JP2013149069A - Load distribution method, distribution processing system, distribution processing device, and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique capable of efficiently executing load distribution in a distribution processing system.SOLUTION: In a distribution processing system including one or more distribution processing devices (nodes), a substitute node aggregate is generated for causing one or more nodes (substitute nodes) to perform all processes taken charge by a node (candidate node) with which some of processes taken charge by a node (comparison node) are shared (step S1). A prediction value (after-substitution load prediction value) of a load after substitution by each substitute node when the substitute node aggregate performs processes in place of the candidate node is obtained (step S2). When the after-substitution load prediction value for each substitute node satisfies predetermined conditions (step S3: Yes), load distribution is executed in which the substitute node aggregate is caused to perform the processes for the candidate node to make the candidate node into a free node, and the candidate node, which has become a free node, is caused to perform some of the processes taken charge by the comparison node in place of the comparison node (step S4).

Description

  The present invention relates to a load distribution method, a distributed processing system, a distributed processing device, and a computer program in a distributed processing system including one or more distributed processing devices.

  There is known a distributed processing system in which a plurality of distributed processing devices (nodes) share processing and storage of a large amount of data. In such a distributed processing system, there is a case where the processing amount of a certain first node is extremely increased and a load such as a CPU (Central Processing Unit) usage rate approaches 100% is high. In this case, such a distributed processing system can distribute the load of the first node by sharing a part of the processing being executed by the first node with other second nodes. . Thus, in a distributed processing system, load distribution among nodes becomes an important issue. An example of a technique for performing such load balancing is described in Non-Patent Document 1. Non-Patent Document 1 mainly describes load distribution in a distributed processing system employing Chord. Chord is a typical algorithm of a distributed hash table (Distributed Hash Table, DHT) known as a technique for assigning data to a node in charge in a distributed processing system. In Chord, a ring-shaped key space is defined. In this key space, the value of the key increases clockwise. The key is a hash function value. The participating node has any key on the key space as a node ID. In the key space, a length in the clockwise direction is defined as a distance from one key to another key. Here, the node closest to a key is called the successor of that key. The node farthest from a key is called the predecessor for that key. In Chord, certain data is handled by a key successor calculated from the data. That is, each node takes a key that is larger than the node ID of the predecessor of the own node and reaches the node ID of the own node. A technique described in Non-Patent Document 1 will be described below for load distribution in a distributed processing system employing such Chord.

Each node (node a) acquires the load value of another node (node x) selected at random, and the relationship between the load value (l_x) of node x and the load value (l_a) of node a is When l_x ≦ ε × l_a (ε is a constant of 0 <ε <1, for example, 0.25), any one of the following cases 1 to 3 is performed. Note that the successor of node x is node y.
<Case 1: When node a is a successor of node x>
A part of the data held by the node a is moved to the node x so that the loads on the node a and the node x are equalized.
<Case 2: When the node a is not the successor of the node x and the load value of the node a is equal to or less than the load value of the node y>
Part of the data held by the node y is moved to the node x so that the load values of the node x and the node y are equalized.
<Case 3: When the node a is not the successor of the node x and the load value of the node a is larger than the load value of the node y>
All data held by the node x is moved to the node y. That is, the node y takes charge of all data processed by the node x. The node a moves a part of the data held by the node a to the node x so that the load values of the node a and the node x are equal.

  Cases 1 to 3 will be described using specific examples. As shown in FIG. 12, a distributed processing system composed of eight nodes from node a to node h is assumed. In FIG. 12, each node has node IDs arranged in the order shown in the key space 900 on the ring. Further, the numerical value shown beside the graphic indicating each node represents the load of each node.

<Case 1>
It is assumed that the node a has acquired the load value (l_h) of the node h. Here, l_h (10%) ≦ ε × l_a (80%), and node a is a successor of node h. Therefore, this case corresponds to case1. Therefore, in order to reduce the load value of the node a, the node a moves a part of the held data to the node h so that the load values of the node a and the node h are equal. As a result, the load values of the node a and the node h are 45%, respectively. Further, it is assumed that the key ranges handled by the node a and the node h are [151, 190] and [100, 150] before the data movement. [K1, k2] represents a key range from k1 to k2. Here, if the data whose key is [151, 170] is moved from node a to node h, the key ranges newly assigned to node a and node h are [171, 190] and [100, 170]. become.

<Case 2>
It is assumed that the node a has acquired the load value (l_b) of the node b. Here, l_b (15%) ≦ ε × l_a (80%). The node a is not the successor of the node b, and the load value of the node a (l_a = 80%) is equal to or less than the load value of the node c that is the successor of the node b (l_c = 85%). Therefore, this case corresponds to case2. Therefore, in order to distribute the load of the node c having a high load value, the node c moves a part of the held data to the node b so that the load values of the node b and the node c are equalized. As a result, the load value of node a is 80% (no change), the load value of node b is 50%, and the load value of node c is 50%. Further, it is assumed that the key ranges handled by the node b and the node c before the data movement are [191, 250] and [251, 300], respectively. Here, if the data whose key is [251, 280] is moved from the node c to the node b, the key ranges newly assigned to the node b and the node c are [191, 280] and [281, 300]. become. In addition, the range of keys that node a is in charge of does not change.

<Case 3>
It is assumed that the node a has acquired the load value (l_d) of the node d. Here, l_d (20%) ≦ ε × l_a (80%). Further, the node a is not the successor of the node d, and the load value of the node a (l_a = 80%) is larger than the load value of the node e which is the successor of the node d (l_e = 30%). Therefore, this case corresponds to case3. Therefore, in order to distribute the load of the node a having a high load value, the node d moves the held data to the node e and becomes an empty node. Then, the node a moves the data to be held to the node d so that the load value of the node a is halved. As a result, the load value of the node a is 40%, the load value of the node d is 40%, and the load value of the node e is 50%. Further, it is assumed that the key ranges assigned to the node a, the node d, and the node e before the data movement are [151, 190], [301, 380], and [381, 500]. Here, if the data whose key is [170, 190] is moved from node a to node d, the key ranges newly assigned to node a, node d, and node e are [151, 169], [170, 190] and [301, 500].

  As described above, the technique described in Non-Patent Document 1 can reduce the load of a node having a high load value in the distributed processing system.

David R. Karger, Matthias Ruhl, “Simple Efficient Load Balancing Algorithms for Peer-to-Peer Systems”, Proceedings of the sixteenth annual ACM symposium on Parallelism in algorithms and architectures (SPAA’04), 2004

  In a distributed processing system, in order to distribute the processing of the first node having a high load to other second nodes, the second node is not executing any processing (hereinafter, a node in such a state). Is also referred to as an empty node), or it is assumed that only a small amount of processing is executed and the load is low. Therefore, in the technique described in Non-Patent Document 1, in case 3, in order to distribute the load of the first node, the second node is selected at random, and all the processes currently being executed by the second node are selected. Is assigned to the third node which is the successor of the second node. As a result, the technique described in Non-Patent Document 1 uses the second node as an empty node.

  However, since the technique described in Non-Patent Document 1 substitutes the processing of the second node without considering how the load of the third node changes, the load of the third node increases. It may be too much. For this reason, it may be necessary to further distribute the load of the third node that has become too heavy. As a result, the technique described in Non-Patent Document 1 may increase the number of times of load distribution and cannot perform load distribution efficiently.

  SUMMARY An advantage of some aspects of the invention is to provide a technique capable of more efficiently executing load distribution in a distributed processing system.

  The load distribution method according to the present invention is a node (candidate node) that is a candidate for sharing a part of processing that a certain node (comparison node) is responsible for in a distributed processing system constituted by one or more distributed processing devices (nodes). ), A proxy node set is created that substitutes all of the processes handled by the candidate node by using one or more nodes (proxy nodes), and the proxy node set acts as a proxy for the candidate node A substitute load prediction value (post-substitution load prediction value) of each of the proxy nodes of the proxy node, and when the post-substitution load prediction value of the proxy node satisfies a predetermined condition, the processing of the candidate node is performed In addition to making the candidate node an empty node on behalf of a node set, a part of the processing handled by the comparison node is assigned to the candidate node that has become an empty node To perform load balancing.

  The distributed processing system according to the present invention is a distributed processing system including one or more distributed processing devices (nodes), and a candidate for sharing a part of processing handled by another node (comparison node). The candidate node consists of a proxy node set creation unit that creates a proxy node set that substitutes one or more nodes (proxy nodes) for all the processes that the node is in charge of, and sets the load value of each proxy node. Based on the proxy node load receiving unit to receive and the load value of each proxy node, a predicted value of the load after proxy of each proxy node when the proxy node set performs processing of the candidate node (after proxy) A post-substitution load prediction value calculation unit for obtaining a load prediction value), a post-substitution load prediction value transmission unit that transmits the post-substitution load prediction value to the comparison node, and a load component determined by the comparison node A load distribution execution unit that executes the load distribution unit, and the proxy node transmits a load value of its own node to the candidate node, and a load that executes load distribution determined by the comparison node A distributed execution unit, and the comparison node receives a post-substitution load prediction value receiving unit, and if the post-substitution load prediction value satisfies a predetermined condition, Performing a load distribution in which the candidate node is made a vacant node by substituting the process to the proxy node set, and part of the process handled by the comparison node is shared by the candidate node that has become a vacant node A load distribution determination unit for determining; and a load distribution execution unit for executing the determined load distribution.

