JP2009140224A - Inter-node data transfer controller, inter-node data transfer control method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently allocate an area to a distributed shared memory shared between nodes, and to release the allocation while maintaining efficiency in performing allocation. <P>SOLUTION: A node address association means 1-B has a means converting a memory address into a node address and a table wherein the memory address and the node address are associated, or has a means corresponding to calculation associating the memory address and the node address. The memory address is converted into the node address based on a prescribed rule such as retrieval of the table, partial masking or a shift of the memory address, or the calculation. When the node address association means 1-B releases the allocation of a part or all of the memory area allocated for the distributed shared memory 1-C on a local memory as the distributed shared memory 1-C, the node address association means 1-B precedingly releases the allocation of the area present in an address different form initial allocation inside the area of the distributed shared memory 1-C to improve use efficiency of a local memory area. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a memory configuration of a transfer control device using a distributed shared memory that transfers data within a node or between nodes, and particularly in a distributed shared memory and a local memory configuration within each node. The present invention relates to an inter-node data transfer control device, an inter-node data transfer control method, and a program that perform other settings on the shared memory and cancel the allocation of the distributed shared memory.

  An inter-processor (or inter-node) data transfer control device using a distributed shared memory usually uses existing transmission systems such as LAN, ATM, TCP / IP, etc., and node addresses (IP Data is transferred by associating an address or the like, but not an IP address, as long as it is an identifier that identifies a node in the network with a distributed shared memory address.

Conventionally, in the inter-processor data transfer control device, when a part of the local memory is allocated to the distributed shared memory and used, the local memory already allocated for the distributed shared memory is unallocated as the distributed shared memory. Depends on the means for allocating for distributed shared memory. For this reason, a technique has been proposed in which the user side prepares an allocation method (see, for example, Patent Document 1).
JP 10-133941 A

  By the way, the distributed shared memory is a part of local memory allocated for inter-node data transfer. Therefore, a sufficient area is not always secured for inter-node data transfer. A means to allocate to the shared memory is required. In the distributed shared memory currently in use, determine whether there is any memory area that exceeds a certain memory size in the set of individual memory areas corresponding to individual nodes in the network. A new memory area is allocated to the set of memory areas that have been allocated, and the function related to the node address handling method is modified for the allocated memory areas, thereby ensuring additional memory areas. Used memory area.

  Since the distributed shared memory is a part of the local memory, the location is identified only by the memory address, and the size of the newly allocated memory area is also the memory area of the currently used distributed shared memory. If it is the same as the size of the set, the assignment is easy. However, since it is necessary to determine the size excess for a set of individual memory areas, it is possible to determine the size excess and perform allocation by applying the memory configuration and processing flow that identifies the usage status. Necessary. In addition, since a new memory allocation is insufficient or wasted, it is necessary to apply a memory configuration and a processing flow in which the usage status is identified. In addition, it is necessary to associate a plurality of divided networks with a memory configuration adapted to the network structure and an allocation process flow. In this case, in order to perform efficient deallocation, a means based on the premise of deallocation is incorporated in the allocation stage, and at the time of deallocation, an efficient memory allocation state based on the current memory usage status and network classification Therefore, it is necessary to prepare a means that does not decrease the memory read / write speed.

  However, in the prior art, although the assigning unit functions effectively, it is not considered to use the assigning unit as an effective release unit. In other words, in the conventional technology, in the distributed shared memory that is currently used, whether a set of individual memory areas corresponding to individual nodes in the network exceeds a certain memory size. And determining a new memory area for the set of excess memory areas, and correcting the node address correspondence function for the reserved memory area, The memory area secured in addition to was used. The size of the newly secured memory area is also the same as the size of the set of memory areas of the distributed shared memory currently used.

  Furthermore, it is determined whether or not a certain memory size has been exceeded for each set of memory areas, and a larger or more memory area is allocated even if it does not require a large amount of newly allocated memory area. However, there is a problem that the allocation size is small. In addition, since the distributed shared memory allocation method is common to all nodes, when the network is divided into a plurality of sections, the correspondence between the distributed shared memory and the nodes is complicated, but these are complicated. There is a problem that the design considering the above is not made.

  Therefore, even when deallocating, since it is randomly deallocated for each fixed size, even if there is a lot of unused memory range, it remains unallocated, and the memory usage efficiency It becomes worse, or distributed shared memory targeting the same node exists discontinuously on local memory, or distributed shared memory targeted to the same network partition exists discontinuously on local memory As a result, there is a problem that the memory read / write time is increased.

  The present invention has been made in consideration of such circumstances, and the purpose thereof is to enable efficient allocation and maintain efficiency at the time of allocation even when allocation becomes unnecessary later. An object is to provide an inter-node data transfer control device, an inter-node data transfer control method, and a program that can be deallocated.

  In order to solve the above-described problem, the present invention provides a data transfer between a plurality of nodes connected to a network using a distributed shared memory allocated on a local memory included in each node. An inter-node data transfer control device for controlling a distributed shared memory whose memory addresses are virtually or physically the same among the plurality of nodes, and an area corresponding to the node according to the addition of the nodes Area allocating means for allocating to the distributed shared memory, and area releasing means for releasing the area allocated to the distributed shared memory in response to the reduction of the node, the area releasing means releasing the area In this case, the inter-node data transfer control device is characterized in that the area allocated later by the area allocating unit is released first.

  In the above invention, the present invention further includes a determination unit that determines whether an area allocated to the distributed shared memory exceeding a predetermined threshold is used in units of groups based on a connection relationship of the plurality of nodes. The area allocating means allocates a new area to the distributed shared memory in units of groups when the determining means determines that the area is used exceeding a predetermined threshold. And

  According to the present invention, in the above invention, the predetermined threshold value is at least one or more, and the area allocating unit stores the distributed shared memory in the group unit based on the number of predetermined threshold values exceeding the predetermined threshold value. The capacity of the area to be allocated is determined.

  The present invention is characterized in that, in the above-mentioned invention, the area releasing means releases an area allocated later in units of groups first.

  Further, in order to solve the above-described problem, the present invention uses a distributed shared memory allocated on a local memory included in each node among a plurality of nodes connected to a network, to communicate with other processes. An inter-node data transfer control method for controlling data transfer, wherein an area corresponding to a node is distributed in such a manner that memory addresses are virtually or physically the same among the plurality of nodes according to the addition of the node. Allocating to shared memory, and releasing the area allocated later when releasing the area allocated to the distributed shared memory in accordance with the reduction of the node. A feature is a data transfer control method between nodes.

