JP2017054190A - Node and method for optimizing slot - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the deviation of load toward a specific node in decreasing or increasing nodes that constitute a distributed processing system.SOLUTION: A node 1 constituting a distributed processing system comprises: a storage unit 30 for storing, in correlation to each of node IDs of a plurality of nodes 1, an allocation ID table 200 in which a slot indicating a concerned area in an ID space is stored; a node load measurement unit 14 for measuring the load information of the own node 1; and a slot optimization processing unit which, when a decrease or increase occurs, determines a node whose node ID is to be changed from among the plurality of nodes 1, changes the node ID of the determined node to a position in the ID space for reducing the deviation of load about the plurality of nodes 1 after the decrease or increase using the load information of the plurality of nodes 1, and stores the changed node ID in the allocation ID information.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a node and slot optimization method for reducing the load bias of each node in a distributed processing system that stores data by clustering nodes distributed on a network.

  In recent years, with the rise of cloud computing, it has been required to efficiently process and retain a large amount of data. Thus, distributed processing technology has been developed that realizes efficient processing by operating a plurality of servers in a coordinated manner.

  When performing distributed processing, it is necessary to determine data to be handled by each server (hereinafter referred to as “node”) constituting a distributed processing system having a cluster configuration. At this time, in order to increase the processing capability of the entire distributed processing system, it is desirable that the number of data handled by each node is averaged.

  As a representative data management method, a remainder obtained by dividing a value obtained by multiplying the key of each data by a hash function (hereinafter referred to as “hash (key)”) by the number of nodes N, that is, “hash (key) mod N”. There is a method in which a node having a number as a number manages data. In this case, it is assumed that numbers “0” to “N−1” are assigned to each node in advance. When such a management method is used, when a node is added or removed, the value of N changes, and the node in charge of storing the data changes for a lot of data. It will be necessary to rearrange.

  Therefore, as a method for suppressing the number of data that the node in charge changes with the addition / detachment of a node to about 1 / N, a data management method using a consistent hashing method (see Non-Patent Document 1). There is. This consistent hash method is used in Amazon Dynamo (see Non-Patent Document 2) and the like.

  In this data management method using the consistent hash method, IDs (IDentifiers) are assigned to both nodes and data. Then, when the closed ID space is traced clockwise from the ID of the data, the node that hits first is assumed to be in charge of the data. An example of how to give an ID to a node is a value (hash (IP address)) obtained by multiplying an IP address by a hash function.

  In a clustered distributed processing system, when the processing performance of each node is equal, the amount of data handled by each node is made equal, that is, a consistent hash method ID space (hereinafter, simply referred to as “ID space”). It is desirable to make the distances between the nodes (hereinafter referred to as “node assigned areas” or “slots”) equal. In order to realize this point, a method of virtually giving a plurality of IDs to each node is used (see Non-Patent Document 3). By having each node have a plurality of virtual IDs, even if the size of the assigned area for each virtual ID is different, the assigned area size for each node is averaged according to the law of large numbers.

David Karger, et al., “Consistent Hashing and Random Trees: Distributed Caching Protocols for Relieving Hot Spots on the World Wide Web”, [online], 1997, ACM, [August 25, 2015 search], Internet <URL : http: //www.akamai.com/dl/technical_publications/ConsistenHashingandRandomTreesDistributedCachingprotocolsforrelievingHotSpotsontheworldwideweb.pdf> Giuseppe DeCandia, et al., “Dynamo: Amazon's Highly Available Key-value Store”, SOSP'07, October 14-17, 2007, Stevenson, Washington, USA, [online], [searched on August 25, 2015] , Internet <URL: http: //www.allthingsdistributed.com/files/amazon-dynamo-sosp2007.pdf> Michio Irie and 4 others, "Load distribution method that is conscious of data replication in consistent hash method", The Institute of Electronics, Information and Communication Engineers, October 2010, IEICE Technical Report, IN2010-77, P.69- 74

  However, in the method of virtually giving a plurality of IDs to each node described in Non-Patent Document 3 (method using a virtual node), a plurality of virtual nodes are set for one node. There is a disadvantage that the number increases and search performance and processing performance deteriorate. Therefore, for example, in providing video services, there are cases where the use of virtual nodes is postponed.