  The distributed processing apparatus of the present invention is a distributed processing apparatus (node) constituting a distributed processing system, and the own node is a candidate for sharing a part of processing handled by another node (comparison node). A proxy node set creation unit that creates a proxy node set for substituting one or more other nodes (proxy nodes) for all the processes handled by the own node when the node is a node (candidate node); A proxy node load receiving unit that receives the value of the load of the proxy node, and the proxy node set after the proxy node when the proxy node set performs the processing of the candidate node based on the load value of the proxy node A post-substitution load prediction value calculation unit that obtains a predicted load value (post-substitution load prediction value), a post-substitution load prediction value transmission unit that transmits the post-substitution load prediction value to the comparison node, and the comparison node And a load distribution execution unit for performing load balancing to be.

  In the computer program of the present invention, the own node shares a part of processing handled by another node (comparison node) with the computer device constituting the distributed processing device (node) included in the distributed processing system. A proxy node set creation step of creating a proxy node set that substitutes one or more other nodes (proxy nodes) for all processes handled by the own node when the node is a candidate node (candidate node); A proxy node load receiving step for receiving the load value of each proxy node; and based on the load value of each proxy node, each proxy node in the case where the proxy node set performs the processing of the candidate node A post-substitution load prediction value calculation step for obtaining a post-substitution load prediction value (post-substitution load prediction value), and the post-substitution load prediction value are transmitted to the comparison node A load prediction value transmission step after row, to execute a load balancing execution step of executing a load distribution which is determined by the comparison node.

  The distributed processing apparatus of the present invention is a distributed processing apparatus (node) that constitutes a distributed processing system, and the own node is a comparison node that shares a part of the processing in charge with other candidate nodes. , A post-substitution load prediction value that receives from the candidate node a predicted value (post-substitution load prediction value) of the proxy node that constitutes the proxy node set that performs all of the processes handled by the candidate node. When the receiving unit and the post-substitution load prediction value satisfy a predetermined condition, the candidate node processing is substituted for the substitution node set to make the candidate node a vacant node, and the comparison node is in charge A load distribution deciding unit that decides to execute load distribution to share a part of the candidate nodes with the candidate nodes that became free nodes, and a load distribution execution unit that executes the decided load distribution; Equipped with a.

  The computer program of the present invention is a comparison node that allows a computer device constituting a distributed processing device (node) included in the distributed processing system to share a part of the processing that the node is responsible for with other candidate nodes. In some cases, after the proxy receiving from the candidate node the predicted load value (post-substitute load predicted value) of each proxy node constituting the proxy node set that performs all of the processes in charge of the candidate node When the load prediction value receiving step and the post-substitution load prediction value satisfy a predetermined condition, the processing of the candidate node is delegated to the proxy node set to make the candidate node an empty node, and the comparison node A load balancing decision step for deciding to execute load balancing for sharing a part of the processing in charge to the candidate nodes that have become free nodes; A load distribution execution step of executing the determined load distribution, thereby to execute.

  The present invention can provide a technique capable of more efficiently executing load distribution in a distributed processing system.

It is a block diagram which shows the structure of the distributed processing system as the 1st Embodiment of this invention. It is a hardware block diagram of the distributed processing apparatus as the first embodiment of the present invention. It is a functional block diagram of the distributed processing system as a 1st embodiment of the present invention. It is a flowchart explaining operation | movement of the distributed processing system as the 1st Embodiment of this invention. It is a block diagram which shows the structure of the distributed processing system as the 2nd Embodiment of this invention. It is a functional block diagram of the distributed processing system as the 2nd Embodiment of this invention. It is a flowchart explaining the load distribution operation | movement of the distributed processing system as the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the distributed processing system as the 3rd Embodiment of this invention. It is a functional block diagram of the distributed processing system as the 3rd Embodiment of this invention. It is a flowchart explaining the load distribution operation | movement of the candidate node in the 3rd Embodiment of this invention. It is a flowchart explaining the load distribution operation | movement of the comparison node in the 3rd Embodiment of this invention. It is a figure which shows an example of the distributed processing system which employ | adopted Chord.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(First embodiment)
FIG. 1 shows the configuration of a distributed processing system 1 as a first embodiment of the present invention. In FIG. 1, the distributed processing system 1 includes one or more distributed processing apparatuses (hereinafter also referred to as nodes) 10. These distributed processing devices 10 are communicably connected to each other via a network constituted by the Internet, a LAN (Local Area Network), a public line network, a wireless communication network, or a combination thereof. Although FIG. 1 shows six distributed processing apparatuses 10, the number of distributed processing apparatuses constituting the distributed processing system targeted by the present invention is not limited.

  Data to be processed by the distributed processing system 1 is processed by one of the distributed processing devices 10. Here, the data may be a content to be stored, or a query for requesting the stored content. When each distributed processing apparatus 10 receives data to be processed in the distributed processing system 1, if the data is handled by the own node, each distributed processing apparatus 10 executes the process at the own node. Each distributed processing apparatus 10 transmits the received data to another distributed processing apparatus 10 that is in charge of processing the received data if the received data is not handled by the own node. Each distributed processing device 10 may not be able to directly communicate with all other distributed processing devices 10. Each distributed processing device 10 is one that can be transferred from its own node so that it finally arrives at another distributed processing device 10 that is in charge of processing that data if the received data is not handled by its own node. The data may be transmitted to the other distributed processing apparatus 10. In such a distributed processing system 1, the load on each node varies depending on the amount of data received by that node.

  In addition, in the distributed processing apparatus 10 illustrated in FIG. 1, a node that is a target for reducing the load by load distribution is referred to as a comparison node 110. In the distributed processing apparatus 10, a node that is a candidate for sharing a part of the processing of the comparison node 110 is referred to as a candidate node 120. Further, in the distributed processing apparatus 10, when the candidate node 120 shares the processing of the comparison node 110, the node acting as the proxy for the candidate node 120 until then is referred to as the proxy node 130. It is assumed that each of the distributed processing devices 10 constituting the distributed processing system 1 can operate as the comparison node 110, the candidate node 120, and the proxy node 130.

  Next, the hardware configuration of the distributed processing apparatus 10 is shown in FIG. In FIG. 2, the distributed processing apparatus 10 includes a CPU (Central Processing Unit) 1001, a RAM (Random Access Memory) 1002, a ROM (Read Only Memory) 1003, a storage device 1004 such as a hard disk, and a network interface 1005. It is comprised by the computer apparatus provided.

  The ROM 1003 and the storage device 1004 store a computer program and various data for causing the computer device to function as the distributed processing device 10 of the present embodiment.

  The CPU 1001 reads the computer program and various data stored in the ROM 1003 and the storage device 1004 into the RAM 1002 and executes them.

  The network interface 1005 is a module that communicates with other distributed processing apparatuses 10 via a network.

  Next, the functional block configuration of the distributed processing system 1 is shown in FIG. In FIG. 3, the distributed processing system 1 includes a proxy node set creation unit 101, a post-substitution load prediction value calculation unit 102, a post-substitution load prediction value determination unit 103, and a load distribution execution unit 104. Here, each functional block includes a network interface 1005 in one of the distributed processing devices 10 and a CPU 1001 that reads a computer program and various data stored in the ROM 1003 and the storage device 1004 into the RAM 1002 and executes them. The Further, each functional block of the distributed processing system 1 may be configured by the same distributed processing device 10, or at least one may be configured by different distributed processing devices 10. In addition, the distributed processing device 10 constituting each functional block needs to communicate with another functional block in its operation, and the other functional block is constituted by another distributed processing device 10. Shall communicate with the other distributed processing apparatus 10 as appropriate.

  The substitute node set creation unit 101 creates a substitute node set for a candidate node 120 that substitutes all of the processes handled by the candidate node 120 using one or more substitute nodes 130. Here, the proxy node set creation unit 101 may create a proxy node set by randomly selecting one or more nodes that can be the proxy node 130 as the proxy node 130. Note that the node that can be the proxy node 130 may be, for example, a node other than the comparison node 110 and the candidate node 120 in the distributed processing device 10 configuring the distributed processing system 1.

  The post-substitution load prediction value calculation unit 102 obtains a post-substitution load prediction value (post-substitution load prediction value) of each proxy node 130 when the proxy node set performs all of the processing of the candidate node 120.

  Here, if the proxy node set is composed of a plurality of proxy nodes 130, the post-substitution load prediction value calculation unit 102 uses all of the processes of the candidate nodes 120 as post-substitution load prediction values of the proxy nodes 130. The total value of the load value allocated to each proxy node 130 and the load value of each proxy node 130 is calculated. For example, the post-substitution load predicted value calculation unit 102 calculates a load value for each proxy node 130 to be substituted so that the post-substitution load prediction value of each proxy node 130 is equalized. An estimated post-substitution load prediction value may be obtained.

  Further, if the proxy node set is configured by one proxy node 130, the post-substitution load prediction value calculation unit 102 uses the load value of the candidate node 120 as the post-substitution load prediction value of the proxy node 130, and What is necessary is just to obtain | require the total value with the value of the load of the proxy node 130. FIG.

  The post-substitution load predicted value determination unit 103 determines whether or not the post-substitution load prediction value of each proxy node 130 satisfies a predetermined condition. Here, the predetermined condition may be, for example, that the maximum value among the post-proxy load prediction values of each proxy node 130 is equal to or less than a threshold value.