  The present invention further includes a step of determining whether an area allocated to the distributed shared memory exceeding a predetermined threshold is used in a group unit based on a connection relation of the plurality of nodes in the above invention, When it is determined that an area is used exceeding the predetermined threshold, a new area is allocated to the distributed shared memory in units of groups.

  According to the present invention, in the above invention, the predetermined threshold is at least one or more, and a capacity of an area to be allocated to the distributed shared memory in units of groups is determined based on the number of the predetermined thresholds exceeding the predetermined threshold. The method further includes the step of determining.

  The present invention is characterized in that, in the above invention, when the allocated area is released, the area allocated in the group unit is released first.

  Further, in order to solve the above-described problem, the present invention uses a distributed shared memory allocated on a local memory included in each node among a plurality of nodes connected to a network, to communicate with other processes. According to the increase in the number of nodes, the computer corresponding to the node-to-node data transfer control device that controls data transfer has an area corresponding to the node, and the memory addresses are virtually or physically the same among the plurality of nodes. A step of allocating to a certain distributed shared memory, and a step of releasing the area allocated later when releasing the area allocated to the distributed shared memory according to the reduction of the node. It is a program characterized by making it carry out.

  According to the present invention, when a node is added, an area corresponding to the node is allocated to the distributed shared memory, and when the area allocated to the distributed shared memory is released according to the reduction of the node, Since the allocated area is released first, distributed shared memory can be used efficiently, and even if allocation becomes unnecessary later, allocation is performed while maintaining efficiency at the time of allocation. The advantage is that the area can be released.

  Further, according to the present invention, it is determined whether the area allocated to the distributed shared memory exceeding the predetermined threshold is used in units of groups based on the connection relationship of the plurality of nodes, and the predetermined threshold is exceeded. When an area is determined to be in use, a new area is allocated to the distributed shared memory in units of groups, so the area used in the distributed shared memory exceeds the specified memory size. In the case where there is a possibility that an area shortage may occur, there is an advantage that it is possible to efficiently determine that the size is exceeded and to perform allocation efficiently.

  According to the present invention, the capacity of the area to be allocated to the distributed shared memory is determined in units of the group based on the number of predetermined thresholds that have been exceeded. Thus, it is possible to obtain an advantage that it is possible to efficiently determine the excess of the size and perform the allocation efficiently.

  Further, according to the present invention, since the area allocated in units of groups later is released first, even if the allocation becomes unnecessary later, the allocation is released while maintaining the efficiency at the time of allocation. The advantage is that it can be done.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
At least one distributed shared memory address is associated with a node address, and data is sent to the corresponding node by designating the memory address. Even with commercially available memory, the number of addresses is sufficiently large, so normally there is no shortage of memory area associated with the node, but if for some reason the number of memory addresses is insufficient, use virtual memory addresses The number of memory addresses to be associated can be increased.

  However, when the number of memory addresses is insufficient or in a network structure in which another network (lower network) is connected to a lower level of the node, a child number is prepared for the node identifier, and hierarchically An area of the distributed shared memory is configured. In this case, the network is divided through the node to which the lower network is connected. In other words, the nodes belonging to the lower network share the parent (upper) node address, and the child (lower) node addresses are different so that the distributed shared memory area is configured. When it is difficult to allocate the memory area to the inter-node structure, such as when the structure between the nodes is ring-shaped, the memory configuration is configured to be close to or close to the structure of the node relationship.

  FIG. 1 is a block diagram showing an internal configuration of a node that communicates with a communication system according to an embodiment of the present invention. In the figure, the communication system includes a network 1-I and a plurality of nodes 1-F and 1-J to 1-N connected to the network 1-I. In the following description, only the configuration of the node 1-F will be described, but the other nodes 1-J to 1-N have the same configuration.

  The node 1-F includes application programs 1-A and 1-A ′, a node address conversion unit 1-B, a distributed shared memory 1-C, and a transfer control unit 1-D. It is connected. In the internal bus 1-E shown in FIG. 1, an alternate long and short dash line and a dotted line indicate the flow of data addresses. For example, the application program 1-A writes data to the distributed shared memory 1-C secured on the local memory. The distributed shared memory 1-C is a memory area secured on the local memory and shared by each node. Node address correspondence means (area allocation means, area release means, determination means) 1-B refers to the memory address of the data flowing on the internal bus 1-E, converts the memory address to the node address, and transfers control means 1 -Notify D. The transfer control unit 1-D sends data to the network 1-I toward the notified node address.

  In the node address correspondence means 1-B, the means for converting the memory address into the node address includes (a) a table in which the memory address and the node address are associated, and this table is searched or ( b) A node address is obtained by processing a memory address based on a predetermined rule such as a shift of a memory address, a partial mask, or an operation.

  Although “notification” is described, the transfer control means 1-D inquires of the node address correspondence means 1-B or refers to the converted node address in the node address correspondence means 1-B. I do not care.

  In the node that has received the data, the memory address of the data is referred to and the data is written to the memory address on the distributed shared memory secured on the local memory of the receiving node. As a result, the transmission node and the reception node share the same data in the same memory address of the distributed shared memory in the own node.

  For example, when the application program 1-A writes the data 1-G to the distributed shared memory 1-C, the node address handling means 1-B sets the memory address portion of the data 1-G flowing on the internal bus 1-E. The memory address “1-C-1” is converted into a node address with reference to the transfer control means 1-D. The transfer control means 1-D sends data 1-G to the network 1-I toward the notified node address 1-H.

  In the node that has received the data 1-G, by referring to the memory address of the data 1-G, the data 1-G is written to the memory address on the distributed shared memory secured on the local memory of the receiving node. The transmission node and the reception node share the same data at the same address of the distributed shared memory in the own node. In FIG. 1, as an example in which the node 1-F and the nodes 1-J to 1-N transmit and receive data, the data is addressed to the node 1-M, that is, the receiving side is the node 1-M. In the node 1-M, the transfer control means 1-D ′ writes the data to the same memory address 1-C-1 as the transmission side on the distributed shared memory 1-C ′. The same operation is performed in the other nodes 1-F to 1-N.

  Next, FIG. 2 is a conceptual diagram showing an outline of a relationship between nodes connected to a network in the communication system according to the present embodiment. FIG. 3 is a conceptual diagram showing an outline of a memory structure inside the node in the communication system according to the present embodiment. As shown in FIG. 2A, when there are a plurality of network segments (1-a, 2-a,..., Ma; 1-b,. If there is only one, as shown in FIG. 3 (b), the shared memory area addressed to the node 1-a (for convenience, the node address or the node address also serves as the node name) is assigned to the upper memory address. 1 and the memory address lower order is a, and the configuration of the distributed shared memory is set, and the conversion method is registered in the node address conversion means 1-B. However, a mask can be set for the memory address, and each memory area is composed of a plurality of buffers (hereinafter, the parts that are considered to be distinguishable are simply referred to as a memory area), so that the same node Multiple data can be sent to the address. That is, if the memory area corresponding to the node 1-a in FIG. 2 (a) is the shared memory area (1, a) in FIG. 3 (b), as shown in FIG. This means that a plurality of configured buffers are collected.