  In the distributed system of the consistent hash method, the assigned area of each node is determined by a clockwise search in the ID space of the consistent hash. Therefore, when a specific node is removed for some reason, the assigned area is transferred to the next node in the clockwise direction of the removed node. As a result, an event in which the load is extremely concentrated on a specific node occurs, causing a problem that the load balance of each node is lost. This point will be described with reference to FIG.

  FIG. 12A shows the nodes and data arranged in the ID space of the consistent hash and the load of each node. In the consistent hash method, as shown in the lower diagram of FIG. 12A, the ID space is traced clockwise from the data located in the ID space (indicated by a circle “◯” in each diagram), and the first hit is performed. The node (indicated by a square “□” in each figure) is the node in charge of the data. In FIG. 12A, each of the nodes “A”, “B”, and “C” is responsible for processing three data, and the node “D” is responsible for processing two data. The load on each node in this case is shown in the upper part of FIG.

  Consider a case where node “C” is removed in the state of FIG. In this case, as shown in FIG. 12B, the data that was assigned to the removed node “C” is assigned to the node “D” that is the next node in the clockwise direction in the ID space. Become. As a result, the load of the node “D” becomes larger than that of other nodes (see the upper diagram of FIG. 12B). That is, the load balance of each node is lost.

  Moreover, even if it is going to cancel load balance by adding nodes, in the existing technology, for example, the position of the increasing node is determined at random, so the optimal arrangement of the nodes is not achieved, and the load balance is not canceled. A case is assumed.

  The present invention has been made in view of such a background, and the present invention is a node that can reduce the load bias to a specific node during the reduction / expansion operation of the nodes constituting the distributed processing system. It is another object of the present invention to provide a slot optimization method.

  In order to solve the above-described problem, the invention according to claim 1 is the node of the distributed processing system in which each of a plurality of nodes constituting the cluster distributes and processes data by a consistent hash method. A storage unit for storing distribution ID information in which a slot indicating a responsible area in the ID space is stored in association with each node ID of the node, and a node load measuring unit for measuring load information of the own node When a reduction or expansion occurs in a plurality of nodes constituting the cluster, a node to change the node ID is determined from the plurality of nodes, and the load information of the plurality of nodes is used. The node of the determined node at a position on the ID space that reduces the load bias of the plurality of nodes after the reduction or addition Change the D, and the node ID and the changed as nodes, characterized in that it comprises a slot optimization processing unit that stores the distribution ID information.

  The invention according to claim 6 is a slot optimization method for the node of a distributed processing system in which data is distributed and processed by a consistent hash method to each of a plurality of nodes constituting a cluster, Is provided with a storage unit that stores distribution ID information in which a slot indicating a responsible area in the ID space is stored in association with each of the node IDs of the plurality of nodes, and measures load information of its own node And when a plurality of nodes constituting the cluster is reduced or added, a node to change the node ID is determined from among the plurality of nodes, and the load information of the plurality of nodes is determined. Using the determined position on the ID space to reduce the load bias for the plurality of nodes after the reduction or expansion It changes the node ID of the over-de, and the node ID and the changed and slot optimization method characterized by performing the steps of: storing in the distribution ID information.

In this way, when a number of nodes constituting a cluster is reduced or added, a node whose node ID is to be changed is determined from the plurality of nodes, and the load is not biased. The node ID of the selected node can be changed.
Therefore, it is possible to reduce the load imbalance on the specific node when the node is reduced or added.

  According to a second aspect of the present invention, the slot optimization processing unit is a node of a next node positioned next in a clockwise direction to a node to be removed on the ID space when a specific node is to be removed. The node IDs of the next node positioned next to the next node in the clockwise direction from the node to be removed and the node ID of the previous node positioned one counterclockwise from the node to be removed 2. The node according to claim 1, wherein the node ID is calculated by equally dividing the total, and the distribution ID information is updated using the calculated node ID of the next node.

  As a result, when a specific node is removed, the node ID of the next node positioned next in the clockwise direction of the node to be removed in the ID space is appropriately set so that the load is not biased between adjacent nodes. Can be changed.

  According to a third aspect of the present invention, before the slot optimization processing unit is positioned one counterclockwise before the node to be removed on the ID space when the specific node is to be removed. The node ID of the node is the node ID of the next node positioned clockwise from the node to be removed, and the node ID of the previous node positioned two times counterclockwise from the node to be removed. And the distribution ID information is updated by using the calculated node ID of the previous node.