  The load distribution execution unit 104 executes load distribution for reducing the load on the comparison node 110 when it is determined that the post-substitution load prediction value satisfies a predetermined condition. Specifically, the load distribution execution unit 104 moves the data handled by the candidate node 120 to the substitute node 130 to make the candidate node 120 an empty node in order to make the substitute node 130 perform the processing of the candidate node 120. Then, the load distribution execution unit 104 moves a part of the data handled by the comparison node 110 to the candidate node 120 in order to share a part of the processing handled by the comparison node 110 with the candidate node 120 that has become an empty node. Let For example, the load distribution execution unit 104 may move data from the comparison node 110 to the candidate node 120 so that the load values after load distribution of the comparison node 110 and the candidate node 120 are equal.

  The operation of the distributed processing system 1 configured as described above will be described with reference to FIG.

  First, the substitute node set creation unit 101 creates a substitute node set including one or more substitute nodes 130 for a candidate node 120 (step S1). For example, as described above, the proxy node set creation unit 101 may create a proxy node set made up of randomly selected nodes that can be the proxy node 130.

  Next, the post-substitution load prediction value calculation unit 102 performs the post-substitution load when the proxy node set performs all the processes of the candidate nodes 120 for each of the proxy nodes 130 that constitute the proxy node set created in step S1. A predicted value is calculated (step S2). For example, as described above, the post-substitution load prediction value calculation unit 102 may obtain the post-substitution load prediction value so that the post-substitution load prediction value of each substitution node 130 becomes equal.

  Next, the post-substitution load predicted value determination unit 103 determines whether or not the post-substitution load prediction value of each proxy node 130 calculated in step S2 satisfies a predetermined condition (step S3). For example, as described above, the post-substitution load prediction value determination unit 103 may determine whether or not the maximum value of the post-substitution load prediction value is equal to or less than a threshold value.

  Here, when it is determined that the post-substitution load predicted value determination unit 103 satisfies a predetermined condition, the load distribution execution unit 104 executes load distribution. Specifically, as described above, the load distribution execution unit 104 substitutes the processing of the candidate node 120 by the proxy node 130 to make the candidate node 120 an empty node. Then, the load distribution execution unit 104 assigns a part of the processing handled by the comparison node 110 to the candidate nodes 120 that have become free nodes. For example, the load distribution execution unit 104 may share a part of the processing of the comparison node 110 with the candidate node 120 so that the load values of the comparison node 110 and the candidate node 120 are equal.

  Thus, the distributed processing system 1 ends the operation.

  Note that the distributed processing system 1 may perform the above-described operation with any one of the distributed processing devices 10 constituting the distributed processing system 1 as the candidate node 120 at every predetermined timing. Further, the distributed processing system 1 may perform the above-described operation as a comparison node 110 having a load value equal to or greater than a threshold among the distributed processing devices 10 constituting the distributed processing system 1. Alternatively, the distributed processing system 1 may perform the above-described operation using a randomly selected one of the distributed processing devices 10 constituting the distributed processing system 1 as the comparison node 110.

  Next, effects of the first exemplary embodiment of the present invention will be described.

  The distributed processing system as the first exemplary embodiment of the present invention can execute load distribution more efficiently.

  The reason is that when a candidate node that is a candidate for sharing the processing of the comparison node that is the target of reducing the load is created, a proxy node set that substitutes the processing is created, and the processing of the candidate node is delegated to the proxy node set Candidate nodes that calculate the post-substitution load prediction value of each substitution node and, when the post-substitution load prediction value satisfies a predetermined condition, make the candidate node process the substitution node as an empty node and become an empty node This is because load balancing is performed to share part of the processing of the comparison node. Thereby, when the post-substitution load prediction value does not satisfy a predetermined condition (for example, the maximum value of the post-substitution load prediction value is larger than the threshold value), the distributed processing system according to the present embodiment does not perform load distribution. Therefore, the load value after load distribution of the proxy node does not become so large that further load reduction is necessary.

(Second Embodiment)
Next, a second embodiment of the present invention will be described in detail with reference to the drawings.

  FIG. 5 shows the configuration of a distributed processing system 2 as a second embodiment of the present invention. In FIG. 5, the distributed processing system 2 is configured by one or more distributed processing devices 20. Similar to the distributed processing apparatus 10 according to the first embodiment of the present invention described with reference to FIG. 1, these distributed processing apparatuses 20 are connected to each other via a network. In FIG. 5, six distributed processing devices 20 are shown, but the number of distributed processing devices constituting the distributed processing system of the present invention is not limited.

  Similarly to the distributed processing system 1 as the first embodiment of the present invention, when the distributed processing system 2 receives data to be processed, the distributed processing system 2 transmits the data to the distributed processing device 20 in charge of the data. To process the data. Each distributed processing device 20 holds the addresses of some other distributed processing devices 20 as a routing table in order to transfer the data received by itself to other distributed processing devices 20 in charge of the data. In such a distributed processing system 2, the load on each node varies depending on the amount of data received by that node.

  In addition, in the distributed processing device 20 illustrated in FIG. 5, a node that is a target for reducing the load by load distribution is referred to as a comparison node 210. In the distributed processing device 20, a node that is a candidate for sharing a part of the processing of the comparison node 210 is referred to as a candidate node 220. Further, in the distributed processing device 20, when the candidate node 220 shares the processing of the comparison node 210, the node acting as the proxy for the candidate node 220 until then is referred to as the proxy node 230. Each distributed processing device 20 can operate as the comparison node 210, the candidate node 220, and the proxy node 230.

  The distributed processing device 20 is configured by a computer device having the same hardware configuration as the distributed processing device 10 according to the first embodiment of the present invention shown in FIG.

  Next, the functional block configuration of the distributed processing system 2 is shown in FIG.

  In FIG. 6, the comparison node 210 includes a post-substitution load predicted value reception unit 211, a load distribution determination unit 212, and a load distribution execution unit 215. In addition, the candidate node 220 includes a proxy node set creation unit 221, a proxy node load reception unit 222, a post-substitution load prediction value calculation unit 223, a post-substitution load prediction value transmission unit 224, and a load distribution execution unit 225. Prepare. The proxy node 230 includes a load transmission unit 231 and a load distribution execution unit 235. Here, the post-proxy load predicted value receiving unit 211, the proxy node load receiving unit 222, the post-proxy load predicted value transmitting unit 224, the load transmitting unit 231, and the load distribution executing units 215, 225, and 235 are network The interface 1005 includes a CPU 1001 that reads a computer program and various data stored in the ROM 1003 and the storage device 1004 into the RAM 1002 and executes them. The load distribution determination unit 212 and the post-substitution load predicted value calculation unit 223 are configured by a CPU 1001 that reads a computer program and various data stored in the ROM 1003 and the storage device 1004 into the RAM 1002 and executes them. Note that the hardware configuration of each functional block of the comparison node 210, the candidate node 220, and the proxy node 230 is not limited to the above configuration.

  Next, each functional block of the candidate node 220 will be described.

  The substitute node set creation unit 221 creates a substitute node set that substitutes one or more substitute nodes 230 for all of the processes handled by the own node. For example, the proxy node set creation unit 221 may create a proxy node set by using, as the proxy node 230, another distributed processing apparatus 20 that can be the proxy node 230 and whose load value is equal to or less than a threshold value. In addition, the proxy node set creation unit 221 creates a proxy node set composed of one proxy node 230 by using the other distributed processing apparatus 20 having the smallest load value as the proxy node 230. May be. The other distributed processing device 20 that can be the proxy node 230 may be another distributed processing device 20 that can acquire the load value.

  The proxy node load receiving unit 222 receives the value of the load of the proxy node 230 for each proxy node 230 constituting the proxy node set created by the proxy node set creation unit 221.

  Based on the load value of each proxy node 230 received by the proxy node load reception unit 222, the proxy post-load predicted value calculation unit 223 substitutes the proxy node set with the processing of its own node. Obtain the estimated post-substitute load value.

  For example, the post-substitution load prediction value calculation unit 223 uses each proxy node 230 as a post-substitution load prediction value of each proxy node 230 and the value of the load of each proxy node 230 at that time and the process in charge of the own node. A total value with the load value allocated to each of the values may be obtained.

  If the proxy node set is configured by one proxy node 230, the post-substitution load prediction value calculation unit 223 uses the proxy node 230 as the post-substitution load prediction value at that point in time and the own node. What is necessary is just to obtain | require the total value with the value of the load of the process in charge of.

  Further, when the proxy node set includes a plurality of proxy nodes 230, the post-substitution load prediction value calculation unit 223 obtains post-substitution load prediction values such that the proxy load prediction values of the proxy nodes 230 are equal. May be.

  The post-substitution load predicted value transmission unit 224 transmits each post-substitution load prediction value calculated by the post-substitution load prediction value calculation unit 223 to the comparison node 210. At this time, when the post-substitution load prediction value is calculated for each of the plurality of proxy nodes 230, the post-substitution load prediction value transmission unit 224 uses the maximum value among the post-substitution load prediction values of each proxy node 230 as a comparison node. You may transmit to 210.

  In addition, for example, the post-substitution load predicted value transmission unit 224 selects a node having a load value equal to or greater than the threshold among the other distributed processing devices 20 as the comparison node 210 of the post-substitution load prediction value transmission destination. Also good. Further, for example, the post-substitution load predicted value transmission unit 224 may select any of the other nodes included in the routing table held by the own node as the destination comparison node 210 of the post-substitution load prediction value. .