  When the distributed shared memory 1-C is divided in a certain unit (equivalent to a hard disk sector), the maximum number of sectors is shared memory size / (maximum number of buffers in the sector × buffer size), so the number of sectors used Assuming that m is the number of buffers used in a sector, m can be increased as long as there is room in the memory to be mounted, but n can only be increased up to the maximum number of buffers in the sector.

  Thus, either n or m can be sufficiently large to the extent that the memory space is used up, but the other is restricted, and as shown in FIG. In this example, m has a margin but the memory size is limited, so that it is difficult to generate a limit compared to n), so that a plurality of addresses are represented as one address and the other is avoided. . Alternatively, as shown in FIG. 3 (d), the portion corresponding to the shortage number is discontinuous in the middle (in FIG. 3 (d), there is a discontinuity between c and d or between w and x. However, in reality, another memory area is allocated using a virtual address, and the node relationship in the network and the memory address relationship can be identified by the same arrangement.

  As described above, in this embodiment, the structure of the distributed shared memory area is made into a block according to the structure of the network partition. For example, when the memory areas indicated by a to m,... And 1 to n in FIG. 3B cannot be secured continuously, they are secured in several parts as shown in FIG. It will be. The network partition configuration shown in FIG. 2 (a) corresponds to the configuration of the distributed shared memory area shown in FIG. 3 (b), and the network partition configuration shown in FIG. 2 (b) and the distribution shown in FIG. 3 (c). The configuration of the shared memory area corresponds to each.

  For example, even if the number of nodes increases or decreases on the network, and the node 1-n is added to the network section of the node 1-a in FIG. 2B, for example, the data on the distributed shared memory 1-C in FIG. An area corresponding to the node 1-n may be allocated at a position following m.

  As shown in FIG. 2B, when the set C-2 including the nodes 1-a ′, 1-b ′, and 1-c ′ increases in the network section C-1, as described above. , 1- (n + 1), 1- (n + 2),... May be added as in the case of adding the node 1-n, but these nodes 1-a ′, 1-b ′, 1- With c ′ as a new network section, a buffer is allocated on the distributed shared memory 1-C so as to be in parallel with the nodes 1-a,..., 1-n and the nodes 2-a,. Also good. In this case, in FIG. 3C, the distributed shared memory area C-1 corresponds to the existing shared distributed memory for the node, that is, the network section C-1 in FIG. 2B, and the distributed shared memory area C- 2 corresponds to the network section C-2 in FIG. That is, a new distributed shared memory area C-2 corresponding to the network section C-2 in FIG. 2B is added to the existing distributed shared memory area (C-1 in FIG. 3C). Become. If the number of nodes decreases, the number of buffers will be decreased.

  For such memory allocation, node change rules (described later) are provided in advance. Therefore, even if a new node is added, a number of nodes is reduced, a new network segment is added or deleted, an application program that generates transmission data and specifies a destination, an OS, etc. Even if the relationship is changed, an operation such as registering or setting the node number after change / addition / deletion of each node to be changed / added / deleted is not performed.

  Here, in order to clarify the effect of the memory configuration change processing according to the present embodiment, the normal memory configuration change processing and the memory configuration change processing according to the present embodiment will be described.

  FIG. 4 is a flowchart for explaining the operation of a normal memory configuration change process. In the figure, first, when a node relationship is changed (step Sa1) in order to cause a distributed shared memory user, an application, or an OS to cause a transition (step Sa1), a transition to a communication stop state is made to change the node relationship (step Sa2). ). Next, change information of the node address (IP number or the like) is input (step Sa3), and the changed location is notified to the application, OS, etc. (step Sa4). The application, the OS, etc. set the updated memory address for each (step Sa5). Thereafter, the application, the OS, and the like are transitioned to a communicable state after the node relationship is changed (step Sa6).

  Next, FIG. 5 is a flowchart for explaining the operation of the memory configuration changing process according to the present embodiment. In the figure, in order for the distributed shared memory user, application, and OS to cause a transition, when the number of nodes increases or decreases in the network or network classification (step Sb1), first, the node address handling means 1-B determines whether the application, The OS or the like is transitioned to a communication stop state to change the node relationship (step Sb2). Further, the node address handling means 1-B determines the memory location according to the distributed shared memory configuration change rule (described later) (step Sb3), and transitions the application, OS, etc. to the communicable state after the node relationship change. (Step Sb4).

  As described above, in this embodiment, as long as the change rule of the memory structure can be referred to, the application, the OS, and the like correspond to the node address corresponding to the change of the node relationship in the distributed shared memory 1-C. By simply writing data to the designated area of the memory structure changed by the means 1-B, the transfer control means 1-D can send the data to the predetermined node after the change.

The above-mentioned change rule means associating the relationship between the memory structure and the node. For example, the following rule can be considered.
(1) Adding a node A memory address that matches the IP address (non-masked portion) of the memory block corresponding to the segment including the node is assigned.

(2) Node reduction The memory address that matches the IP address (unmasked portion) of the memory block corresponding to the segment including the node is released from the allocation as the distributed shared memory 1-C.

  Depending on how to determine the change rule, the segment and the memory block are associated with each other, the node address (IP address or the like) is matched with a part of the memory address, or the association can be performed by a simple calculation. For example, in the case of a change rule that divides a memory block for each segment and matches the non-masked portion of the IP address with the memory address, if the information on the correspondence between the segment and the memory block changes, the node address It is necessary to register in the conversion means 1-B.

  However, when a new node is added, that is, when a memory address corresponding to a new node is to be allocated, it is not always necessary to register the correspondence relationship of the allocated memory address. That is, if the application can even refer to the change rule, data can be written (transmitted) by designating a memory address corresponding to the target node. The state that can be referred to means that a change rule to be used is stored as software, a table, a program, or embedded logic stored in a memory, and an application, an OS, or the like transmits data, or If reading is performed before that, or a target node name, a process name, or the like is specified, a memory address is output, and the memory address can be acquired.

