in the table pointer held by the node having the node number stored in the management table information 16.

  When the common resource management table access unit 17 cannot access the common resource management table, the common resource management table rebuilding unit 13 acquires the usage status of the common resource from the communicable node, and stores the common resource management table in the own node. Rebuild. Thereafter, the common resource management table restructuring unit 13 records the node having the reconstructed common resource management table, that is, the node number of the own node, in the table pointer. As a situation in which the common resource management table access means 17 cannot access the common resource management table, a situation in which the node that holds the common resource management table goes down and the common resource management table cannot be acquired from that node is conceivable. In addition, a situation may be considered in which the node that holds the table pointer goes down and the node that holds the common resource management table in use becomes unknown.

  FIG. 2 shows an operation procedure when accessing the common resource management table. Here, description will be made assuming that the node 10 accesses the common resource management table. The common resource management table access unit 17 refers to the management table information 16 when the process 19 refers to the common resource management table, and is stored in the management table information 16 as a node having a table pointer used as a primary. Is specified (step A1). The common resource management table access unit 17 determines whether it is possible to access the table pointer held by the identified node, that is, the table pointer used as the primary (step A2). When the common resource management table access unit 17 is accessible, the common resource management table access unit 17 refers to the table pointer and identifies the node that holds the common resource management table being used (step A3).

  When the common resource management table access unit 17 determines that access is not possible in step A2, the common resource management table access unit 17 refers to the management table information 16 and identifies the node recorded as a node having a table pointer used as a secondary in the management table information 16 (step A4). The common resource management table access unit 17 determines whether it is possible to access the table pointer held by the identified node, that is, the table pointer used as the secondary (step A5). When the common resource management table access unit 17 is accessible, the common resource management table access unit 17 proceeds to step A3 and refers to the table pointer to identify the node that holds the common resource management table in use.

  If the common resource management table access unit 17 determines that access is not possible in step A5, the common resource management table is reconstructed assuming that the node that holds the currently used common resource management table is unknown (step A6). . This reconstruction process will be described later. The same node number is recorded in the primary table pointer and the secondary table pointer. For this reason, whether the primary table pointer or the secondary table pointer is used, the common resource management table holding node specified in step A3 is the same node.

  The common resource management table access unit 17 determines whether or not the common resource management table possessed by the node identified in step A3 is accessible (step A7). If the common resource management table access unit 17 determines that access is not possible, the common resource management table currently in use is determined to be unusable, and the process proceeds to step A6 to reconstruct the common resource management table. When the common resource management table access unit 17 determines that access is possible in step A7, the common resource management table access unit 17 obtains the movement time from the common resource management table, and this coincides with the last movement time recorded in the primary or secondary table pointer. It is determined whether or not to perform (step A8). If the two do not match, the common resource management table access unit 17 determines that the common resource management table is invalid, proceeds to step A6, and reconstructs the common resource management table. The common resource management table access unit 17 determines that the common resource management table is valid when the times coincide.

  When the validity of the common resource management table is confirmed, the common resource management table moving means 11 determines the difference between the current time and the last movement time of the common resource management table, that is, since the common resource management table has moved last time. Find the elapsed time. The common resource management table moving unit 11 determines whether or not the elapsed time exceeds a predetermined time (maximum stop time) (step A9). This maximum stop time is determined in advance based on the system scale, operation mode, and the like. The common resource management table moving means 11 moves the common resource management table to its own node when the elapsed time from the previous movement exceeds the maximum stop time (step A10). That is, the common resource management table moving unit 11 stores the contents of the common resource management table accessed by the common resource management table access unit 17 in the common resource management table 15 of its own node. The common resource management table moving means 11 does not move the common resource management table when it is determined in step A9 that the elapsed time from the previous movement is within the maximum retention time.

  When the common resource management table is moved, the common resource management table moving unit 11 records the movement time in the common resource management table 15 moved to its own node (step A11). The table pointer updating unit 12 records the node number of the own node in the table pointer held by the node having the node number recorded in the management table information 16. The table pointer updating unit 12 records the movement time of the common resource management table recorded in step A11 as the last movement time of the table pointer (step A12). When the management table information 16 records two nodes, that is, a node that holds the primary table pointer and a node that holds the secondary table pointer, the table pointer updating unit 12 stores the table pointers held by both nodes. Record the node number of the node and the last movement time.

  The common resource management table access unit 17 updates the common resource management table in accordance with the resource acquisition or resource release of the process 19 (step A13). When the common resource management table has been moved to the own node, the common resource management table access unit 17 reflects the updated contents in the common resource management table 15 of the own node. When the common resource management table is not moved, the common resource management table access unit 17 reflects the updated contents in the common resource management table held by the common resource management table holding node specified in step A3.