  As a result, when a specific node is removed, the load of the node ID of the previous node located one counterclockwise before the node to be removed in the ID space is not biased between adjacent nodes. Can be changed appropriately.

  According to a fourth aspect of the present invention, the slot optimization processing unit refers to the distribution ID information when a specific node is added, and the size of the slot is the largest among the plurality of nodes. Nodes are extracted and assigned so that nodes to be added are arranged at positions where the slots of the extracted nodes are divided, and node IDs of the nodes to be added are assigned to the extracted nodes whose slots are divided by the nodes to be added Is calculated by equally dividing the total of the node ID of the previous node and the node ID of the previous node positioned one counterclockwise from the extracted node, and using the calculated node ID of the additional node, The node according to claim 1, wherein the distribution ID information is updated.

  Thereby, when a specific node is added, the node ID in the ID space of the added node can be set appropriately so that the load is not biased between adjacent nodes.

  In the invention according to claim 5, the slot optimization processing unit selects a starting node so that the amount of data movement of each node is minimized when a specific node is reduced or added, Using the starting node as a reference, determine the node ID of each node so that the slot sizes of each node are equal, and update the distribution ID information using the determined node ID of each node The node according to claim 1 is provided.

As a result, when a specific node is removed or added, the node that is the starting point that minimizes the load on each node is selected, and the size of the slot of each node is determined based on the starting node. The node ID of each node can be determined to be equal.
Therefore, it is possible to prevent the load on each node from being biased when the number of nodes is reduced or increased.

  ADVANTAGE OF THE INVENTION According to this invention, the node and slot optimization method which reduces the bias | biasing of the load to a specific node at the time of the reduction | decrease / addition operation of the node which comprises a distributed processing system can be provided.

It is a figure which shows the whole structure of the distributed processing system containing the node which concerns on this embodiment. It is a figure for demonstrating the outline | summary of the load distribution process which the node which concerns on this embodiment performs. It is a functional block diagram of a node concerning this embodiment. It is a figure which shows the data structural example of the node identifier management table which concerns on this embodiment. It is a figure which shows the data structural example of the distribution ID table which concerns on this embodiment. It is a figure for demonstrating the load distribution process at the time of reduction of the method (1) Minimum Adjustment method. It is a figure for demonstrating the load distribution process at the time of expansion of the method (1) Minimum Adjustment method. It is a figure for demonstrating the origin node selection process at the time of reduction of the technique (2) Full Justification method. It is a figure for demonstrating the origin node selection process at the time of expansion of the method (2) Full Justification method. It is a flowchart which shows the flow of a process in case the node which concerns on this embodiment performs the method (1) Minimum Adjustment method. It is a flowchart which shows the flow of a process in case the node which concerns on this embodiment performs the method (2) Full Justification method. It is a figure for demonstrating the problem in a prior art.

<Overall configuration>
First, a distributed processing system 1000 including a node 1 according to a mode for carrying out the present invention (hereinafter referred to as “this embodiment”) will be described.
FIG. 1 is a diagram showing an overall configuration of a distributed processing system 1000 including a node 1 according to the present embodiment.

  This distributed processing system 1000 includes a plurality of nodes 1. Each node 1 is a physical device such as a computer or a logical device such as a virtual machine. The load balancer 3 distributes the message received from the client 2 by simple round robin or the like and transmits it to each node 1. Then, the distribution unit 12 of the node 1 distributes the message from the client 2 to the node 1 in charge of the message based on, for example, a consistent hash method. In the node 1 in charge of the message, the signal processing unit 13 performs signal processing and provides a service to the client 2.

  Note that the load balancer 3 does not exist, and a message can be transmitted from the client 2 to an arbitrary node 1 (distribution unit 12). Further, the distribution unit 12 and the signal processing unit 13 may exist on the same node 1 at the same time, or may exist on different nodes 1.

<Node>
Next, the node 1 constituting the distributed processing system 1000 will be specifically described.

≪Overview≫
First, an outline of processing of the node 1 according to the present embodiment will be described.
The node 1 executes any one of a method (1) Minimum Adjustment method and a method (2) Full Justification method as a method of load distribution processing (processing for reducing load imbalance).