  When the change rule is used, even if a shortage occurs in the area of the distributed shared memory 1-C, each memory area (for example, the memory portion indicated by (1, a) in FIG. Even if there is a shortage in the area addressed to the node 1-a in FIG. 2A, it is not always necessary to determine the shortage for each individual memory. In other words, a shortage is determined in a unit composed of a plurality of memory areas, and there is a shortage of memory areas in the unit of the plurality of memory areas as long as a certain shortage criterion is not exceeded. It is possible to apply a method of adding memory from a place where there is a surplus only by correcting the change rule.

  Also, even when the number of nodes decreases, it is not always necessary to directly identify the correspondence of the allocated memory address and delete it from the allocation. This is because the processing load such as search increases. When the number of nodes decreases, it is a process in the opposite direction to the case of adding a node as described above, and a part of a node address (such as an IP address) and a memory address are matched with each other or are associated by a simple calculation. As a result, the output memory address may be deleted from the correspondence with the node.

  The device described as a node is not limited as long as it has a communication function such as a computer, a server, a router, a sensor (node), a driving device, a control device, or the like. Further, the memory arrangement may not be physical, and virtual memory may be used. The memory configuration described here may be performed in advance at the stage of initializing the communication function before connection to the network, or the memory may be reconfigured in a situation where communication is already performed. In addition, the relationship between nodes is expressed as a bus structure for convenience, but the actual physical connection state such as a ring shape or a star shape may be in any form.

  Even if the number of nodes is increased or decreased, or the memory configuration is changed, the node address conversion unit 1-B holds a table indicating the correspondence between the node addresses and the memory addresses, and the node address and the memory address in the table. The correspondence of is to be changed. Alternatively, the node address conversion means 1-B is a node address obtained by processing a memory address based on a predetermined rule such as a shift of a memory address, a partial mask, or an operation. The memory address and the node address are converted by the above means. Therefore, even if the area of the distributed shared memory 1-C is insufficient, the node address conversion unit 1-B is stored in the table of the node address conversion unit 1-B without rewriting or correcting it on a large scale. It is possible to solve the shortage by simply adding a new correspondence between the memory address and the node, or by correcting the operation of the node address conversion means 1-B so as to be a new correspondence between the memory address and the node. it can.

  Next, a memory configuration setting method will be described by taking as an example a case where the network structure is only the network section C-1 and the network section C-2 as shown in FIG.

  FIG. 6 is a conceptual diagram for explaining an example of a memory configuration setting method according to the present embodiment. In FIG. 6 (a), a part with 1, 2, q (1) originally indicates a total of q (1) area sets of 1, 2, 3,..., Q (1). However, it is omitted for simplification of description. The same applies to other regions. In addition, a region next to 1, 2, q (5) indicated as a reference G5 indicating a group of regions is indicated as 1, 2, q (s) indicated as a reference Gs indicating a region group. This is also a group of regions denoted G1 indicating a group of regions, 1, 2, q (1), followed by 1, 2, q of the next region group denoted G2 indicating a region group. There is q (2), and similarly, there are 1, 2 and q (5) of the group of the next region indicated by the symbol G5 on the leftmost side, and the next region indicated by the symbol Gs indicating the region group This means that there are s groups of regions of the same kind, such as 1, 2 and q (s) of the above groups, but this is omitted for the sake of simplicity.

  Further, although the memory area M1 and the memory area M2 shown in FIG. 6A are written horizontally, any of C-1, C-2,..., C-i in FIG. Or a memory area equivalent to the memory area indicated by the square areas “1, 2,..., N” and “a, b,..., D, f” shown in FIG. It corresponds to either. The size of the memory area differs between FIG. 3 and FIG. 6, but these are only examples of the memory configuration, and the size of the memory area is not considered and the configurations correspond to each other. Show.

  Therefore, there are a plurality of memory areas indicated by the square areas in FIG. 3C and FIG. 3D in the local memory, and M1 and M2 are two memory areas. However, it is not necessary to be secured in physically continuous positions. Further, due to the virtual memory, each of the memory areas indicated by the square areas in FIG. 3C and FIG. 3D is divided into a plurality of physical positions and is physically discontinuous. It doesn't matter.

  Further, in FIG. 6, addresses [G1-1, a], [G1-1, b],..., [Gs-q (s), c], etc., which are indicated as the minimum areas of the memory areas M1 and M2. May not be the minimum physical memory configuration. In other words, these are the minimum units of data transfer as seen from the upper layer (denoted as the minimum data transfer unit area). “When data is written in this unit area, the data reaches the size of this unit area. It is an area for use such as “there is no transfer to the target node”, and has a certain size.

  In the drawing, a combination of numbers and alphabets is used as a symbol indicating the position, but one character may be one address, and a plurality of addresses (or FIG. 3 (e) as shown in FIG. ), A larger number of addresses) may be indicated.

  When a shortage of a memory area currently used as the distributed shared memory 1-C is predicted, a new memory area needs to be allocated to the distributed shared memory 1-C. Therefore, in the present embodiment, if a shortage of the memory area M1 currently used is predicted, if there is a memory area that exists as a free area in the local memory and can be allocated as the distributed shared memory 1-C, By adding this as the memory area M2 to the memory area M1, the shortage of the memory area of the distributed shared memory 1-C is solved.

  In the memory area M1 already used as the distributed shared memory 1-C, it is assumed that the shaded data transfer minimum unit area is currently associated with a node. That is, it is assumed that the address [G1-1, a] in FIG. 6A corresponds to the node 1-J in FIG. 1 or corresponds to the node 1-a in FIG. . In addition, the area groups G1, G2,..., Gs are, for example, [G1-1, a], [G1-1, b],. In FIG. 2A, [G1-2, a], [G1-2, b],..., [G1-2, x] of the region group G1 correspond to the node 1-a in FIG. [G2-1, a], [G2-1, b],..., [G2-1, x] of the area group G2 correspond to the node 1-b in FIG. Suppose that it corresponds to the network classification.

  In such a state, FIG. 6A shows a situation in which the region group G1, the region group G3, the region group G4, and the region group G5 have exceeded a predetermined shortage criterion. The predetermined shortage criterion usually has an appropriate value that differs depending on the system to be applied. Therefore, an appropriate value may be selected and set depending on the system. A plurality of them may exist. For example, in FIG. 6A, in the memory area M1, there are two shortage criteria, the shortage criterion L1 and the shortage criterion L2. It is a value corresponding to the threshold value.

  The shaded portion shown in FIG. 6A is an area in which the node address and the memory address are already associated, that is, in the node address conversion means 1-B, if the memory address is input, the node address is output. An area in the distributed shared memory 1-C that is in a state to be set is shown. For example, if the associated memory area exists in the area group at a position that exceeds the section c in the area set, that is, if it exists at a position that exceeds the shortage criterion L1 on the right side, the shortage criterion is exceeded. In addition, if it exists at a position that exceeds the shortage criterion L2 on the right side, it is determined that the shortage criterion has been exceeded twice.