  The load distribution execution unit 225 executes load distribution of contents determined by a load distribution execution unit 215 of the comparison node 210 described later. Specifically, the load distribution execution unit 225 transmits data handled by the own node to the substitute node 230 in order to cause the substitute node 230 to substitute the process handled by the own node. In addition, the load distribution execution unit 225 receives a part of data handled by the comparison node 210 from the comparison node 210 in order to share a part of the processing of the comparison node 210.

  Next, each functional block of the proxy node 230 will be described.

  The load transmission unit 231 transmits the load value of the own node to the candidate node 220. For example, the load transmission unit 231 may transmit a load value in response to a request from the candidate node 220.

  The load distribution execution unit 235 executes load distribution of the content determined by the load distribution execution unit 215 of the comparison node 210 described later. Specifically, the load distribution execution unit 235 receives the data handled by the candidate node 220 from the candidate node 220 in order to perform the processing handled by the candidate node 220.

  Next, each functional block of the comparison node 210 will be described.

  The post-substitution load prediction value receiving unit 211 receives the post-substitution load prediction value from the candidate node 220.

  The load distribution determining unit 212 determines whether or not the received post-proxy load prediction value and the load value of the own node have a predetermined relationship. Specifically, the predetermined relationship may be that the load value of the own node is equal to or greater than the post-proxy load prediction value.

  Further, the load distribution determining unit 212 determines the load distribution processing content when the received post-substitution load prediction value and the load value of the own node have a predetermined relationship. The processing contents of this load distribution are to make the candidate node 220 a vacant node by substituting the processing of the candidate node 220 to the proxy node set, and to part of the processing handled by the own node as a vacant node 220 is a process shared by 220. Then, the load distribution determination unit 212 notifies the candidate node 220 of the processing content of the determined load distribution.

  The load distribution execution unit 215 executes the load distribution determined by the load distribution determination unit 212. Specifically, the load distribution execution unit 215 transmits a part of the data handled by the own node to the candidate node 220 in order to share a part of the process of the own node with the candidate node 220. For example, the load distribution execution unit 215 may transmit a part of data handled by the own node to the candidate node 220 so that the load value of the own node is equal to the load value of the candidate node 220. Good.

  The load distribution operation of the distributed processing system 2 configured as described above will be described with reference to FIG. In FIG. 7, the left diagram shows the operation of the candidate node 220, the right diagram shows the operation of the comparison node 210, and the dashed arrows connecting the left and right represent the data flow between the nodes.

  Here, first, the proxy node set creation unit 221 of the candidate node 220 creates a proxy node set (step S11).

  Next, the proxy node load receiving unit 222 receives the load value of each proxy node 230 constituting the proxy node set created in step S11 (step S12).

  Next, the post-substitution load prediction value calculation unit 223 calculates the post-substitution load prediction value of each proxy node 230 when the proxy node set created in step S11 is delegated to the processing of the own node ( Step S13).

  Next, the post-substitution load predicted value transmission unit 224 transmits the post-substitution load prediction value calculated in step S13 to the comparison node 210 (step S14). For example, as described above, the post-substitution load predicted value transmission unit 224 may transmit the maximum value of the post-substitution load prediction value calculated in step S13 to the comparison node 210. In this step, the post-substitution load predicted value transmission unit 224 may select the other distributed processing apparatus 20 having a load value equal to or less than a threshold value as the transmission destination comparison node 210. Alternatively, the post-substitution load predicted value transmission unit 224 may select any other distributed processing device 20 held by the own node in the routing table as the comparison node 210 of the transmission destination.

  Next, the post-substitution load prediction value receiving unit 211 of the comparison node 210 receives the post-substitution load prediction value (step S15).

  Next, the load distribution determining unit 212 determines whether or not the post-proxy load prediction value received in step S15 and the load value of its own note have a predetermined relationship (step S16). For example, as described above, the load distribution determination unit 212 may determine whether or not the load value of the own node is equal to or greater than the post-proxy load prediction value.

  Here, when it is determined that the post-substitution load prediction value and the load value of the own note have a predetermined relationship, the load distribution determination unit 212 determines the content of the load distribution process (step S17). Specifically, it is determined that the processing of the candidate node 220 is delegated to the proxy node 230 to make the candidate node 220 an empty node and to share a part of the processing of the own node with the candidate node 220. Then, the load distribution determining unit 212 notifies the candidate node 220 of the load distribution processing content.

  Next, the load distribution execution unit 215 executes load distribution of the determined content (step S18). Specifically, the load distribution execution unit 215 transmits a part of data handled by the own node to the candidate node 220.

  Next, when the content of load distribution is notified (Yes in step S19), the load distribution execution unit 225 of the candidate node 220 executes load distribution of the content (step S20). Specifically, the load distribution execution unit 225 transmits data handled by the own node to the proxy node set. Then, the load distribution execution unit 225 receives part of the data handled by the comparison node 210.

  As described above, the distributed processing system 2 ends the load distribution operation.

  Note that the candidate node 220 may perform the post-substitution load predicted value transmission processing in step S14 at predetermined timings (for example, at predetermined time intervals). Alternatively, when at least one of the load value of the proxy node 230 and the load value of the own node received in step S12 is equal to or less than the threshold value determined for each candidate node 220, The load predicted value transmission process may be executed. Alternatively, the candidate node 220 may execute the post-substitution predicted load value transmission process in step S14 at the timing requested from the comparison node 210.

  Next, the effect of the second exemplary embodiment of the present invention will be described.

  The distributed processing system as the second exemplary embodiment of the present invention can execute load distribution more efficiently.

  The reason is that, among the nodes that make up the distributed processing system, the candidate node that is a candidate for sharing the processing of the comparison node whose load is to be reduced, delegates all of its own processing to the proxy node set. After the proxy load prediction value of each proxy node is calculated and transmitted to the comparison node, the comparison node performs the processing of the candidate node when the load value of the own node and the proxy load prediction value are in a predetermined relationship. This is because it is determined to execute load distribution in which a part of the processing of the comparison node is shared by the candidate node by substituting for the proxy node. As a result, when the load value of the comparison node and the post-substitution load prediction value are not in a predetermined relationship (for example, the post-substitution load prediction value is larger than the load value of the comparison node), the distributed processing system executes load distribution. do not do. Therefore, the maximum value of the load value after load distribution among the comparison node, the candidate node, and the proxy node does not become larger than the maximum value of the load value before load distribution. As a result, the distributed processing system according to the second exemplary embodiment of the present invention does not cause a case where it is necessary to reduce the load on another node by the load distribution processing that reduces the load on one node. Can be executed more efficiently.

(Third embodiment)
Next, a third embodiment of the present invention will be described in detail with reference to the drawings.

  In the present embodiment, an example in which the present invention is applied to a distributed processing system based on the Chord technique described in Non-Patent Document 1 will be described. FIG. 8 shows the configuration of a distributed processing system 3 as a third embodiment of the present invention. In FIG. 8, the distributed processing system 3 includes one or more distributed processing devices 30. Also, these distributed processing devices 30 are connected to each other via a network, like the distributed processing device 30 according to the second embodiment of the present invention described with reference to FIG. In FIG. 8, six distributed processing devices 30 are shown, but the number of distributed processing devices constituting the distributed processing system of the present invention is not limited.

  Each distributed processing device 30 has any key on the key space as a node ID. Here, the key space is a conceptual ring-shaped key space in which the key value increases clockwise. In this key space, the distance between keys is defined in the clockwise direction.

  Further, in the distributed processing system 3, data to be processed is processed by a node having a node ID closest to the key calculated from the data (successor of the key). That is, each distributed processing device 30 is related to the key in the range from the node ID of the node having the farthest node ID from the node ID of the own node (the predecessor of the own node) to the node ID of the own node. Responsible for data processing. Here, the data may be a content to be stored, or a query for requesting the stored content.

In addition, the distributed processing system 3 sequentially transfers data received from any one of the distributed processing devices 30 to the other distributed processing devices 30, so that the distributed processing device 30 in charge of processing the data finally Delivered to Each distributed processing device 30 holds addresses of some other distributed processing devices 30 as a routing table in order to transfer data to other distributed processing devices 30. For example, in Chord, each distributed processing device 30 determines the address of the successor of the key whose distance from the key that is the node ID of its own node is 2 ⁱ⁻¹ (i = 1, 2,..., M), Store in the routing table.

  In addition, in the distributed processing device 30 configuring the distributed processing system 3, a node that is a target for reducing the load by load distribution is referred to as a comparison node 310. In the distributed processing device 30, a node that is a candidate for sharing a part of the processing of the comparison node 310 is referred to as a candidate node 320. Further, in the distributed processing device 30, when the candidate node 320 shares the processing of the comparison node 310, the node acting as the proxy for the candidate node 320 until then is referred to as the proxy node 330. Each of the distributed processing devices 30 can operate as the comparison node 310, the candidate node 320, and the proxy node 330.

  Next, the functional block configuration of the distributed processing system 3 is shown in FIG.

  In FIG. 9, the comparison node 310 includes a load information reception unit 311, a load distribution determination unit 312, and a load distribution execution unit 315. In addition, the candidate node 320 includes a proxy node set creation unit 321, a proxy node load reception unit 322, a load information creation unit 323, a load information transmission unit 324, and a load distribution execution unit 325. The proxy node 330 also includes a load transmission unit 331 and a load distribution execution unit 335. The load information receiving unit 311 constitutes an embodiment of the post-substitution load predicted value receiving unit of the present invention. Moreover, the load information creation unit 323 constitutes an embodiment of the post-substitution load predicted value calculation unit of the present invention. Moreover, the load information transmission part 324 comprises one Embodiment of the post-substitution load predicted value transmission part of this invention.