  Method (1) The Minimum Adjustment method is a node to be reduced (hereinafter, may be referred to as “reduced node”) when the node 1 constituting the distributed processing system 1000 is reduced or added. The node ID is divided so that the assigned area is equally divided between the nodes adjacent to the nodes to be added (hereinafter sometimes referred to as “additional nodes”) (hereinafter also referred to as “adjacent nodes”). change.

Specifically, as shown in FIG. 2A, in the case where nodes “A” to “D” are arranged on the consistent hash ID space and each data processing is executed, for example, Assume that node “C” is removed. At this time, in the method (1) Minimum Adjustment method, the load deviation is reduced by changing the node ID of either the node “D” or the node “B” that is adjacent to the node “C”. To do. Here, description will be made assuming that the node ID of the node “D”, which is the next node on the ID space of the reduced node (node “C”), is changed. Method (1) In the Minimum Adjustment method, as shown in FIG. 2B, the node “D”, which is the next node in the ID space of the node “C”, which is the reduced node, and the next node ( The node ID of the node “D” is changed so as to equally divide the area in charge of the node “D” and the node “A” with respect to the node “A” which is the node one after another (here, the node “D”) Move counterclockwise.)
Thereby, in the ID space, it is possible to prevent the load of the next node (node “D”) from becoming extremely high in the clockwise direction of the reduced node (node “C”). In addition, since the node that needs to move the assigned data is only the node “D” adjacent to the reduced node (node “C”) among all the nodes, the amount of data movement can be reduced. It can cope with the reduction at high speed.

  Method (2) The Full Justification method is specified so that when the node 1 belonging to the distributed processing system 1000 is removed or when the node 1 is added, the assigned area of each node 1 is equally divided. The node IDs of all the nodes 1 other than the starting node are changed using the node (“starting node” described later) as a reference.

Specifically, as shown in FIG. 2A, when nodes “A” to “D” are arranged on the consistent hash ID space and each data is processed, for example, the node It is assumed that “C” has been removed. At this time, in the method (2) Full Justification method, as shown in FIG. 2 (c), the area in charge of each node 1 (node “B” in FIG. 2 (c)) is set as the origin node. The node IDs of the nodes 1 other than the starting node are changed (reset) so that the slot) is equally divided. Here, the node IDs of the nodes “A” and “D” are changed.
As a result, the load on each node 1 can be kept uniform without being biased.

  The node 1 according to the present embodiment reduces the number of nodes 1 constituting the distributed processing system 1000 by executing the load distribution process of any one of the method (1) Minimum Adjustment method and the method (2) Full Justification method. During the expansion operation, it is possible to reduce the load imbalance on a specific node.

<< Node configuration >>
Next, the node 1 constituting the distributed processing system 1000 according to the present embodiment will be specifically described. The node 1 according to the present embodiment is a privileged node that manages a node identifier management table 100 (see FIG. 4) and a distribution ID table 200 (see FIG. 5), which will be described later, among the plurality of nodes 1 of the distributed processing system 1000. And the case where the node identifier management table 100 and the distribution ID table 200 are received from the privileged node and the node identifier management table 100 and the distribution ID table 200 are updated. The processing performed by the privileged node will be described later.

As shown in FIG. 1, the node 1 is communicably connected to the load balancer 3 and is communicably connected to other nodes 1 other than itself constituting the cluster. Further, when the node 1 receives a message from the client 2 via the load balancer 3, the node 1 distributes the message to the responsible node 1 (including itself), and executes signal processing of the message. In addition, the node 1 that is a privileged node receives information on the reduction / expansion of the node 1 belonging to the distributed processing system 1000, and any one of the method (1) Minimum Adjustment method and the method (2) Full Justification method described above. By executing the above, the node ID of the target node 1 is changed in the ID space (specifically, the distribution ID table 200 described later is updated), thereby reducing the load bias.
As illustrated in FIG. 3, the node 1 includes a control unit 10, an input / output unit 20, and a storage unit 30.

  The input / output unit 20 inputs / outputs information to / from the load balancer 3 and other nodes 1 other than itself. Further, the input / output unit 20 is an input / output interface that performs input / output between a communication interface (not shown) that transmits / receives information via a communication line and an input unit such as a keyboard or an output unit such as a monitor. (Not shown).