  By determining whether or not there is a region group that exceeds the shortage criterion L1 in the memory region M1, it is first determined that the region group G1, the region group G3, the region group G4, and the region group G5 have exceeded the shortage criterion L1. Is done. The shortage criterion L1 described here is treated as a memory region allocation criterion in the region groove. It is determined whether or not the memory area allocation standard is exceeded for each of the area group G1, area group G3, area group G4, and area group G5 that exceeds the shortage criterion L1.

  However, there may be a plurality of memory area allocation criteria (for example, in the memory area M1 in FIG. 6A, there are two memory area allocation standards L1 and L2), and there are a plurality of cases. Determines whether the memory area allocation criterion is exceeded for each memory area allocation criterion. That is, in the memory area M1 of FIG. 6A, there are two memory area allocation criteria L1 and L2, and the area group G1, area group G3, area group G4, and area group G5 are allocated to the memory area. The reference L1 is exceeded, and the area group G1 and the area group G5 exceed the memory area allocation reference L2.

  A new memory area is additionally allocated to the area group that exceeds the memory area allocation criteria. By determining how many memory area allocation standards are exceeded, the size of the memory area to be additionally allocated can be determined. A larger memory area is newly allocated to a memory group having a large shortage of memory areas. That is, how large a memory area is allocated is determined by which section (partition of an area set in an area group) is determined as a memory area allocation criterion. For this reason, as described in this example, the memory area usage of the area set that uses the largest memory area in the area group is not subject to the comparison determination with the memory area allocation criterion, but other values. May be the target of comparison determination with the memory area allocation criterion. That is, depending on the system to be applied or the usage status of the memory area, another value may be a target for comparison with the memory area allocation reference.

The memory area M2 in FIG. 6A is an example of a memory area that is newly allocated to an area group that is insufficient in the memory area M1. The size of the newly allocated memory is based on how many memory area allocation criteria are exceeded. For example, in the memory area M1 in FIG. 6A, a memory having a size corresponding to one new area group for the area group G3 and the area group G4 that exceeds the memory area allocation reference L1 by one. and allocate space, it is indicated by respective G3 ₂ and G4 _2. In addition, for the area group G1 and the area group G5 that exceed the memory area allocation reference L2, the memory areas having the sizes corresponding to the two new area groups are allocated, and G1 ₂ and G5, respectively. ₂ . Note that no new memory area is allocated to the area group G2 and the area group Gs that do not exceed the memory area allocation criteria L1 and L2.

  In this embodiment, a memory area having a size corresponding to the area group is assigned as a unit. However, this is merely an example, and depending on the system to be applied or the usage status of the memory area, another size may be allocated as a unit, and the area that does not exceed the memory area allocation standards L1 and L2 Also for the group, a new allocation having a smaller size than the memory size allocated to the area group exceeding the memory area allocation criterion may be performed.

  In FIG. 6A, an arrow line (dotted line) connecting the memory area M1 and the memory area M2 indicates where the boundary of the area set in the memory area M1 corresponds in the memory area M2. An example of a situation in which the distributed shared memory 1-C is continuously used after a new memory area is allocated is shown as a memory area M1 'and a memory area M2' in FIG. 6B. When the distributed shared memory 1-C is used at the same usage rate increase rate as the memory area M1 of the distributed shared memory 1-C was used before the new allocation, the area group G1 and the area group G5 In FIG. 2, the memory area M1 ′ is all used, and in addition, the memory area M2 ′ is used. In the area group G1, the area group G3, and the area group G4, the memory area M1 ′ is not yet used. However, the memory area M2 ′ is also used. Although the memory area M1 'and the memory area M2' are described adjacent to each other, the physical memory positions need not be adjacent to each other.

  Next, FIG. 7 is a conceptual diagram showing an outline of the situation of the distributed shared memory 1-C after the memory area is changed in the present embodiment. In FIG. 7A, FIG. 7A shows an example of the state of the memory after the memory configuration is changed due to a new node being added or the distributed shared memory 1-C allocated to the node being insufficient. Show. In FIG. 7B, since a part of the distributed shared memory 1-C assigned to the node becomes unnecessary or the number of nodes decreases, the assignment of the distributed shared memory 1-C of the target part is released. It shows the situation for normal local memory usage.

  First, in FIG. 7A, when the region [G1-1, a] belonging to the region group G1 is added due to a region shortage, the region [G1-1, a] Is not allocated as a continuous area to the area of the area group G1, but is allocated for the area [G1-1, a] at discontinuous positions ([G1-1, a] (addition) or G1 (addition)) Where). In addition, the allocation is originally based on the premise that the allocation unit is increased in the memory width direction when a shortage occurs after the initial allocation as in the area [G1-1, b]. That is, it is assumed that two allocation units can be allocated as in the area [G1-1, e].

  For example, for efficiency, allocation is performed in units smaller than the allocation unit size by the method described with reference to FIGS. 6A and 6B. That is, FIG. 6A corresponds to a situation in which the allocation unit area to be added in the memory width direction is placed in the vertical direction. In FIG. 6A, the memory area M1 and the memory area M2 are arranged horizontally. 6 corresponds to FIG. 6B. However, as shown in FIG. 6B, the allocation unit is divided on the premise of the original allocation method (the same applies to the memory areas M1 and M2 in FIG. 6A).

  In FIG. 7 (a), the area [G1-1, a] is already assigned to [G1-1, b] before being assigned as [G1-1, a] (addition). * 1) or a state that was originally assigned to the area (* 2) for [G1-1, c], but was assigned as [G1-1, a] (additional) because it was insufficient in itself It is shown. Initially, the position is different from the range used as the distributed shared memory 1-C.

  FIG. 7B shows a state in which a part for [G1-1, a] is no longer necessary, and is removed in the reverse order of the procedure of addition. First, [G1-1, a] (addition) is deallocated as the distributed shared memory 1-C. Further, when an unnecessary amount increases for [G1-1, a], an area (* 2) originally intended for [G1-1, c] and an area originally intended for [G1-1, b] (* 1) 1 is deassigned for [G1-1, a]. However, since this is an area in the area group G1, the allocation as the distributed shared memory area is not released, and when the necessary amount of [G1-1, b] and [G1-1, c] increases, It is secured as an area allocated to. In FIG. 7B, [G1-1, c] disappears when the node corresponding to [G1-1, c] is deleted from the network, and the area (* 3) is Allocation as [G1-1, c] is released, but since it is in the area group G1, allocation as a distributed shared memory area is not released, and is allocated when a new node is added in the area group G1. Maintain the state reserved for.