  Next, each functional block of the candidate node 320 will be described.

  The substitute node set creation unit 321 creates a substitute node set composed of one substitute node 330 by using the successor of the own node (candidate node 320) as the substitute node 330.

  The proxy node load receiving unit 322 receives the load value of the proxy node 330 that is the successor of the local node.

  When the comparison node 310 is a successor of its own node, the load information creation unit 323 creates load information including a numerical value representing the load value of the own node and the node ID of the own node.

  In addition, when the comparison node 310 is not the successor of the own node, the load information creation unit 323 calculates the total value of the load value of the own node and the load value of the successor (that is, the estimated post-substitution load value of the substitute node 330 that is the successor). ) And the node ID of the own node.

The load information transmission unit 324 transmits the load information created by the load information creation unit 323 to the comparison node 310. Note that the load information transmission unit 324 may select another distributed processing device 30 as the comparison node 310 that is the transmission destination of the load information, for example, by the following method.
<Load Information Transmission Destination Selection Method (1)> A key in the key space is selected at random, and another distributed processing device 30 in charge of the selected key is selected as the comparison node 310 of the load information transmission destination.
<Load Information Transmission Destination Selection Method (2)> Random selection from nodes stored in the routing table held by the candidate node 320.
<Load Information Transmission Destination Selection Method (3)> Among the nodes stored in the routing table held by the candidate node 320, the amount per unit time of data transferred from other nodes to the node is equal to or greater than the threshold value Select a node.
<Load Information Transmission Destination Selection Method (4)> In the candidate node 320, when a transmission queue is prepared for each of other nodes serving as data transfer destinations, the amount of data stored in the transmission queue is equal to or greater than a threshold value. Select the node that is.

  The load distribution execution unit 325 executes load distribution of the content determined by the comparison node 310. Details of specific load distribution processing will be described later.

  Next, each functional block of the proxy node 330 will be described.

  The load transmission unit 331 transmits to the candidate node 320 the load value of the processing that the node is in charge of. For example, the load transmission unit 331 may transmit a load value in response to a request from the candidate node 320.

  The load distribution execution unit 335 executes load distribution of the content determined by the load distribution execution unit 315 of the comparison node 310 described later. Details of specific load distribution processing will be described later.

  Next, each functional block of the comparison node 310 will be described.

  The load information receiving unit 311 receives load information from the candidate node 320. When the own node is a successor of the candidate node 320, the load information includes the load value of the candidate node 320 and the node ID of the candidate node 320. Further, when the own node is not the successor of the candidate node 320, the load information includes the load value of the candidate node 320 and the total value of the load values of the successor (the predicted load value after the successor of the successor) and the node of the candidate node 320. It consists of ID.

When the load value included in the received load information and the load value of the own node satisfy a predetermined relationship, the load distribution determination unit 312 determines the load distribution processing content. Here, the predetermined relationship may be a relationship represented by the following equation (1).
[Load value of comparison node 310] ≧ [Load value included in received load information] (1)
Specifically, when the load distribution determination unit 312 satisfies such a predetermined relationship and the candidate node 320 that is the transmission source of the load information is the predecessor of the own node, the load distribution determination unit 312 ) And a load distribution process for moving data so that the load is evenly distributed among the candidate nodes 320. In addition, when the load distribution determination unit 312 satisfies such a predetermined relationship and the candidate node 320 that is the load information transmission source is not the predecessor of the own node, the load distribution determination unit 312 performs the processing of the candidate node 320 as a successor (proxy) The node 330) is delegated, and a load distribution process for allocating a part of the process of the own node (comparison node 310) to the candidate node 320 is determined. At this time, the load distribution determining unit 312 moves from the own node to the candidate node 320 so that the load value after load distribution of the own node is equal to the load value after load distribution of the candidate node 320. Data may be determined. In addition, when the predetermined relationship is not satisfied, the load distribution determining unit 312 determines not to execute load distribution between the candidate node 320 and the own node.

  In addition, when the load distribution determination unit 312 determines to execute load distribution, the load distribution determination unit 312 notifies the candidate node 320 of the processing content.

  The load distribution execution unit 315 executes load distribution of the content determined by the load distribution determination unit 312.

  The operation of the distributed processing system 3 configured as described above will be described in detail with reference to FIGS. 10 and 11. 10 shows the operation of the candidate node 320, and FIG. 11 shows the operation of the comparison node 310. 10 and 11 indicate the flow of data between the candidate node 320 and the comparison node 310.

  In FIG. 10, first, the proxy node set creation unit 321 of the candidate node 320 selects the successor of its own node as the proxy node 330 constituting the proxy node set (step S <b> 21).

  Next, the proxy node load receiving unit 322 receives the load value from the successor (step S22).

  Next, the load information creation unit 323 determines whether or not the comparison node 310 is a successor of its own node (step S23). Here, the load information creation unit 323 uses the load information transmission unit 324 for, for example, the comparison node 310 selected using any one of the load information transmission destination selection methods (1) to (4) described above. Judgment should be made.

  Here, when the comparison node 310 is the successor of the own node, the load information creation unit 323 creates load information including the load value of the own node and the own node ID (step S24).

  On the other hand, when the comparison node 310 is not the successor of the own node, the load information creation unit 323 determines the total value of the load value of the own node and the load value of the successor (predicted post-substitute load prediction value), the own node ID, Is created (step S25).

  Next, the load information transmission unit 324 transmits the load information created in step S24 or step S25 to the comparison node 310 (step S26).

  Next, the operation of the comparison node 310 that has received the load information will be described with reference to FIG.

  In FIG. 11, first, the load information receiving unit 311 of the comparison node 310 receives load information from the candidate node 320 (step S27).

  Next, the load distribution determining unit 312 determines whether or not the load value included in the received load information and the load value of the own node satisfy a predetermined relationship (step S28). For example, as described above, the predetermined relationship may be that the load value of the own node is equal to or greater than the load value included in the received load information.

  If it is determined that the predetermined relationship is satisfied, the load distribution determining unit 312 further determines whether or not the candidate node 320 that is the load information transmission source is a predecessor of the own node (comparison node 310). (Step S29).

  Here, when the candidate node 320 is a predecessor of the own node, the load distribution determination unit 312 moves the data between the two nodes so that the load values are equal between the own node and the candidate node 320. Is determined (step S30). Then, the candidate node 320 is notified of the processing contents of load distribution.

  On the other hand, when the candidate node 320 is not the predecessor of the own node, the load distribution determining unit 312 substitutes the successor for the processing of the candidate node 320 and automatically sets the load value between the own node and the candidate node 320 to be equal. Load distribution for moving data from the node to the candidate node 320 is determined (step S31). Then, the candidate node 320 is notified of the processing contents of load distribution.

  If it is determined in step S28 that the predetermined relationship is not satisfied, the distributed processing system 3 does not execute load distribution.

  Next, the load distribution execution unit 315 of the comparison node 310 transmits / receives data to / from the candidate node 320 according to the content of the load distribution determined in step S30 or step S31 (step S32).

  Next, in FIG. 10, when the load distribution execution unit 325 of the candidate node 320 is notified of the load distribution processing content from the comparison node 310 (Yes in step S33), the load distribution is executed according to the notified processing content. (Step S34). Specifically, if the comparison node 310 is a successor of its own node, the load distribution execution unit 325 transmits and receives data so that the load values after load distribution are equalized between the comparison node 310 and its own node. . If the comparison node 310 is not the successor of its own node, the load distribution execution unit 325 transmits all of the data handled by the own node to the successor. Then, the load distribution execution unit 325 receives data from the comparison node 310 so that the load values after load distribution are equal between the comparison node 310 and the own node.

  Thus, the distributed processing system 3 ends the operation.

  Next, the load distribution operation of the distributed processing system 3 will be described with a specific example. Here, load distribution operation when the distributed processing system 3 according to the present embodiment is applied to a distributed processing system constituted by eight distributed processing devices from the node a to the node h as shown in FIG. explain. As described above, in FIG. 12, each node has a node ID arranged in the order shown in the key space 900 on the ring. Further, the numerical value shown beside the graphic indicating each node represents the load value of each node.

  First, an operation example when the node e is the candidate node 320 and the node e is selected as the comparison node 310 of the load information transmission destination by the node e will be described.

  In this case, first, the proxy node set creation unit 321 of the node e selects the node f that is the successor of the node e as the proxy node 330 (step S21).

  Then, the proxy node load receiving unit 322 of the node e receives the load value “20%” of the node f from the node f (step S22).

  Next, since the node a as the comparison node 310 is not the successor of the own node (No in step S23), the load information creation unit 323 of the node e determines the load value “20%” of the node f and the load of the own node. Load information consisting of "50%" (successor's surrogate post-substitute load prediction value) that is the total value of "30%" and the node ID of the own node is created (step S25).

  Next, the load information transmission unit 324 transmits the created load information to the node a as the comparison node 310 (step S26).

  Next, in the node a that receives the load information, the load distribution determination unit 312 has the load value “80%” of the own node equal to or greater than the load value “50%” included in the received load information ( In step S28, Yes), it is determined that the node e is not the predecessor of the node a (No in step S29). Therefore, the load distribution determining unit 312 of the node a determines load distribution that causes the node f to perform the process of the node e and also shares a part of the process of the node a with the node e (step S31). Then, the load distribution execution units 315, 325, and 335 of the node a, the node e, and the node f execute the determined load distribution. As a result, the load values after the load distribution of the node a and the node e are each 40%. Further, the load value after load distribution of the node f is 50%.