  FIG. 3 shows an operation procedure of the reconstruction process. The common resource management table reconstructing means 13 collects the usage status of the common resource of each node from all the nodes including its own node and records it in the common resource management table 15 (step B1). More specifically, the common resource management table restructuring means 13 inquires each node (information of the common resource currently used by each node) of each node. The common resource management table restructuring unit 13 reflects the usage status of the common resource included in the response from the node in the common resource management table 15. When the response from the node is not obtained, the common resource management table restructuring unit 13 sets the node to non-operating and does not collect the usage status of the common resource.

  In step B1, the common resource management table restructuring unit 13 inquires the usage status of the common resource in order from the node having the weakest node number. When the common resource management table restructuring unit 13 completes the inquiry about the usage status of the common resource for all nodes, the common resource management table restructuring unit 13 records the current time as the movement time of the common resource management table 15 (step B2). The common resource management table rebuilding means 13 determines whether or not the only node existing in the system is its own node (step B3). The common resource management table rebuilding means 13 creates management table information 16 when the only node existing in the system is the own node, and records the node number of the own node in the primary of the management table information 16 ( Step B4). The secondary field is left blank (not set).

  When the common resource management table restructuring unit 13 determines in step B3 that another node exists in the system, it issues a broadcast to search for a node having a table pointer used as a primary (step B5). The common resource management table restructuring unit 13 determines whether or not a response is obtained to the issued broadcast (step B6). When the response is obtained, the common resource management table reconstruction unit 13 records the node number of the responding node in the primary of the management table information 16 (step B7).

  If the common resource management table rebuilding means 13 determines that no response is obtained in step B6, it records the node number of its own node in the primary of the management table information 16 (step B8). Thereafter, the common resource management table rebuilding means 13 notifies other nodes by broadcast that the primary of the management table information 16 has been changed in order to notify that the own node has become primary (step B9). The other nodes record the node number of the node 10 in the primary of the management table information according to the notification contents.

  The common resource management table rebuilding means 13 searches for a node having a table pointer used as a secondary after searching for a node having a table pointer used as a primary. The common resource management table restructuring unit 13 issues a broadcast to search for a node having a table pointer used as a secondary (step B10). The common resource management table restructuring unit 13 determines whether a response to the issued broadcast has been obtained (step B11). When the response is obtained, the common resource management table restructuring unit 13 records the node number of the responding node in the secondary of the management table information 16 (step B12).

  If the common resource management table restructuring unit 13 determines that no response is obtained in step B11, it determines whether or not the own node is a node having a table pointer used as a primary (step B13). The common resource management table rebuilding means 13 becomes a node having a table pointer used as a secondary with respect to another node, for example, an adjacent node, when the own node is a node having a primary table pointer. Is requested (step B14). The common resource management table restructuring unit 13 records the node number of the own node in the secondary of the management table information 16 when the own node is not the node having the primary table pointer (step B15). Thereafter, the common resource management table restructuring unit 13 notifies other nodes by broadcasting that the secondary of the management table information 16 has been changed in order to notify that the own node has become the secondary (Step S1). B16).

  The common resource management table restructuring unit 13 records the node number of the own node in the table pointer held by the node having the node number recorded in the management table information 16 (step B17). Further, the common resource management table restructuring unit 13 records the movement time recorded in the common resource management table 15 in step B2 on the table pointer held by the node having the node number recorded in the management table information 16. When the process proceeds from step B3 to step B4, the node holding the table pointer used as the primary is its own node, and there is no node used as the secondary and holding the table pointer. Therefore, the common resource management table restructuring unit 13 records the node number and movement time of the own node in the table pointer 14 of the own node. When the process proceeds from step B3 to step B4, the common resource management table restructuring unit 13 records the node number and movement time of the own node in both the table pointers used as primary and secondary.

  FIG. 4 shows an operation procedure when the node is activated. When the node starts operation, it tries to acquire management table information from another node, and determines whether or not the management table information has been acquired (step C1). If there is a node already connected to the network and the node is operating, the newly started node can acquire management table information from the operating node. If the node determines that the management table information cannot be acquired, the node performs a rebuild process of the common resource management table using the common resource management table rebuilding means (step C2).

  The procedure of the common resource management table reconstruction process in step C2 is the same as the procedure shown in FIG. The newly operated node performs a process of reconstructing the common resource management table, thereby creating a common resource management table in its own node and generating a table pointer and management table information. The node records the node number of its own node in the common resource management table location node of the table pointer. Further, the node records the node number of its own node in the primary of the management table information. The management table information secondary is left unset.

  When it is determined that the management table information has been acquired in step C1, the newly operating node stores the contents of the acquired management table information in the management table information in its own node (step C3). The node determines whether or not the node number of the node holding the table pointer used as the primary and the node number of the node holding the table pointer used as the secondary are recorded in the management table information (step C4). When both the nodes holding the table pointers used as primary and secondary are recorded, the process is terminated.