  Next, the operation of the main part of the memory configuration setting process according to the present embodiment will be described in comparison with the prior art.

  FIG. 8 is a flowchart for explaining the operation of the main part of the memory configuration setting process according to the prior art. In FIG. 8, in the conventional technique, since the necessary distributed shared memory 1-C is increased, a part of the local memory is allocated as the distributed shared memory 1-C, and as a result, the distributed shared memory 1-C is added. The outline of the operation | movement flow of the main part of the setting method of a memory structure is shown. In the figure, what are indicated as areas M1 and M2 correspond to the areas M1 and M2 shown in FIG. 6A, and are newly used as the areas already used as the distributed shared memory 1-C. An area allocated as the distributed shared memory 1-C is shown. The area group length unit is a minimum area surrounded by a thick line in FIG. 6A, and is a unit of a memory area when a new memory area is allocated in the memory allocation process.

  First, the node address handling means determines whether or not there is an area set that exceeds the memory area allocation criterion in the area M1 of the distributed shared memory 1-C (step Sc1). If there is, it is determined whether or not the size of the area set can be secured as the distributed shared memory 1-C in the unused area (step Sc2). If it cannot be secured, an error is notified (step Sc3). On the other hand, if the unused area can be secured as the distributed shared memory 1-C, the node address handling means stores the address of the area in which the size of the area set can be secured in the table referred to by the node address conversion means. It is added after the address area M1 (step Sc4). Thereafter, the process waits for a predetermined time (step Sc5), and then returns to step Sc1. Here, the description is based on the assumption that a table is used. However, when node address conversion is performed by calculation, the calculation means is corrected.

  As described above, in the prior art, it is determined whether there is any memory area set in the distributed shared memory 1-C that is currently used that exceeds the memory allocation criterion, and the memory that exceeds the memory allocation standard. If there is an area set, a new memory area is secured, and in order to associate the reserved memory area with the same node address as the node address to which the excess memory area set corresponds, the node address handling means The table or calculation means referred to is corrected.

  As a result, the reserved memory area can be used as a successor of the excess memory area set, that is, the excess memory area set is changed to “exceeded memory area set and reserved memory area”. Will be replaced. The size of the secured memory area is usually the same as the size of the memory area set of the distributed shared memory 1-C currently used. This is because all the memory area sets are made the same size, This is because by making the added area the same size, the allocation from the local memory to the distributed shared memory 1-C becomes easy. Whether or not there is an excess memory area set is determined periodically or irregularly in response to an instruction from the user of the inter-node data transfer control device or the usage system. In the prior art, since a set of areas and groups as shown in FIG. 6A are not configured, an efficient memory allocation according to the usage situation of the distributed shared memory 1-C and the relationship between nodes is possible. Have difficulty.

  Next, FIGS. 9 and 10 are flowcharts for explaining the operation of the main part of the memory configuration setting process according to this embodiment. Similarly to FIG. 8, the flowcharts shown in FIGS. 9 and 10 also allocate a part of the local memory as the distributed shared memory 1-C because the necessary distributed shared memory 1-C has increased. The outline of the operation | movement flow of the main part of the setting method of a memory structure when C is added is shown.

  First, the node address converting means 1-B is a memory group M1 of the currently used distributed shared memory 1-C, in which a plurality of memory area sets are in excess of the shortage criterion. It is determined whether there is a shortage criterion (step Sd1), and if there are those that exceed the shortage criterion, the memory area set that uses the most memory area in the target area group is assigned the memory area allocation. It is determined whether the reference is exceeded (step Sd2). Normally, the shortage criterion and the memory area allocation criterion are the same value, but they may be set to different values, thereby ensuring a margin, changing the timing of adding the memory area, and saving the memory to other area groups. It is also possible to add a region and time.

  If there is a memory area set that exceeds the memory area allocation criteria, the number of new area group allocations to the area group including the memory area set is incremented by 1 (step Sd3). Next, the node address conversion unit 1-B uses the value of the memory usage of the memory area set that uses the most memory area in the target area group as the maximum memory usage in the target area group. The area group length unit is subtracted from the maximum memory usage in the target area group (step Sd4), the group excess number is added by 1 (step Sd5), and the process returns to step Sd2. Thereafter, steps Sd2 to Sd5 are repeated until the memory area set that uses the most memory area in the target area group does not exceed the memory area allocation criterion.

  When the memory area set that uses the most memory area in the target area group does not exceed the memory area allocation standard, the node address conversion unit 1-B determines that the area set ([Gx−x, x ] Is added by 1 (step Sd6), and steps Sd2 to Sd5 are repeated until there are no region groups that exceed the shortage criterion in the region threshold M1 (step Sd7). When there are no region groups that exceed the shortage criterion in the region threshold M1, the product of the number of region group allocations and the region group length unit is taken and the products are summed for each region group (step Sd8).

  Next, the node address conversion unit 1-B determines whether or not the total size can be secured as the distributed shared memory 1-C in the unused area (step Sd9), and the total size cannot be secured. The (free size ÷ total size) × area group length is set to a newly allocated area group length to reduce the total size value to a size that can be secured (step Sd10), and the total size is an unused area. To determine whether it can be secured as the distributed shared memory 1-C (step Sd11). If the total size does not reach the size that can be allocated as the distributed shared memory 1-C, an error is notified to the user of the inter-node data transfer control device and the use system (step Sd12).

  On the other hand, when the total size can be secured in the local memory that can be allocated as the distributed shared memory 1-C in step Sd9 or step Sd11, an area corresponding to the total size is selected from the securable areas. Region M2 is set (step Sd13). Next, the memory address of the area M2 is added to the succeeding area M1 of the address in the table referred to by the node address converting means 1-B (step Sd14). Then, it waits for a predetermined time (step Sd15), and then returns to step Sd1.

  In addition, when performing node address conversion by calculation, a calculation means is corrected. The initial value of the number of area group assignments is 0. Whether or not there is an area group that exceeds the shortage criterion may be determined periodically or irregularly in response to an instruction from the user of the inter-node data transfer control device or the usage system.