  The storage unit 30 includes storage means such as a hard disk, a flash memory, and a RAM (Random Access Memory), and includes data 300 to be processed, a node identifier management table 100 (see FIG. 4), a distribution ID table (distribution ID information). ) 200 (see FIG. 5), node load measurement information 400, which is information obtained by measuring the node load of the node 1 itself, is stored. Further, when the node 1 is a privileged node, system load information 500, which is information obtained by collecting (collecting) node load information of each node 1, is stored. Details of each piece of information stored in the storage unit 30 will be described later.

  The control unit 10 controls the entire node 1 and includes a node identifier management unit 11, a distribution unit 12, a signal processing unit 13, a node load measurement unit 14, and a slot optimization processing unit 15. The control unit 10 is realized, for example, by a CPU (Central Processing Unit) (not shown) developing and executing a program stored in the storage unit 30 on a RAM (not shown).

The node identifier management unit 11 manages the node information (IP address and the like) of each node 1 configuring the cluster in the distributed processing system 1000 and the ID space handled by each node 1.
Specifically, the node identifier management unit 11 receives information from the outside when a node is detached (removed) or added (added) to the distributed processing system 1000 to which the node identifier belongs, and the distributed processing system The node identifier management table 100 in which the identification information and the like of the node 1 constituting 1000 is stored is updated.

FIG. 4 is a diagram illustrating a data configuration example of the node identifier management table 100 according to the present embodiment.
As shown in FIG. 4, the node identifier management table 100 stores the node identifier 101 and the address 102 (for example, IP address) of each node 1 constituting the distributed processing system 1000 in association with each other.

  The node identifier 101 is given by, for example, the node identifier management unit 11 of a specific node (for example, set in ascending order of the node identifier 101) set in advance in the distributed processing system 1000. Distributed to each node 1. The node identifier 101 is assigned to each virtual ID when a virtual ID is used in the consistent hash ID space.

  In addition, the node identifier management unit 11 outputs the received information about the deinstallation / expansion of the node 1 to the slot optimization processing unit 15 and the processing result of the slot optimization processing unit 15 (node ID change information, etc.) ), The node identifier management table 100 is updated (reflecting the reduction / expansion of the node 1), and the distribution ID table 200 is updated in order to change the ID space (area in charge: slot) in charge of the node 1. To do.

FIG. 5 is a diagram illustrating a data configuration example of a distribution ID table (distribution ID information) 200 according to the present embodiment.
As shown in FIG. 5, the distribution ID table 200 stores an ID space 202 (in charge area: slot) that the node 1 is in charge of in association with the node identifier 201. This node identifier 201 is the same information as the node identifier 101 of FIG. In the example shown in FIG. 5, the total number of IDs in the ID space is “800” of “0” to “799”. For example, the node 1 with the node identifier 201 “A” is “0” as the ID space 202 in charge. To 199 ”. In the distribution ID table 200, the node ID in the ID space of the node 1 (node “A”) with the node identifier 201 “A” is “199”, and similarly, the ID of the node “B” is the same. The node ID on the space is “399”, the node ID of the node “C” on the ID space is “599”, and the node ID of the node “D” on the ID space is “799”. . Then, the node identifier management unit 11 sorts the node IDs of the respective nodes 1 in ascending order in the distribution ID table 200 and manages them as a continuous ID space 202.
In the present embodiment, each ID is arranged clockwise in a closed ID space, and the node that first hits when the data is traced clockwise from the data ID will be described as the charge of the data. However, each ID may be arranged in a counterclockwise direction in the ID space, and when the data is traced counterclockwise from the ID of the data, the node that hits first may be assigned to the data. That is, an ID on the ID space can be set around a predetermined direction.

  The node identifier management unit 11 of the privileged node in the distributed processing system 1000 transmits the latest node identifier management table 100 and the distribution ID table 200 to each node 1. Accordingly, the node identifier management unit 11 of each node 1 always updates and holds the node identifier management table 100 and the distribution ID table 200 in the latest state. By doing so, each node 1 in the distributed processing system 1000 holds the same node identifier management table 100 and distribution ID table 200.