  When the node determines that one of the node holding the table pointer used as the primary and the node holding the table pointer used as the secondary is recorded in Step C3, the node from the node of the recorded node number Attempts to get the table pointer held by the node. The node determines whether or not the table pointer has been acquired (step C5). When the node cannot acquire the table pointer, the node proceeds to step C2 and reconstructs the common resource management table.

  When the node determines that the table pointer has been acquired in step C5, the node records the content of the acquired table pointer in the table pointer of its own node (step C6). Thereafter, the node records the node number of its own node in the management table information as an unset node among a node holding a table pointer used as a primary and a node holding a table pointer used as a secondary. The node declares by broadcast that its own node has become a node holding a table pointer used as a primary or a table pointer used as a secondary (step C7).

  Hereinafter, a specific case will be described. FIG. 5 shows two nodes in operation. FIG. 5 shows a case where no failure or the like has occurred. The node 20 (node number 20) is a node holding a table pointer 24 used as a primary, and the node 10 (node number 10) is a node holding a table pointer 14 used as a secondary. The node number “20” is recorded in the primary (P) of the management table information 16 of the node 10, and the node number “10” is recorded in the secondary (S). Similarly, in the node 20, the node number “20” is recorded in the primary (P) of the management table information 26, and the node number “10” is recorded in the secondary (S).

  The node that holds the common resource management table information in use is the node 20. The movement time of the common resource management table 25 held by the node 20 is time t1. In the table pointer 24 of the node 20 used as the primary, the node number “20” is recorded as the common resource management table location node. The table pointer 24 records time t1 as the last movement time. Similarly, in the table pointer 14 of the node 10 used as the secondary, the node number “20” is recorded as the common resource management table location node, and the time t1 is recorded as the final movement time.

  In FIG. 5, it is assumed that the process (proc1) 19 belonging to the node 10 needs to access the common resource management table. The common resource management table access means 17 (FIG. 1) refers to the primary (P) of the management table information 16 in the own node and acquires the node number “20”. The common resource management table access unit 17 accesses the table pointer 24 held by the node 20 and acquires the node number “20” of the node holding the common resource management table in use. Further, the common resource management table access unit 17 acquires the last movement time t1 from the table pointer 24.

  The common resource management table access unit 17 acquires the time (movement time) t1 when the common resource management table 25 has moved to the node 20 from the common resource management table 25 held by the node 20. The common resource management table access unit 17 compares the last movement time t1 recorded in the table pointer 24 with the movement time t1 recorded in the common resource management table 25. Since the two match, the common resource management table access unit 17 determines that the common resource management table 25 held by the node 20 is valid.

  The common resource management table moving unit 11 of the node 10 determines whether or not to move the common resource management table to the own node because the common resource management table is in the remote node. More specifically, first, an elapsed time after the common resource management table moves to the node 20 is obtained from the difference between the current time and the movement time t1 of the common resource management table 25. Next, the elapsed time is compared with a predetermined maximum stop time, and it is determined whether or not the elapsed time exceeds the maximum stop time. The common resource management table moving means 11 determines to move the common resource management table to the own node when the elapsed time exceeds the maximum stop time. If the elapsed time is within the maximum stop time, it is determined not to move the common resource management table.

  FIG. 6 shows a state after moving the common resource management table. The common resource management table moving unit 11 duplicates the contents of the common resource management table 25 and stores them as the common resource management table 15. In addition, the movement time t2 is recorded in the common resource management table 15. As a result, the common resource management table 15 held by the node 10 becomes available. Thereafter, the table pointer updating unit 12 refers to the management table information 16 and updates the contents of the table pointer 24 used as the primary and the table pointer 14 used as the secondary. In this update, the node number of the common resource management table location node is changed from the node number “20” to the node number “10”, and the last movement time is changed from the time t1 to the time t2.

  The process 19 performs processing such as updating on the common resource management table. If the common resource management table has moved, the process 19 reflects the updated contents in the common resource management table 15 in the own node. If there is no movement of the common resource management table, the process 19 reflects the updated contents in the common resource management table 25 held by the node 20 which is a remote node.

  FIG. 7 shows a case where the table pointer used as the primary cannot be accessed. Assume that the process of the node 40 needs to access the common resource management table. The common resource management table access means of the node 40 refers to the management table information 46 and tries to access the table pointer 34 held by the node 30 used as a primary. When the operation of the node 30 is stopped due to a down or the like, the common resource management table access unit of the node 40 fails to access.