Next, the deallocation process according to this embodiment will be described.
9 and 10, when there is a region group that exceeds the shortage criterion in the region M1, it is determined that the group is exceeded for each region set ([Gx-x, x], etc. in FIG. 6A). Record with the address of the allocation unit (denoted as the number of excess groups). After allocating the distributed shared memory 1-C to the group, if the number of the distributed shared memory 1-C may be small for the group, the allocation will be canceled. Refers to the record of the number of excess groups, and if there is any value in the record, until the value becomes 0, addresses in the range not initially allocated as the distributed shared memory 1-C (see FIG. 7A) are targeted. Then, the table (table indicating the correspondence between the node address and the memory address) held by the node address converting means 1-B is searched to identify the additional area of the corresponding node, and the distributed shared memory 1- Continue to deallocate as C.

  If the number of distributed shared memories 1-C may be small for the group and the record of the number of excess groups is 0, the area set ([Gx-x, x], etc. in FIG. 6A) is set. With reference to the newly added allocation unit number, until the value of the allocation unit number becomes 0, the table held by the node address conversion means 1-B is searched for the address in the group, and the corresponding node The additional area is identified, and the allocation of the area as the distributed shared memory 1-C is continued. As the order of releasing the allocation, “area added due to excess of group” is released first, and “area added in group” is released next. Therefore, normally, the process of deleting the memory addresses added to the subsequent in the table in order from the back may be performed. In addition, the use of the area | region after canceling | release is as having demonstrated in Fig.7 (a).

Next, the operation of deallocation processing according to this embodiment will be described.
FIG. 11 is a flowchart for explaining the operation of deallocation processing according to this embodiment. The node address converting means 1-B first determines whether or not the area set needs to be reduced (step Se1). If the area set needs to be reduced, the number of excess groups (of the corresponding area set) Is determined to be 0 (step Se2). If the number of group excesses (of the corresponding area set) is greater than 0, an address corresponding to one of the allocation units of the area set existing at the address corresponding to the group excess is reduced (step Se3). This corresponds to reducing the allocation unit by 1 in the [G1-1, a] (addition) portion in FIG. 6A, and reducing the number of excess groups (of the corresponding area set) by 1 (step Se4), returning to step Se1, and repeating the above-described processing.

  On the other hand, if the group excess number is 0, it is determined whether or not the number of allocation units newly added to the area set is 0 (step Se5). If the number of allocation units newly added to the area set is larger than 0, an address corresponding to one of the allocation units newly added to the area set is reduced (step Se6). This corresponds to the reduction of the area (* 1) and (* 2) in FIG. 7A, and the number of allocation units newly added to the area set = assignment newly added to the area set. The number of units is reduced by 1 (step Se7), and the node address conversion means 1-B is changed as the area of the area set assigned at the time of initial allocation of the reduced area (step Se8). At this time, since the state where the memory area allocation criterion is exceeded is maintained, the changed area is not used as the area set allocated at the time of initial allocation, but the area of another area set in the area group. It becomes available when there is not enough. Then, it returns to step Se1 and repeats the process mentioned above.

  In the above step Se5, “whether the number of allocation units newly added to the region set ([Gx−x, x], etc. in FIG. 6A) ≠ 0” is “the initial value of the region set in the region group” It is determined whether there is an area different from the target assignment. For example, if an area group corresponds to a segment of the network, a certain node in the segment is deleted, so that local memory is placed at a location corresponding to the area set in the area group, and there are multiple area groups. Can be prevented from being separated.

  That is, in the initial allocation state, when the correspondence between the transmission destination / source node and the memory address is made, the correspondence can be made in segment units, and the search can be performed in segment units. Such inefficiencies can be avoided because the operation becomes impossible. Therefore, if the number of allocation units newly added to the area set is 0, an address corresponding to one of the allocation units assigned to the area set at the time of initial allocation is reduced (step Se9). . This is to reduce the area (* 4) and (* 5) in FIG. 7A, and to reduce the area (* 5) also reduces the nodes for the area set. It corresponds to that. An address corresponding to one of the allocation units is reduced for the area set ([Gx-x, x], etc. in FIG. 6A), but the node address conversion means 1-B is not changed (memory) If it does not exceed the area allocation criteria, it can be used when the area of another area set in the area group is insufficient). Thereafter, the standby state is set (step Se10).

  In the present invention, the description has been made assuming that the memory configuration is physically assumed, but by using virtual memory, the memory address to be used may be different from the real memory address. In this case, the memory configuration according to the present embodiment is the description on the virtual memory, and the processing is performed based on the physical memory address through the conversion between the virtual address and the real address. Further, even if the virtual memory addresses are continuous for a certain memory area, the physical memory arrangement may be discontinuous. For example, the area M1 in FIG. However, this is on virtual memory, and may be discretely allocated on real memory.

  The node address conversion may use a table that associates the node address with the memory address of the distributed shared memory 1-C. Alternatively, for example, when the lower address of the memory address is masked, the node address may be set to be a node address, and the node address and the memory address of the distributed shared memory 1-C may be associated with each other by calculation or memory operation. is there.

  Here, a simple example will be described in the case where the node address converting unit 1-B converts the node address by calculation. FIG. 12 is a conceptual diagram for explaining a case where the node address conversion unit 1-B converts a node address by calculation in the present embodiment. If the node address, memory address, etc. take complex values, or there are a large number of nodes corresponding to the same area group, the node numbers, memory addresses, etc. are (a) aligned before calculation ( b) Give an offset. (c) By making a value such as 10 times or 100 times in the following example an appropriate large value, the same processing can be executed.

  When the number of nodes corresponding to the same area group is 10 or less, as shown in FIGS. 12A and 12B, the value obtained by multiplying the node address by “10 × area set size” is set as the start address of the area set. A group is set for each “100 × area set size multiple”. The area set that does not need to be added is the “area set in units of 10 times”. However, for the area set in which the memory is added, “10 times” processing is increased to 11 times, for example, “11 times unit. As a region set ”. For the area set to be removed, the process of “10 times” is listed as “area set in units of 9 times” by multiplying, for example, 9 times. For the “area set in units of 10”, the memory address can be obtained by multiplying the node address by “10 × area set size” (the reverse process is performed to obtain the node address from the memory address).

  For the area set listed in “area set with 11 (9) times unit”, the node address is obtained by “11 (9) × area set size multiple”. As shown in FIG. 12C, the increase / decrease in the case of associating the node address with the memory address of the distributed shared memory 1-C by an operation or a memory operation or the like is as follows. This is done by adding or deleting an area set address or node address to the "area set" list. The case where the area group itself is subject to increase / decrease is the same as the case where the area set is increased / decreased. Note that a process corresponding to “multiply by 10” or the like may be performed by a shift or a mask, or an operation may be used.