  The privileged nodes are set so as to become privileged nodes in order from the node 1 in the top row of the node identifier management table 100 (FIG. 4), for example. When node 1 newly becomes a privileged node, information indicating that it is a privileged node is transmitted to each node 1 or the like. Then, the privileged node updates its own distribution ID table 200 when there is an arrangement change in the ID space (node ID change or the like) for the node 1 in the cluster, and the update information is updated for each node. Deliver to 1.

Returning to FIG. 3, the distribution unit 12 calculates “hash (key)” based on information (“distribution key”) in the message received from the client 2 via the load balancer 3 or the like, and distributes the distribution ID table 200. To identify the node 1 in charge of processing the message. Then, the distribution unit 12 acquires information on the address of the identified node 1 with reference to the node identifier management table 100, and distributes (transmits) the message to the identified node 1.
In addition, the signal processing unit 13 performs signal processing of messages related to data handled by the node 1 itself.

For example, when a node load measurement request or the like is received from a privileged node in a predetermined cycle, the node load measurement unit 14 measures the node load (load information) of its own node 1 and stores it in its storage unit 30. Store as node load measurement information 400. Then, the node load measurement unit 14 transmits the measured node load (load information) to the privileged node.
The node load measured by the node load measuring unit 14 is, for example, a memory usage amount, a CPU usage rate, or the number of data.

The slot optimization processing unit 15 described below is provided in each node 1, but operates when the node 1 itself is a privileged node.
The slot optimization processing unit 15 of the node 1 changes the node ID in the ID space by executing any one of the method (1) Minimum Adjustment method and the method (2) Full Justification method, and reduces the load bias. Is realized. Note that an administrator of the distributed processing system 1000 or the like sets in advance which of the method (1) Minimum Adjustment method and the method (2) Full Justification method the slot optimization processing unit 15 executes.

(Method (1) Minimum Adjustment method)
Method (1) The Minimum Adjustment method is a method in which, when a node 1 constituting the distributed processing system 1000 is reduced or added, the assigned area is set between the reduced node and the node adjacent to the added node (adjacent node). The node ID is changed so as to be divided.

[Processing at the time of reduction]
A process executed by the slot optimization processing unit 15 at the time of node reduction by the method (1) Minimum Adjustment method will be described.
First, the slot optimization processing unit 15 sets the node 1 whose node ID is to be changed as the node 1 located on the clockwise side of the reduced node (hereinafter, may be referred to as “next node”) or the reduced number. Node 1 located on the counterclockwise side of the node (hereinafter, may be referred to as “previous node”) is determined. Here, the slot optimization processing unit 15 determines the area in charge for the node 1 (hereinafter sometimes referred to as “second node”) positioned second on the clockwise side of the reduced node and the previous node. Are compared (slot size), load size, or the number of data. Then, if the value of the next node is smaller, the slot optimization processing unit 15 determines the next node as the node 1 whose node ID is to be changed. If the value of the previous node is smaller, the slot optimization processing unit 15 sets the previous node as the node ID. The node 1 to be changed is determined.

  Specifically, the slot optimization processing unit 15 compares the size of the slot of the next node and the previous node for the reduced node when the size of the assigned area (slot size) is used as a reference, Node 1 with the smaller slot size is selected. When the load size is used as a reference, the slot optimization processing unit 15 compares the loads of the reduced node with the load of the next node and the previous node, and selects the node 1 on the side with the lower load. Further, when the number of data is based on the number of data, the slot optimization processing unit 15 compares the number of data of the next node and the previous node with respect to the reduced node, and determines the node 1 with the smaller number of data. select. By doing so, it is possible to determine the node 1 whose node ID is to be changed while avoiding an increase in the load bias between the nodes 1 as much as possible.

  The slot optimization processing unit 15 sets the node 1 whose node ID is to be changed as the node 1 (next node) positioned on the clockwise side of the reduced node or the node 1 (located on the counterclockwise side of the reduced node) ( If any of the previous nodes is determined, the node ID is calculated based on the following equation.

[If determined as the next node]
When the node 1 whose node ID is to be changed is determined as the next node, the slot optimization processing unit 15 calculates the node ID to be changed according to the following equation (1).