  When the access to the table pointer 34 used as the primary fails, the common resource management table access means of the node 40 refers again to the management table information 46 and accesses the table pointer 24 held by the node 20 used as the secondary. Since node 20 is in operation, this access is successful. The common resource management table access unit refers to the table pointer 24 and acquires the node number “10” of the node that holds the common resource management table in use and the last movement time t5. The subsequent steps are the same as the case where the failure described with reference to FIG. 5 has not occurred.

  FIG. 8 shows a case where both the table pointers used as primary and secondary cannot be accessed. Assume that the process of the node 40 needs to access the common resource management table. The common resource management table access means of the node 40 refers to the management table information 46 and tries to access the table pointer 34 held by the node 30 used as a primary. When the operation of the node 30 is stopped due to a down or the like, the common resource management table access unit of the node 40 fails to access. Up to this point, the case is the same as that described with reference to FIG.

  When the access to the table pointer 34 used as the primary fails, the common resource management table access means of the node 40 refers again to the management table information 46 and tries to access the table pointer 24 held by the node 20 used as the secondary. . Node 20 is down and this access also fails. That is, the common resource management table access unit fails to access the table pointer 34 used as the primary and the table pointer 24 used as the secondary.

  If access to the table pointer fails, the node that holds the common resource management table being used becomes unknown, and the common resource management table is reconstructed according to the procedure shown in FIG. By reconstructing the common resource management table, the common resource management table is generated in the node 40. After the common resource management table is reconstructed, it is the same as the case where there is no failure shown in FIG.

  FIG. 9 shows a case where the common resource management table being used cannot be accessed. Assume that the process of the node 40 needs to access the common resource management table. The common resource management table access means of the node 40 refers to the management table information 46 and accesses the table pointer 34 held by the node 30 used as a primary. The common resource management table access unit obtains the node number “10” of the location node of the common resource management table and the last movement time t5 from the table pointer 34.

  The common resource management table access means of the node 40 accesses the common resource management table 15 of the node 10 and tries to acquire the movement time, but the node 10 is down and access fails. Since the common resource management table cannot be accessed, the common resource management table is reconstructed according to the procedure shown in FIG. By reconstructing the common resource management table, the common resource management table is generated in the node 40. After the common resource management table is reconstructed, it is the same as the case where there is no failure shown in FIG.

  FIG. 10 shows a case where the common resource management table is reconstructed in the node 30. The node 20 is a node holding a table pointer 24 used as a primary, and the node 10 is a node holding a table pointer 14 used as a secondary. The node having the common resource management table in use is the node 10. The common resource management table access unit of the node 30 accesses the common resource management table 15 of the node 10 after accessing the table pointer 24 used as the primary and held by the node 20. The node 10 is down, and the common resource management table access unit fails this access. If the access to the common resource management table 15 fails, the common resource management table reconstructing means of the node 30 reconstructs the common resource management table in its own node.

  The common resource management table rebuilding means of the node 30 inquires the information (usage status) of the currently used common resource to all nodes including the own node, and collects the usage status of the common resource. The common resource management table restructuring unit reflects the collected usage status in the common resource management table 35 in the own node. At this time, a node that does not receive a response to the inquiry does not collect the usage status because it is not operating. The common resource management table restructuring unit collects the usage status of the common resource from all operating nodes, and when the common resource management table is completed, records the current time as the movement time of the common resource management table.

  Subsequently, the common resource management table restructuring unit of the node 30 broadcasts and waits for a response in order to search for a node having a table pointer used as a primary. In FIG. 10, the node 20 corresponds and the node 20 responds to the broadcast. The common resource management table reconstruction means records the node number “20” of the node 20 in the primary (P) of the management table information 36. The common resource management table reconstructing means continues broadcasting to wait for a response in order to search for a node having a table pointer to be used as a secondary. In FIG. 10, although the node 10 corresponds, the node 10 is down and there is no node that responds to the broadcast.

  When there is no response from the node having the secondary table pointer, the common resource management table rebuilding means of the node 30 records the node number “30” of the own node in the secondary (S) of the management table information 36. Further, the common resource management table restructuring means performs broadcasting for declaring that the own node becomes a node having a table pointer used as a secondary. Thereafter, the common resource management table restructuring unit refers to the management table information 36 and updates the table pointer 24 held by the node 20 used as the primary and the table pointer 34 of the own node used as the secondary.

  FIG. 11 shows a state after the common resource management table is reconstructed. The common resource management table restructuring means of the node 30 declares that the node becomes a node having a table pointer used as a secondary. The other nodes 20 and 40 rewrite the node number recorded in the secondary (S) in the management table information 26 and 46 from the node number “10” to the node number “30” according to the contents of the declaration. Further, after referring to the management table information 36, the common resource management table restructuring unit records the node number “30” in the common resource management table location node of the table pointer 24 used as the primary and the table pointer 34 used as the secondary. By doing so, the node accessing the common resource management table thereafter accesses the common resource management table 35 held by the node 30.