  This is the end of the description of the specific example of the operation when the node e is the candidate node 320 and the node a is the comparison node 310.

  Next, an operation example when the node f is the candidate node 320 and the node a is selected as the load information comparison node 310 by the node f will be described.

  In this case, first, the proxy node set creation unit 321 of the node f selects the node g that is the successor of the node f as the proxy node 330 (step S21).

  Then, the proxy node load receiving unit 322 of the node f receives the load value “79%” of the node g from the node g (step S22).

  Next, since the node a as the comparison node 310 is not the successor of the own node (No in step S23), the load information creation unit 323 of the node f determines the load value “79%” of the node g and the load of the own node. Load information consisting of “99%” (successor's proxy post-substitution predicted load value), which is the total value of “20%”, and the node ID of the own node is created (step S25).

  Next, the load information transmission unit 324 transmits the created load information to the node a as the comparison node 310 (step S26).

  Next, in the node a that has received the load information, the load distribution determining unit 312 determines that the load value “80%” of the own node is not equal to or greater than the load value “99%” included in the received load information. (No in step S28), it is determined not to perform load distribution.

  Above, description of the specific example of load distribution operation | movement of the distributed processing system 3 is complete | finished.

  Next, effects of the third exemplary embodiment of the present invention will be described.

  The distributed processing system according to the third exemplary embodiment of the present invention is a node group related to load distribution processing for reducing the load of a certain node, after the load distribution is greater than the maximum load value before load distribution. It is possible to prevent the maximum value of the load value from increasing.

  The reason is that the load distribution determination unit of the comparison node has a load value of the own node that is equal to or greater than the predicted post-substitution load value of the proxy node that performs the process of the candidate node that shares the process of the self node (predetermined relationship). This is because it is decided to execute load balancing.

  Here, the load value before load distribution of the comparison node is l_a, the load value after load distribution is l_a ', the load value before load distribution of the candidate node is l_x, and the load value after load distribution is l_x It is assumed that the load value before load distribution of the proxy node is l_y, and the load value after load distribution is l_y. At this time, the predetermined relationship is expressed as l_a ≧ l_x + l_y. Further, the post-substitution load prediction value l_y ′ is calculated by l_x + l_y.

  If the post-proxy load prediction value l_y ′ (= l_x + l_y) is larger than l_a (No in step S28), the distributed processing system according to the present embodiment does not execute load distribution. Therefore, max {l_a, l_x, l_y} = max {l_a ′, l_x ′, l_y ′}. Here, max {} represents the maximum value among the elements. Accordingly, in the related node group (comparison node, candidate node, and proxy node), the load value after load distribution does not become larger than the maximum load value before load distribution.

Further, when the post-substitution load predicted value l_y ′ (= l_x + l_y) is equal to or less than l_a (Yes in step S28), max {l_a, l_x, l_y} is l_a. In this case, the distributed processing system as the present embodiment performs load balancing,
l_a '= l_a × 0.5,
l_x '= l_a × 0.5,
l_y '= l_x + l_y
It becomes. At this time, max {l_a ′, l_x ′, l_y ′} is l_a × 0.5 or l_x + l_y, and both values are equal to or less than l_a. Therefore, max {l_a, l_x, l_y} ≧ max {l_a ', l_x', l_y '}, and the maximum value of the load value before load distribution in the related node group (comparison node, candidate node, and proxy node) Thus, the load value after load distribution does not increase.

  As described above, in the distributed processing system according to the present embodiment, when the load value of the comparison node is equal to or higher than the post-proxy load prediction value of the proxy node, by executing load distribution, the related node group (comparison Node, candidate node, and proxy node), the load value after load distribution does not become larger than the maximum load value before load distribution. As a result, in the distributed processing system according to the present embodiment, cases where further load distribution is necessary after load distribution can be reduced, and the number of load distribution processes can be reduced.

  For example, in FIG. 12, it has been described that when the node f is the candidate node 320 and the node a is the comparison node 310, the distributed processing system according to the present embodiment does not perform load distribution. Here, consider the case where the related technique described in Non-Patent Document 1 is applied in FIG. At this time, it is assumed that the node a has acquired the load value (l_f) of the node f. Here, l_f (20%) ≦ ε × l_a (80%). In addition, the node a is not the successor of the node f, and the load value of the node a (l_a = 80%) is larger than the load value of the node g that is the successor of the node f (l_c = 79%). Therefore, this case corresponds to case 3 in the related art. Therefore, the node f becomes a free node by moving the held data to the node g. The node a moves the data to be held to the node f so that the load value of the node a is halved. As a result, the load value of the node a and the node f is 40%, and the load value of the node g is 99%. As a result, in the related art, load distribution for reducing the load of the node g having the load value of 99% is further executed. On the other hand, when the load value is transmitted from the node f to the node a, the distributed processing system according to the present embodiment does not perform load distribution as described above, and transmits the load value from the node e to the node a, for example. If this happens, load balancing will be executed. Therefore, in such a specific example, the distributed processing system according to the present embodiment reduces the number of load distribution processes.

  In the present embodiment, the proxy node set creation unit for candidate nodes has been described as creating a proxy node set composed of one proxy node using the successor of the candidate node as a proxy node. In addition, the proxy node set creation unit of candidate nodes uses n (n is an integer of 1 or more) nodes as proxy nodes in order from the node whose distance in the clockwise direction is close to the candidate node. A configured proxy node set may be created.

Further, in the present embodiment, the following can be adopted as the timing at which the candidate node transmits load information to the comparison node.
<Load transmission timing (1)> Predetermined timing. For example, every certain time or timing according to a predetermined schedule.
<Load transmission timing (2)> Timing at which the load value of at least one of the candidate node and the proxy node becomes equal to or less than a threshold value determined for each.
<Load transmission timing (3)> When there is a node whose number or amount of data transferred per unit time is equal to or greater than a threshold among other nodes whose addresses are stored in the routing table of the candidate node.
<Load transmission timing (4)> When a transmission queue is prepared for each data transfer destination node in the candidate node, and the number or amount of data stored in the transmission queue for a certain node exceeds a threshold value .

  The candidate node may determine such timing and transmit the post-substitution load prediction value to the comparison node, or the comparison node may determine such timing and send the post-substitution load prediction value to the candidate node. You may request to send a value.

  In the second and third embodiments of the present invention, the post-substitution load prediction value reception unit (or load information reception unit) of the comparison node may receive post-substitution load prediction values from a plurality of candidate nodes. Good. In this case, the load distribution determining unit of the comparison node may determine the load distribution using the candidate node of the transmission source predicted load value with the smallest value and the proxy node.

  In the second and third embodiments of the present invention, when the load distribution determination unit of the comparison node determines not to execute load distribution, the post-proxy load prediction value received from this candidate node is May be transferred to the comparison node. In this case, it is only necessary to determine whether or not to perform load distribution by the other comparison nodes to which the post-proxy load prediction value has been transferred operating in the same manner. As a result, the distributed processing system according to each embodiment can execute load distribution that reduces the load on any node with a high load value based on the post-proxy load prediction value calculated by a certain candidate node. become.

  In the second and third embodiments of the present invention, the candidate node acts as a substitute for the comparison node when at least one of the load value of the own node or the load value of the substitute node is equal to or greater than the threshold value. The post-load predicted value may not be transmitted. As a result, the distributed processing system according to each embodiment determines whether or not to execute post-substitution load prediction value transmission processing and load distribution even though there is a high possibility of determining not to execute load distribution. The load distribution can be executed more efficiently without performing the determination process.

  In addition, the distributed processing system according to each embodiment described above performs the following processing before executing load balancing when it is determined to perform load balancing using candidate nodes and proxy nodes. May be. In other words, in this case, the distributed processing system according to each embodiment, when the load value of the proxy node is equal to or greater than the load value of the proxy node, part of the processing of the proxy node is processed by the proxy node. You may make it perform the load distribution to share in advance. Thereby, the distributed processing system as each embodiment can further reduce the increase in the load value of the proxy node due to load distribution.

  In the second and third embodiments of the present invention, the candidate node creates a proxy node set and calculates a post-substitution load prediction value, and the comparison node determines that the post-substitution load prediction value satisfies a predetermined condition. An example has been described in which it is determined whether or not to perform load balancing based on whether or not it is satisfied. Not limited to this, in the distributed processing system of the present invention, the proxy node set creation process may be performed by a node other than the candidate nodes. In the distributed processing system of the present invention, the post-substitution load prediction value calculation process may be performed by a node other than the candidate node. Further, in the distributed processing system of the present invention, the determination process as to whether or not the post-substitution load prediction value satisfies a predetermined condition may be performed by a node other than the comparison node.

  In each embodiment of the present invention described above, the operation of each node described with reference to each flowchart is stored as a computer program of the present invention in a storage device (storage medium) of the computer device. The computer program may be read and executed by the CPU. In such a case, the present invention is constituted by the code of the computer program or a storage medium.

  Moreover, each embodiment mentioned above can be implemented in combination as appropriate.

  The present invention is not limited to the above-described embodiments, and can be implemented in various modes.