ID of the node (next node) located on the clockwise side of the reduced node =
(ID of the node (previous node) located on the counterclockwise side of the reduced node +
ID of the second node (next node) located on the clockwise side) ÷ 2 Equation (1)

In the example shown in FIG. 2B described above, the ID of the node (previous node: node “B”) located counterclockwise of the reduced node (node “C”) and the reduced node (node “C”) )) Is added to the ID of the second node (second node: node “A”) located on the clockwise side, and divided by 2 (equally divided), so that it is positioned on the clockwise side of the reduced node. ID of the node (next node: node “D”) to be calculated is calculated.
If the equally divided ID value is not divisible, for example, a value after the decimal point is discarded (the same applies hereinafter). Further, since the ID space of the consistent hash is a closed ID space (that is, a ring), the ID next to the set maximum ID value returns to “0”. Therefore, when the ID crosses “0”, Equation (1) is used in consideration thereof. The same applies to formulas (2), (3), and (5) described later.

  In this way, the slot optimization processing unit 15 calculates a new ID of the next node of the reduced node. Then, the slot optimization processing unit 15 outputs the calculated next node ID to the node identifier management unit 11 to update the distribution ID table 200. Thereby, it is possible to reduce the load bias to the node (next node) located on the clockwise side of the reduced node.

[When the previous node is determined]
When the node 1 whose node ID is to be changed is determined as the previous node, the slot optimization processing unit 15 calculates the node ID to be changed according to the following equation (2).

ID of the node (previous node) located on the counterclockwise side of the reduced node =
(ID of the second node (previous node) on the counterclockwise side of the reduced node +
ID of the node (next node) located in the clockwise direction) ÷ 2 Equation (2)

  In the example shown in FIG. 6, when the node “C” is removed on the ID space shown in FIG. 6A, as shown in FIG. 6B, the reduced node (node “C”) The ID of the second node (previous node: node “A”) located on the counterclockwise side and the node (next node: node “D” located on the clockwise side of the reduced node (node “C”) ")) And dividing by 2 (equal division), the ID of the node (previous node: node" B ") located on the counterclockwise side of the reduced node is calculated.

  In this way, the slot optimization processing unit 15 calculates a new ID of a node (previous node) located on the counterclockwise side of the reduced node. Then, the slot optimization processing unit 15 outputs the calculated ID of the previous node to the node identifier management unit 11 to update the distribution ID table 200. As a result, the ID of the previous node moves clockwise in the ID space. Therefore, it is possible to reduce the load bias to the node (next node) located on the clockwise side of the reduced node.

  In the above, the slot optimization processing unit 15 first sets the node 1 whose node ID is changed as the node 1 (next node) positioned on the clockwise side of the reduced node or the counterclockwise of the reduced node. It has been described that one of the nodes 1 (previous node) located on the rotation side is determined. However, the node 1 whose node ID is to be changed is set in the slot optimization processing unit 15 in advance as either the next node of the reduced node or the previous node of the reduced node, and Expression (1) or Expression (2) ) May be used to calculate the changed node ID.

[Processing for expansion]
A process executed by the slot optimization processing unit 15 when adding a node in the method (1) Minimum Adjustment method will be described. The slot optimization processing unit 15 compares the sizes of the slots of each node 1 and arranges a node to be added at a position where the largest slot is equally divided. This will be specifically described below.

  First, the slot optimization processing unit 15 refers to the distribution ID table 200 (FIG. 5), extracts the node 1 with the largest size (slot size) of the assigned area of each node 1 before the addition, Allocation is performed so that a newly added node 1 (additional node) is arranged at a position where the assigned area of the extracted node 1 is divided. Note that the node 1 into which the assigned area is divided may be hereinafter referred to as “divided node”.

  Subsequently, the slot optimization processing unit 15 calculates the node ID of the additional node by the following equation (3).

Expansion node ID =
(ID of the node (divided node) whose slot is divided by the expansion node +
ID of the node (previous node) located on the counterclockwise side of the split node) / 2
... Formula (3)

  Specifically, as shown in FIG. 7A, when nodes “A”, “B”, and “D” are arranged in the ID space and each data is processed, the node “C” is added. Assume that node 1 has received the instruction. Here, the slot optimization processing unit 15 refers to the distribution ID table 200 and extracts the node “D” as the node 1 having the largest assigned area size (slot size). In this case, the slot optimization processing unit 15 adds the node ID of the node “D” and the node ID of the node “B” by adding 2 (equal division) using Expression (3). Then, the node ID of the node “C” which is the additional node is calculated (see FIG. 7B).