  When a plurality of nodes detect that access to the common resource management table is disabled, there is a possibility that the common resource management table is reconstructed in parallel at the plurality of nodes. In order to prevent this, the common resource management table rebuilding means notifies all nodes that the rebuilding of the common resource management table is completed. The node that has received the notification of the completion of rebuilding cancels the rebuilding process and discards the created common resource management table. By doing so, it is possible to prevent the common resource management table from being created in duplicate.

  FIG. 12 shows a state when the node is started up. FIG. 12 shows a state in which there is no operation node in the system and the node 30 has newly started operation. The node 30 tries to acquire the management table information from the adjacent node. However, since the active node is only the node 30, the management table information cannot be acquired. The node 30 reconstructs the common resource management table and generates a table pointer 34, a common resource management table 35, and management table information 36.

  After creating the common resource management table 35, the node 30 records the created time t4 in the common resource management table 35. Further, the node 30 records the node number “30” of its own node in the common resource management table location node of the table pointer 34. As the final movement time, time t4, which is the time when the common resource management table is created, is recorded. The node 30 records the node number “30” of the own node in the primary (P) of the management table information 36. The secondary (S) is left blank (no setting).

  FIG. 13 shows a state in which the node 20 has started operation. When the node 20 starts operating from the state shown in FIG. 12, the node 20 acquires the contents of the management table information 36 from the node 30 that is an adjacent node, and reflects the contents in the management table information 26 of the own node. The contents of the management table information acquired at this time are the contents of the management table information 36 shown in FIG. 12, where only the primary (P) is set (node number “30”), and the secondary (S) is not set. .

  The node 20 refers to the primary (P) of the management table information 26 and checks whether the table pointer 34 of the node 30 is accessible. The node 20 acquires the contents of the table pointer 34 and reflects this in the table pointer 24 of its own node. That is, the node number “30” is recorded in the common resource management table location node of the table pointer 24, and the time t4 is recorded as the last movement time. Thereafter, the node 20 records the node number “20” of the own node in the secondary (S) of the management table information 26. In addition, the node 20 performs a broadcast declaring that it becomes a node having a table pointer used also as a secondary. The node 30 records the node number “20” in the secondary (S) of the management table information 36 by declaring that the node 20 becomes a node having a secondary table pointer.

  FIG. 14 shows a state where the node 40 has further started operation. When the node 40 starts operating from the state shown in FIG. 13, the node 40 acquires the contents of the management table information from the adjacent nodes and reflects the contents in the management table information 46 of the own node. The acquisition source of the contents of the management table information may be either the management table information 26 of the node 20 or the management table information 36 of the node 30. In the system, there are already a table pointer 34 used as a primary and a table pointer 24 used as a secondary. Therefore, it is not necessary to create a new table pointer in the node 40, and the node 40 has only the management table information 46. Thereafter, similarly to the node 40, the node that newly starts operation generates management table information.

  The common resource management system of this embodiment includes a common resource management table moving unit and a table pointer updating unit as a minimum configuration. The common resource management table moving means moves the common resource management table from one node to another node. When the common resource management table is moved, the table pointer update means stores identification information of the destination node in the table pointer. In this embodiment, since the common resource management table moving means moves the common resource management table, the common resource management table is not fixed to one node. For this reason, access to the common resource management table is not concentrated on a specific node, and load distribution can be expected as compared with the case where the common resource management table of the specific node is fixed.

  In Patent Document 1, since it is necessary to circulate the free resource management list to all nodes, the time required for resource acquisition becomes longer depending on the number of nodes. In the present embodiment, since it is not necessary to make the common resource management table go through all the nodes, the time required for resource acquisition does not increase depending on the network scale. Further, in Patent Document 1, each resource using entity can acquire resources only at the timing when the free resource management list has been circulated. In this embodiment, since the node that holds the common resource management table in use is managed using the table pointer, each node refers to the table pointer at any point in time in the common resource management table. Accessible. Further, in Patent Document 1, in order to circulate the resource management list, it is necessary to be able to write logically for each node in the system. However, in the present embodiment, there is no such restriction.

  In this embodiment, the common resource management table is moved to the node that has accessed the common resource management table. After the movement of the common resource management table, the destination node can access the common resource management table by local access. For example, when a process of a node performs a process such as a search on the common resource management table, the process often accesses the common resource management table a plurality of times in a short time. In the present embodiment, since the common resource management table is moved to the accessed node, the access after the movement can be speeded up. As a result, the processing speed can be increased.