  In FIG. 12, for example, “node address 6 has 11 area sets, and the start address of the area set is a position reduced by 10 addresses compared to the case of normal initial allocation (10 nodes each). " By registering the node address in the “α multiple unit” of this list, the memory expansion / reduction using the calculation, that is, “determine the position of the area set corresponding to the node address by the node address × 10 × area set” is performed. enable. By registering the node registered as “11 times” as “10 times”, the memory allocation in the initial state is restored. If it is changed from “9 times” to “11 times”, it will be added (reversely, it will be removed).

  According to the above-described embodiment, in the case where there is an area used in the distributed shared memory 1-C or a buffer that exceeds a certain memory size, and an area shortage may occur. In this case, it is possible to efficiently determine that the size has been exceeded and perform the allocation efficiently. At this time, even if the allocation becomes unnecessary later, the allocation can be canceled while maintaining the efficiency at the time of allocation.

  In addition, the distributed shared memory 1-C can be used efficiently, and local memory can be allocated efficiently. In this case, even when allocation becomes unnecessary later, deallocation is performed while maintaining the efficiency at the time of allocation. It can be carried out.

  Further, according to the present embodiment, when using the node change rule, it is not always necessary to add a memory to the shortage of individual areas, and by correcting the change rule, for a certain unit area Thus, it is possible to efficiently allocate such as determining and adding a shortage, and the same effect can be obtained by modifying the change rule when canceling the allocation.

  In addition, since it is possible to easily associate with a network divided into a plurality of areas, it is necessary to perform an oversize determination / removal necessity determination for each set of memory areas for both allocation and deallocation. In addition, it is possible to reduce memory shortage and waste as a result of new memory allocation / memory deallocation, and memory allocation / deallocation in accordance with the node relationship becomes possible.

  In the above-described embodiment, a program for realizing each step by the node address converting unit 1-B described above is recorded on a computer-readable recording medium, and the node address converting unit 1-B or the like is recorded. A program for realizing the function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read by a computer system and executed. Here, the “computer system” may include an OS and hardware such as peripheral devices. Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. The “computer-readable recording medium” means a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a CD-ROM, a hard disk built in a computer system, etc. This is a storage device.

  Further, the “computer-readable recording medium” means a volatile memory (for example, DRAM (Dynamic DRAM) in a computer system that becomes a server or a client when a program is transmitted through a network such as the Internet or a communication line such as a telephone line. Random Access Memory)), etc., which hold programs for a certain period of time. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

It is a block diagram which shows the structure inside the node which communicates with the communication system by embodiment of this invention. It is a communication system by this embodiment, Comprising: It is a conceptual diagram which shows the outline of the relationship of the node connected to the network. It is a communication system by this embodiment, Comprising: It is a conceptual diagram which shows the outline of the structure of the memory inside a node. It is a flowchart for demonstrating operation | movement of a normal memory structure change process. It is a flowchart for demonstrating operation | movement of the memory structure change process by this embodiment. It is a conceptual diagram for demonstrating an example of the setting method of a memory structure by this embodiment. In this embodiment, it is a conceptual diagram which shows the outline of the condition of the distributed shared memory 1-C after a memory area is changed. It is a flowchart for demonstrating operation | movement of the main part of the setting method of a memory structure by a prior art. It is a flowchart for demonstrating operation | movement of the main part of the setting method of a memory structure by this embodiment. It is a flowchart for demonstrating operation | movement of the main part of the setting method of a memory structure by this embodiment. It is a flowchart for demonstrating the procedure of deallocation by this embodiment. In this embodiment, it is a conceptual diagram for demonstrating the case where the node address conversion means 1-B converts a node address by calculation.

Explanation of symbols

1-A application program or OS 1-B node address conversion means 1-C, 1-C ′ distributed shared memory 1-C-1, 1-Cn memory address 1-D, 1-D ′ transfer control means 1-E internal bus 1-F, 1-J,..., 1-N node 1-G data 1-H node address 1-I network

Claims (9)

A node-to-node data transfer control device that controls data transfer with other processes using a distributed shared memory allocated on a local memory of each node among a plurality of nodes connected to a network,
A distributed shared memory in which memory addresses are virtually or physically the same among the plurality of nodes;
An area allocation means for allocating an area corresponding to the node to the distributed shared memory in response to the addition of the node;
An area releasing means for releasing an area allocated to the distributed shared memory in accordance with the reduction of the node,
The area releasing means is
An inter-node data transfer control device, wherein when an area is released, an area assigned later by the area assigning unit is released first. A determination unit for determining whether an area allocated to the distributed shared memory exceeding a predetermined threshold is used in units of groups based on a connection relationship of the plurality of nodes;
The area allocating means includes
The area according to claim 1, wherein when the determination unit determines that an area is used exceeding a predetermined threshold, an area is newly allocated to the distributed shared memory in units of groups. Internode data transfer control device. The predetermined threshold is at least one or more,
3. The inter-node data transfer control according to claim 2, wherein the area allocating unit determines a capacity of an area to be allocated to the distributed shared memory in units of groups based on a predetermined number of thresholds that have been exceeded. apparatus. The area releasing means is
The inter-node data transfer control device according to claim 2 or 3, wherein an area allocated in group units later is released first. A node-to-node data transfer control method for controlling data transfer with another process using a distributed shared memory allocated on a local memory of each node between a plurality of nodes connected to a network,
Allocating an area corresponding to the node to a distributed shared memory whose memory address is virtually or physically the same among the plurality of nodes in response to the addition of the node;
Releasing the area allocated later when releasing the area allocated to the distributed shared memory according to the reduction of the node;
A data transfer control method between nodes. A step of determining whether or not an area allocated to the distributed shared memory exceeding a predetermined threshold is used in units of groups based on a connection relationship of the plurality of nodes;
6. The inter-node data transfer control according to claim 5, wherein when it is determined that an area is used exceeding the predetermined threshold, a new area is allocated to the distributed shared memory in units of groups. Method. The predetermined threshold is at least one or more,
7. The inter-node data transfer control method according to claim 6, further comprising a step of determining a capacity of an area to be allocated to the distributed shared memory in units of groups based on the number of the predetermined threshold values that have been exceeded. .   8. The inter-node data transfer control method according to claim 6 or 7, wherein when the allocated area is released, an area allocated later in units of groups is released first. A computer for controlling an inter-node data transfer control device that controls data transfer with other processes using a distributed shared memory allocated on a local memory of each node among a plurality of nodes connected to a network In addition,
Allocating an area corresponding to the node to a distributed shared memory whose memory address is virtually or physically the same among the plurality of nodes in response to the addition of the node;
Releasing the area allocated later when releasing the area allocated to the distributed shared memory according to the reduction of the node;
A program characterized by having executed.

Images