Moreover, a part or all of each embodiment mentioned above can be described as the following supplementary notes, but is not limited to the following.
(Appendix 1)
In a distributed processing system composed of one or more distributed processing devices (nodes),
For a node (candidate node) that is a candidate for sharing a part of processing handled by a certain node (comparison node), all of the processing handled by the candidate node is performed using one or more nodes (proxy nodes). Create a proxy node set for proxy,
Obtaining a predicted load value after proxy of each proxy node when the proxy node set performs processing of the candidate node (post-substitute load prediction value),
When the post-substitution load prediction value of each proxy node satisfies a predetermined condition, the candidate node processing is substituted for the proxy node set to make the candidate node an empty node, and the comparison node is in charge Load distribution method for executing load distribution in which a part of the node is shared by the candidate nodes that have become free nodes.
(Appendix 2)
The load distribution is performed when the predetermined condition is satisfied with the predetermined condition that the predicted load value after the proxy of each proxy node and the load value before the load distribution of the comparison node are in a predetermined relationship. The load balancing method according to supplementary note 1, wherein:
(Appendix 3)
In a key space in which a key value increases clockwise and a distance between keys is defined in a clockwise direction, each node constituting the distributed processing system is a conceptual ring-shaped key space. One of the keys is used as a node ID, and the data to be processed by the distributed processing system is determined by the node having the closest node ID from the key calculated from the data (the successor of the key). When processed
Note 1 or 2, wherein, as the proxy node set, a set composed of n nodes (n is an integer of 1 or more) in order from a node having a distance close to the candidate node in the key space is created. The load distribution method according to attachment 2.
(Appendix 4)
The proxy node set is constituted by one proxy node (successor of candidate nodes),
When the post-proxy load prediction value of the proxy node and the load value before load distribution of the comparison node are in the predetermined relationship,
If the comparison node is a successor of the candidate node, load distribution is performed to move data between the comparison node and the candidate node so that the loads of the comparison node and the candidate node are equal.
If the comparison node is not a successor of the candidate node, the comparison node is moved so that all the data held by the candidate node is moved to the proxy node and the loads of the comparison node and the candidate node are equalized. The load balancing method according to appendix 3, wherein load balancing is performed to move data from a node to the candidate node.
(Appendix 5)
5. The load distribution is performed by using, as a comparison node, a node having a load value equal to or greater than a threshold value among the nodes constituting the distributed processing system, according to any one of appendix 1 to appendix 4, Load balancing method.
(Appendix 6)
The load distribution according to any one of appendix 1 to appendix 4, wherein the load distribution is executed by using a randomly selected node as the comparison node among the nodes constituting the distributed processing system. Method.
(Appendix 7)
Any one of Supplementary Note 1 to Supplementary Note 4, wherein the load distribution is executed by using any one of the other nodes included in the routing table holding information on the other nodes in the candidate node as the comparison node. Load balancing method described in 1.
(Appendix 8)
A proxy post-substitution load prediction value of each proxy node is obtained at a timing when the amount of data transferred per unit time exceeds a threshold with respect to any of the other nodes included in the routing table in the candidate node The load distribution method according to appendix 7, wherein the load distribution is executed using the other node as the comparison node.
(Appendix 9)
When a transmission queue is prepared for each other node that is a data transfer destination in the candidate node, at a timing when the amount of data stored in the transmission queue for any other node is equal to or more than a threshold, The load distribution according to any one of appendix 1 to appendix 4, wherein the load distribution is executed by using the other node as the comparison node by obtaining a post-proxy load prediction value of each of the proxy nodes. Method.
(Appendix 10)
The load distribution method according to any one of appendix 1 to appendix 7, wherein the load distribution is executed by obtaining a post-proxy load prediction value of each of the proxy nodes at every predetermined timing.
(Appendix 11)
The load distribution is obtained by obtaining a post-proxy load prediction value of each proxy node at a timing when at least one of the load value of the candidate node and the load value of the proxy node is equal to or less than a threshold value defined for each of the candidate nodes. The load distribution method according to any one of appendix 1 to appendix 7, wherein:
(Appendix 12)
The proxy node set is created for each of the plurality of candidate nodes, one of the candidate nodes is selected based on the post-proxy load prediction value in each of the generated proxy node sets, and the selected candidate node and the proxy node set are selected. The load distribution method according to any one of appendix 1 to appendix 11, wherein the load distribution of the comparison node is executed.
(Appendix 13)
When the post-proxy load prediction value and the load value before load distribution of the comparison node are not in the predetermined relationship, the post-proxy load prediction value is the predetermined relationship with another node as a new comparison node. 13. The load distribution method according to any one of appendix 2 to appendix 12, wherein whether or not to execute the load distribution for the new comparison node is determined based on whether or not the condition is satisfied.
(Appendix 14)
As the post-substitution load prediction value of each proxy node, a total value of a load value to be substituted for each proxy node and a load value of each proxy node among all the processes of the candidate nodes is obtained. 14. The load distribution method according to any one of supplementary note 1 to supplementary note 13, characterized by:
(Appendix 15)
The post-substitution load prediction value of each proxy node is obtained by calculating a load value for proxying to each of the proxy nodes so that the post-substitution load prediction values of the proxy nodes are equal. The load balancing method according to appendix 14.
(Appendix 16)
When the proxy node set consists of one proxy node, a total value of a load value of the candidate node and a load value of the proxy node is obtained as a post-substitution load prediction value of the proxy node. The load distribution method according to appendix 14.
(Appendix 17)
The load distribution before load distribution of the comparison node is equal to or greater than the maximum value of the post-proxy load prediction value of each proxy node, and the load distribution is performed when the predetermined relationship is satisfied. The load balancing method according to any one of Supplementary Note 2 to Supplementary Note 16, which is executed.
(Appendix 18)
The supplementary note 1 is characterized in that the load distribution is executed when the predetermined condition is satisfied, with the predetermined condition that the maximum value of the post-proxy load prediction value of each proxy node is equal to or less than a threshold value. Load balancing method.
(Appendix 19)
The load distribution method according to Supplementary Note 1 or Supplementary Note 2, wherein the proxy node set is created by using, as the proxy node, a node having a load value equal to or less than a threshold among the nodes that can be the proxy node.
(Appendix 20)
20. The load distribution method according to appendix 19, wherein the proxy node set is created by using, as the proxy node, a node having a minimum load value among the nodes that can be the proxy node.
(Appendix 21)
A distributed processing system including one or more distributed processing devices (nodes),
Candidate nodes that are candidates for sharing a part of the processing handled by other nodes (comparison nodes)
A proxy node set creation unit that creates a proxy node set that substitutes one or more nodes (proxy nodes) for all of the processes handled by the own node;
A proxy node load receiving unit that receives a load value of each proxy node;
A post-substitution load that obtains a predicted value (post-substitution load prediction value) after the substitution of each substitution node when the substitution node set substitutes the processing of the candidate node based on the load value of each substitution node A predicted value calculation unit;
A post-substitution load prediction value transmission unit that transmits the post-substitution load prediction value to the comparison node;
A load distribution execution unit for executing load distribution determined by the comparison node;
With
The proxy node is
A load transmitter that transmits a load value of the own node to the candidate node;
A load distribution execution unit for executing load distribution determined by the comparison node;
With
The comparison node is
A post-substitution load prediction value receiving unit that receives the post-substitution load prediction value;
When the post-substitution load prediction value satisfies a predetermined condition, the candidate node process is substituted for the substitute node set to make the candidate node an empty node, and a part of the process that the comparison node is in charge of A load distribution determination unit that determines to execute load distribution to be shared among the candidate nodes that have become free nodes;
A load balancing execution unit for executing the determined load balancing;
A distributed processing system.
(Appendix 22)
A distributed processing device (node) constituting a distributed processing system,
When the own node is a node (candidate node) that is a candidate for sharing a part of the process handled by another node (comparison node), one or more other nodes handle all of the processes handled by the own node A proxy node set creation unit that creates a proxy node set to be delegated using (proxy node);
A proxy node load receiving unit that receives a load value of each proxy node;
A post-substitution load that obtains a predicted value (post-substitution load prediction value) after the substitution of each substitution node when the substitution node set substitutes the processing of the candidate node based on the load value of each substitution node A predicted value calculation unit;
A post-substitution load prediction value transmission unit that transmits the post-substitution load prediction value to the comparison node;
A load distribution execution unit for executing load distribution determined by the comparison node;
A distributed processing apparatus.
(Appendix 23)
Using the distributed processing devices (nodes) constituting the distributed processing system,
When the own node is a node (candidate node) that is a candidate for sharing a part of the process handled by another node (comparison node), one or more other nodes handle all of the processes handled by the own node Create a proxy node set to proxy using (proxy node),
Receiving the load value of each proxy node;
Based on the load value of each proxy node, obtain a predicted value (post-substitution load prediction value) of the proxy node after the proxy node when the proxy node set performs the processing of the candidate node,
Sending the post-substitution load prediction value to the comparison node;
A load distribution method for executing load distribution determined by the comparison node.
(Appendix 24)
In the computer device constituting the distributed processing device (node) included in the distributed processing system,
When the own node is a node (candidate node) that is a candidate for sharing a part of the process handled by another node (comparison node), one or more other nodes handle all of the processes handled by the own node A proxy node set creation step of creating a proxy node set to be delegated using (proxy node);
A proxy node load receiving step of receiving a load value of each proxy node;
A post-substitution load that obtains a predicted value (post-substitution load prediction value) after the substitution of each substitution node when the substitution node set substitutes the processing of the candidate node based on the load value of each substitution node A predicted value calculating step;
A post-substitution load prediction value transmission step of transmitting the post-substitution load prediction value to the comparison node;
A load balancing execution step for performing load balancing determined by the comparison node;
A computer program that runs
(Appendix 25)
A distributed processing device (node) constituting a distributed processing system,
When the own node is a comparison node that shares a part of the processing in charge with other candidate nodes,
Receiving a post-proxy load prediction value receiving from the candidate node a predicted load value (post-replacement load prediction value) of each proxy node that constitutes a proxy node set that performs all of the processes handled by the candidate node And
When the post-substitution load prediction value satisfies a predetermined condition, the candidate node process is substituted for the substitute node set to make the candidate node an empty node, and a part of the process that the comparison node is in charge of A load distribution determination unit that determines to execute load distribution to be shared among the candidate nodes that have become free nodes;
A load balancing execution unit for executing the determined load balancing;
A distributed processing apparatus.
(Appendix 26)
Using the distributed processing devices (nodes) constituting the distributed processing system,
When the own node is a comparison node that shares a part of the processing in charge with other candidate nodes,
Receiving from the candidate node the predicted value of the post-proxy load of each proxy node that constitutes the proxy node set that performs all of the processes in charge of the candidate node (post-substitute load predicted value);
When the post-substitution load prediction value satisfies a predetermined condition, the candidate node process is substituted for the substitute node set to make the candidate node an empty node, and a part of the process that the comparison node is in charge of , Decide to execute load distribution to be shared among the candidate nodes that have become free nodes,
A load balancing method that performs the determined load balancing.
(Appendix 27)
In the computer device constituting the distributed processing device (node) included in the distributed processing system,
When the own node is a comparison node that divides a part of the processing in charge to other candidate nodes, the proxy node of each proxy node constituting the proxy node set that performs all of the processing in charge of the candidate node A post-substitution load prediction value receiving step of receiving a predicted load value (post-substitution load prediction value) from the candidate node;
When the post-substitution load prediction value satisfies a predetermined condition, the candidate node process is substituted for the substitute node set to make the candidate node an empty node, and a part of the process that the comparison node is in charge of A load distribution determining step for determining to execute load distribution to be shared by the candidate nodes that have become empty nodes;
A load balancing execution step for executing the determined load balancing;
A computer program that runs