  In this embodiment, two primary and secondary table pointers indicating nodes that hold a common resource management table are prepared, and the two nodes hold table pointers. If the table pointer is lost, the location of the common resource management table becomes unknown, and each node cannot access the common resource management table. In this embodiment, since there are two table pointers, primary and secondary, even if the primary table pointer cannot be referenced due to a node going down, the secondary table pointer is accessed and the common resource management table is held. It is possible to specify a node to perform. Thus, by providing a plurality of nodes having table pointers, improvement in fault tolerance can be expected.

  In this embodiment, when it becomes impossible to access the common resource management table, the common resource management table reconstructing means collects the usage status of the common resource from each node and reconstructs the common resource management table. The common resource management table restructuring unit records the identification information of the node having the reconstructed common resource management table in the table pointer, so that each node can access the reconstructed common resource management table. Therefore, in this embodiment, even when the node that holds the common resource management table goes down, for example, when the common resource management table becomes inaccessible, the common resource management table is reconstructed to remove the down node. Can be operated continuously.

  FIG. 15 shows a common resource management system according to the second embodiment of this invention. The common resource management system includes nodes 10a and 20a and a crossbar switch 50. The crossbar switch 50 is a network switch that connects nodes. The crossbar switch 50 includes a system shared register 51 and a user shared register 52. The system shared register 51 is a register used by the system, and the user shared register 52 is a register that can be used by a user in a distributed processing program or the like.

  The user shared register 52 corresponds to a common resource shared by a plurality of nodes. The system shared register 51 stores a table pointer 53. The system shared register 51 can be uniquely accessed from each node. Each node has a common resource management table (user shared register free list) as a storage area. Although not shown in FIG. 15, each node has a common resource management table moving unit, a common resource management table restructuring unit, and a common resource management table access unit.

  In FIG. 15, the node 20 a holds the user shared register free list 25 in use, and the node number “20” of the node 20 a is recorded in the table pointer 53 in the common resource management table location node. Yes. In the first embodiment, table pointers are arranged in a plurality of nodes, and a node holding a table pointer used as a primary and a node holding a table pointer used as a secondary are managed using management table information. In the present embodiment, since the table pointer 53 is fixedly arranged on the crossbar switch 50, management table information is not necessary.

  Only one table pointer 53 is arranged in the system shared register 51 of the crossbar switch 50 for the following reason. That is, if the crossbar switch 50 fails for some reason, in addition to being unable to access the connected nodes, each node cannot access the system shared register 51 and the user shared register 52. That is, the table pointer 53 cannot be accessed and the common resource cannot be used. Even if some node has a table pointer used as a secondary, if the crossbar switch 50 goes down, the common resource cannot be used. Therefore, it is not necessary to provide redundancy for the table pointer, and only one table pointer is sufficient for the crossbar switch 50.

  Also in the present embodiment, the movement of the user shared register free list, which is a common resource management table, is the same as in the first embodiment. That is, each node moves the user shared register free list to its own node when accessing the user shared register free list. Therefore, in this embodiment, the same effect as that of the first embodiment can be obtained. That is, it is possible to achieve load distribution by preventing concentration of accesses to specific nodes. Further, the access when the same node accesses after moving the user shared register free list can be a local access, and the processing speed can be increased.

  In the present embodiment, the table pointer 53 is arranged in the system shared register 51 in the crossbar switch 50. Accordingly, each node accesses only the crossbar switch 50 when referring to the table pointer. When any node has a table pointer, the access source node accesses the access destination node through the crossbar switch 50. In this embodiment, since the table pointer 53 can be accessed by accessing up to the crossbar switch 50, an effect of reducing the network load can be expected as compared with the configuration in which the node has the table pointer.

  In the present embodiment, the table pointer 53 is fixedly arranged in the crossbar switch 50, so that management table information is not necessary. When using the management table information, in order to access the user shared register free list, it is necessary to first refer to the management table information and specify the node that holds the table pointer. In this embodiment, the procedure for identifying the node that holds the table pointer by referring to the management table information can be omitted, so that the access to the user shared register free list can be speeded up.

  In the present embodiment, the table pointer 53 is disposed on the crossbar switch 50. By moving the table pointer from the node to the crossbar switch 50, the node dependency ratio of the fault tolerance as the system can be reduced. This means that the fault tolerance of the system operation can be improved by increasing the fault tolerance of the crossbar switch 50. In particular, in a large-scale system with a large number of nodes, the configuration in which the table pointer is arranged in the crossbar switch 50 is an advantageous configuration.

  Also in the present embodiment, when the user shared register free list is lost due to a node down or the like, the common resource management table rebuilding means inquires the usage status of the user shared register 52 to each node, and shares the user shared list. Rebuild the register free list. By reconstructing the shared register free list for users, it is possible to continue the operation of the system except for the down node.

  In each of the above embodiments, each node has a common resource management table moving unit, a table pointer updating unit, and a common resource management table reconstruction. However, all the nodes have these three units. It is not necessary to have one means in the system. The common resource management table moving means moves the common resource management table to the node that has accessed the common resource management table, but the present invention is not limited to this. From the viewpoint of load distribution, the determination of the destination is arbitrary, and the effect of load distribution can be obtained even if the destination is not the accessed node.

  As mentioned above, although this invention was demonstrated based on the suitable embodiment, the common resource management system of this invention, a method, and a program are not limited only to the said embodiment, From the structure of the said embodiment. Various modifications and changes are also included in the scope of the present invention.

  The present invention can be applied to the field of HPC (High Performance Computing).

10, 20, 30, 40: nodes 11, 21: common resource management table moving means 12, 22: table pointer updating means 13, 23: common resource management table restructuring means 14, 24, 34: table pointers 15, 25, 35: Common resource management tables 16, 26, 36, 46: Management table information 17, 27: Common resource management table access means 19, 29: Process 50: Crossbar switch 51: System shared register 52: User shared register 53 : Table pointer

Claims (19)

A common resource management system comprising a plurality of nodes connected to each other via a network,
At least one of the plurality of nodes is
An identification information storage means for storing identification information of a node having a common resource management table for managing a use status of a common resource that can be commonly used by the plurality of nodes;
At least some of the plurality of nodes are respectively
Management table information storage means for storing management table information indicating identification information of a node comprising the identification information storage means;
Identification information acquisition means for acquiring identification information of a node holding the common resource management table from a node specified by the identification information indicated by the management table information;
Common resource management table acquisition means for acquiring the common resource management table from a node specified by the identification information acquired by the identification information acquisition means;
Common resource management table storage means for storing the common resource management table acquired by the common resource management table acquisition means;
In response to said common resource management table is acquired, and an identification information updating means for updating the identification information the identification information storage means stores the identification information of the node that acquired the common resource management table, the common Resource management system. Wherein each of the plurality of nodes, comprising said management table information storage means, the identification information acquiring means, and said common resource management table obtaining means,
Each said common resource management table obtaining means accesses the node that has the pre-Symbol Common resource management table, and acquires the common resource management table, the common resource management system according to claim 1. The common resource management table obtaining unit, the access to the common resource nodes with the management table, when the time and the difference between the current time of the common resource management table is stored in the above predetermined time, the common from the node Gets the resource management table, the common resource management table to update the time to store the current time, the common resource management system according to claim 1 or 2. At least two of the plurality of nodes include the identification information storage unit,
The management table information includes identification information of a node provided with an identification information storage means used as a primary among nodes provided with the identification information storage means, and identification information of a node provided with an identification information storage means used as a secondary ,
The identification information obtaining means, when said identification information storage means used as the primary is available by referring to the identification information storage means, when the identification information storage means for use as the primary is not available, identifying information used as the secondary The common resource management system according to any one of claims 1 to 3 , wherein the storage unit is referred to. Each of at least some of the plurality of nodes is configured so that, when the common resource management table acquisition unit cannot access the node having the common resource management table, the at least one communicable node transmits the common resource. get the usage, as well as reconstructing the common resource management table, the common resource management for updating identification information the identification information storage unit stores the identification information of the node that has a common resource management table reconstructed the The common resource management system according to any one of claims 1 to 4 , further comprising a table restructuring unit. Each of the plurality of nodes includes the common resource management table restructuring unit ,
6. The common resource management system according to claim 5 , wherein each of the common resource management table rebuilding means rebuilds a common resource management table in a node including the common resource management table rebuilding means . The common resource management table restructuring unit is configured such that when the common resource management table acquisition unit accesses the node having the common resource management table and there is no response from the node, or the identification information acquisition unit performs the identification 7. The common resource management system according to claim 5 or 6 , wherein when the information storage means cannot be referred to, the common resource management table is reconstructed. A common resource management method for managing a common resource that can be commonly used by a plurality of nodes by a computer provided in each of a plurality of nodes connected to each other via a network,
An identification information storage step in which a computer provided in at least one of the plurality of nodes stores identification information of a node having a common resource management table for managing a usage status of the common resource;
A management table information storage step in which a computer provided in each of at least some of the plurality of nodes stores management table information indicating identification information of the at least one node;
Among the computers provided in each of the at least some nodes, the computer that is to acquire the common resource management table holds the common resource management table from the node specified by the identification information indicated by the management table information The common resource management table is acquired from the node specified by the identification information , the common resource management table is stored in the common resource management table storage means included in the node having the computer that has acquired the common resource management table. A common resource management table storing step for storing a table and updating identification information stored in a computer included in the at least one node to identification information of a node including the computer that acquired the common resource management table. Method. In the management table information storage step, a computer provided in each of the plurality of nodes stores the management table information,