1, 2, 3 Distributed processing system 101 Proxy node set creation unit 102 Post-substitution load prediction value calculation unit 103 Post-substitution load prediction value determination unit 104 Load distribution execution unit 10, 20, 30 Distributed processing device 110, 210, 310 Comparison node 120, 220, 320 Candidate nodes 130, 230, 330 Proxy node 211 Load prediction value receiving unit 212, 312 Load distribution determination unit 215, 315 Load distribution execution unit 221, 321 Proxy node set creation unit 222, 322 Proxy node load Receiving unit 223 Post-substitution load prediction value calculation unit 224 Sub-substitute load prediction value transmission unit 225, 325 Load distribution execution unit 231, 331 Load transmission unit 235, 335 Load distribution execution unit 311 Load information reception unit 323 Load information creation unit 324 Load Information transmission unit 1001 CPU
1002 RAM
1003 ROM
1004 Storage device 1005 Network interface

Claims (10)

In a distributed processing system composed of one or more distributed processing devices (nodes),
For a node (candidate node) that is a candidate for sharing a part of processing handled by a certain node (comparison node), all of the processing handled by the candidate node is performed using one or more nodes (proxy nodes). Create a proxy node set for proxy,
Obtaining a predicted load value after proxy of each proxy node when the proxy node set performs processing of the candidate node (post-substitute load prediction value),
When the post-substitution load prediction value of each proxy node satisfies a predetermined condition, the candidate node processing is substituted for the proxy node set to make the candidate node an empty node, and the comparison node is in charge Load distribution method for executing load distribution in which a part of the node is shared by the candidate nodes that have become free nodes.   The load distribution is performed when the predetermined condition is satisfied with the predetermined condition that the predicted load value after the proxy of each proxy node and the load value before the load distribution of the comparison node are in a predetermined relationship. The load distribution method according to claim 1, wherein: In a key space in which a key value increases clockwise and a distance between keys is defined in a clockwise direction, each node constituting the distributed processing system is a conceptual ring-shaped key space. One of the keys is used as a node ID, and the data to be processed by the distributed processing system is determined by the node having the closest node ID from the key calculated from the data (the successor of the key). When processed
2. The proxy node set is configured to create a set composed of n nodes (n is an integer of 1 or more) in order from a node closest to the candidate node in the key space. Or the load distribution method of Claim 2. The proxy node set is constituted by one proxy node (successor of candidate nodes),
When the post-proxy load prediction value of the proxy node and the load value before load distribution of the comparison node are in the predetermined relationship,
If the comparison node is a successor of the candidate node, load distribution is performed to move data between the comparison node and the candidate node so that the loads of the comparison node and the candidate node are equal.
If the comparison node is not a successor of the candidate node, the comparison node is moved so that all the data held by the candidate node is moved to the proxy node and the loads of the comparison node and the candidate node are equalized. 4. The load balancing method according to claim 3, wherein load balancing is performed to move data from a node to the candidate node.   5. The load distribution is performed by using, as the comparison node, a node having a load value equal to or greater than a threshold value among the nodes constituting the distributed processing system. Load balancing method described in 1. A distributed processing system including one or more distributed processing devices (nodes),
Candidate nodes that are candidates for sharing a part of the processing handled by other nodes (comparison nodes)
A proxy node set creation unit that creates a proxy node set that substitutes one or more nodes (proxy nodes) for all of the processes handled by the own node;
A proxy node load receiving unit that receives a load value of each proxy node;
A post-substitution load that obtains a predicted value (post-substitution load prediction value) after the substitution of each substitution node when the substitution node set substitutes the processing of the candidate node based on the load value of each substitution node A predicted value calculation unit;
A post-substitution load prediction value transmission unit that transmits the post-substitution load prediction value to the comparison node;
A load distribution execution unit for executing load distribution determined by the comparison node;
With
The proxy node is
A load transmitter that transmits a load value of the own node to the candidate node;
A load distribution execution unit for executing load distribution determined by the comparison node;
With
The comparison node is
A post-substitution load prediction value receiving unit that receives the post-substitution load prediction value;
When the post-substitution load prediction value satisfies a predetermined condition, the candidate node process is substituted for the substitute node set to make the candidate node an empty node, and a part of the process that the comparison node is in charge of A load distribution determination unit that determines to execute load distribution to be shared among the candidate nodes that have become free nodes;
A load balancing execution unit for executing the determined load balancing;
A distributed processing system. A distributed processing device (node) constituting a distributed processing system,
When the own node is a node (candidate node) that is a candidate for sharing a part of the process handled by another node (comparison node), one or more other nodes handle all of the processes handled by the own node A proxy node set creation unit that creates a proxy node set to be delegated using (proxy node);
A proxy node load receiving unit that receives a load value of each proxy node;
A post-substitution load that obtains a predicted value (post-substitution load prediction value) after the substitution of each substitution node when the substitution node set substitutes the processing of the candidate node based on the load value of each substitution node A predicted value calculation unit;
A post-substitution load prediction value transmission unit that transmits the post-substitution load prediction value to the comparison node;
A load distribution execution unit for executing load distribution determined by the comparison node;
A distributed processing apparatus. In the computer device constituting the distributed processing device (node) included in the distributed processing system,
When the own node is a node (candidate node) that is a candidate for sharing a part of the process handled by another node (comparison node), one or more other nodes handle all of the processes handled by the own node A proxy node set creation step of creating a proxy node set to be delegated using (proxy node);
A proxy node load receiving step of receiving a load value of each proxy node;
A post-substitution load that obtains a predicted value (post-substitution load prediction value) after the substitution of each substitution node when the substitution node set substitutes the processing of the candidate node based on the load value of each substitution node A predicted value calculating step;
A post-substitution load prediction value transmission step of transmitting the post-substitution load prediction value to the comparison node;
A load balancing execution step for performing load balancing determined by the comparison node;
A computer program that runs A distributed processing device (node) constituting a distributed processing system,
When the own node is a comparison node that shares a part of the processing in charge with other candidate nodes,
Receiving a post-proxy load prediction value receiving from the candidate node a predicted load value (post-replacement load prediction value) of each proxy node that constitutes a proxy node set that performs all of the processes handled by the candidate node And
When the post-substitution load prediction value satisfies a predetermined condition, the candidate node process is substituted for the substitute node set to make the candidate node an empty node, and a part of the process that the comparison node is in charge of A load distribution determination unit that determines to execute load distribution to be shared among the candidate nodes that have become free nodes;
A load balancing execution unit for executing the determined load balancing;
A distributed processing apparatus. In the computer device constituting the distributed processing device (node) included in the distributed processing system,
When the own node is a comparison node that divides a part of the processing in charge to other candidate nodes, the proxy node of each proxy node constituting the proxy node set that performs all of the processing in charge of the candidate node A post-substitution load prediction value receiving step of receiving a predicted load value (post-substitution load prediction value) from the candidate node;
When the post-substitution load prediction value satisfies a predetermined condition, the candidate node process is substituted for the substitute node set to make the candidate node an empty node, and a part of the process that the comparison node is in charge of A load distribution determining step for determining to execute load distribution to be shared by the candidate nodes that have become empty nodes;
A load balancing execution step for executing the determined load balancing;
A computer program that runs

Images