In the common resource management table acquisition step, among the computers included in each of the plurality of nodes, a computer that intends to acquire the common resource management table starts from the node specified by the identification information indicated by the management table information. The common resource management method according to claim 8 , wherein identification information of a node having a resource management table is acquired , the node specified by the identification information is accessed, and the common resource management table is acquired . Wherein in the common resource management table acquisition step, the computer trying to acquire the common resource management table, accesses the node that has the common resource management table, the difference between the time and the current time at which the common resource management table stores 10. The common resource management method according to claim 8 , wherein the common resource management table is acquired from the node and the time stored in the common resource management table is updated to the current time when the time is equal to or longer than a predetermined time. In the identification information storing step, a computer provided in at least two of the plurality of nodes stores identification information of a node that holds the common resource management table;
The management table information among the nodes comprising a computer which stores the identification information includes identification information of nodes that Ru is used as the primary, and the identification information of the node that Ru is used as a secondary,
Prior Symbol Common Resource Management table acquisition step, the computer trying to acquire the common resource management table, when the node is used as the primary is available by referring to the identification information computer to which the node is provided to store, The common resource management method according to claim 8 , wherein when a node used as the primary cannot be used, identification information stored in a computer included in the node used as the secondary is referred to. When the computer which is going to acquire the common resource management table cannot access the node having the common resource management table , it acquires the usage status of the common resource from at least one communicable node, and the common resource management Reconstructing the common resource management table for reconstructing the table and updating the identification information stored in the computer storing the identification information in the identification information storage step to the identification information of the node having the reconstructed common resource management table The common resource management method according to claim 8 , further comprising a step. Wherein in the common resource management table rebuild step, the common resource management trying to get a table computer, wherein when there is no response from the common resources the node when accessing nodes with the management table, or the 13. The common resource management method according to claim 12 , wherein when the identification information of a node that holds the common resource management table cannot be referred to, the common resource management table is reconstructed. A program for causing a computer included in each of a plurality of nodes connected to each other via a network to execute processing for managing common resources that can be commonly used by the plurality of nodes,
Causing a computer included in at least one of the plurality of nodes to execute an identification information storage process for storing identification information of a node having a common resource management table for managing a usage state of the common resource;
Causing a computer included in each of at least some of the plurality of nodes to execute management table information storage processing for storing management table information indicating identification information of the at least one node;
Among the computers included in each of the at least some nodes, a node that holds the common resource management table from a node specified by the identification information indicated by the management table information in a computer that is to acquire the common resource management table The common resource management table is acquired from the node specified by the identification information , the common resource management table is stored in the common resource management table storage means included in the node having the computer that has acquired the common resource management table. A program for storing a table and executing a common resource management table acquisition process for updating identification information stored in a computer included in the at least one node to identification information of a node including the computer that acquired the common resource management table . In the management table information storage process, the management table information is stored in a computer provided in each of the plurality of nodes,
In the common resource management table acquisition process, among the computers included in each of the plurality of nodes, the computer that is trying to acquire the common resource management table receives the common from the node specified by the identification information indicated by the management table information. The program according to claim 14 , wherein identification information of a node having a resource management table is acquired, the node specified by the identification information is accessed, and the common resource management table is acquired . In the common resource management table acquisition process, the computer that is acquiring the common resource management table is made to access the node that holds the common resource management table, and the difference between the time stored in the common resource management table and the current time is stored. The program according to claim 14 or 15 , wherein when the value is equal to or longer than a predetermined time, the common resource management table is acquired from the node, and the time stored in the common resource management table is updated to the current time . In the identification information storing process, the identification information of the node holding the common resource management table is stored in a computer provided in at least two of the plurality of nodes,
The management table information among the nodes comprising a computer which stores the identification information includes identification information of nodes that Ru is used as the primary, and the identification information of the node that Ru is used as a secondary,
In the previous SL common resource management table obtaining process, to a computer trying to acquire the common resource management table, when the node is used as the primary is available to see the identification information computer to which the node is provided to store the The program according to any one of claims 14 to 16 , wherein when a node used as a primary cannot be used, identification information stored in a computer included in the node used as a secondary is referred to . When the computer that is trying to acquire the common resource management table cannot access the node that holds the common resource management table , the usage status of the common resource is acquired from at least one communicable node, and the common resource management Reconstructing the table and reconstructing the common resource management table for updating the identification information stored in the computer storing the identification information in the identification information storage process to the identification information of the node having the reconstructed common resource management table The program according to any one of claims 14 to 17 , further executing a process . Wherein a common resource management table rebuild process, the computer trying to acquire the common resource management table, when there is no response from the node when accessing node that has the common resource management table, or the common when it is not possible to refer to the identification information of the node that has the resource management table, makes the reconstruction of the common resource management table, the program of claim 18.

Images