  In this way, the slot optimization processing unit 15 calculates the node ID of the extension node. Then, the slot optimization processing unit 15 outputs the calculated node ID of the additional node to the node identifier management unit 11 to update the distribution ID table 200. As a result, when the node 1 is added, it is possible to avoid a load bias between the split node and the added node.

(Method (2) Full Justification method)
Next, method (2) Full Justification method will be described.
Method (2) The Full Justification method is specified so that when the node 1 belonging to the distributed processing system 1000 is removed or when the node 1 is added, the assigned area of each node 1 is equally divided. The node IDs of all the nodes 1 other than the node (“starting node” described later) are changed.

[Processing at the time of reduction]
The processing that the slot optimization processing unit 15 executes at the time of node reduction by the method (2) Full Justification method will be described.
The slot optimization processing unit 15 first selects a node having the smallest data movement amount as a starting node (starting node selection processing), and then the size of the assigned area of each node 1 with the selected starting node as a reference. The node ID of each node 1 other than the starting node is determined so that (slot size) is equally divided (node ID update processing).

FIG. 8 is a diagram for explaining the starting node selection process when the method (2) Full Justification method is removed. The slot optimization processing unit 15 selects the starting node so that the data movement amount of each node 1 is minimized.
Therefore, the slot optimization processing unit 15 calculates the average load (node average load μ) of each node 1 after the reduction. Then, as shown in Expression (4), the node “N” having the smallest Euclidean distance total d _N from the node average load μ is selected.

As shown in FIG. 8A, when four nodes 1 (nodes “A”, “B”, “C”, and “D”) are arranged in the ID space, the node “C” is removed. think of. In FIG. 8, in order to make the explanation easy to understand, the load of each node is indicated by data (circle “◯” in the figure), and the load of each data “◯” is assumed to be the same “1”.
First, the slot optimization processing unit 15 calculates the node average load μ after the node “C” is removed as μ = 1/12/3 = 4, where N (number of nodes) is “3”. And the number of data is “12”.

At this time, as shown in FIG. 8B, the node “A” is the starting point, and the size of the assigned area (slot size) of each node 1 (nodes “A”, “B”, and “D”) is equally divided. In this case, the total Euclidean distance d _A is summed in the order of the node “A”, the node “B”, and the node “D”,
d _A = | 3-4 | + | 5-4 | + | 4-4 | = 1 + 1 + 0 = 2
It becomes.

Further, as shown in FIG. 8C, the size (slot size) of the assigned area of each node 1 (nodes “A”, “B”, “D”) is equally divided starting from the node “B”. in the case, the Euclidean distance sum d _B sums in the order of node "a", the node "B", the node "D",
d _B = | 3-4 | + | 5-4 | + | 4-4 | = 1 + 1 + 0 = 2
It becomes.

Further, as shown in FIG. 8D, the size (slot size) of the assigned area of each node 1 (nodes “A”, “B”, and “D”) is equally divided starting from the node “D”. In this case, the Euclidean distance sum d _D is summed in the order of the node “A”, the node “B”, and the node “D”,
d _D = | 4-4 | + | 4-4 | + | 4-4 | = 0 + 0 + 0 = 0
It becomes.

As described above, in the case shown in FIG. 8, the slot optimization processing unit 15 uses the node “D” (d _D = 0) having the smallest total Euclidean distance d _N from the node average load μ (= 4) as the starting node. Choose as.

  Then, the slot optimization processing unit 15 sets the node ID of each node 1 to the following formula (5) so that the size of the assigned area (slot size) is equally divided with reference to the selected starting node. Calculate using.

Node ID = size of the entire ID space / N × k + node ID of the starting node Expression (5)
Here, N = number of nodes (after reduction), 0 ≦ k <N (k is a positive integer).

  In this way, when the node 1 is removed, the slot optimization processing unit 15 selects the starting node with the smallest amount of data movement, and the size of the area in charge of each node 1 is based on the starting node. The node IDs of the nodes 1 other than the starting node are calculated so that the (slot size) is equally divided. Then, the slot optimization processing unit 15 outputs the node ID of each node 1 other than the calculated starting node to the node identifier management unit 11 to update the distribution ID table 200. Thereby, it is possible to prevent the load on each node from being biased.

[Processing for expansion]
The processing executed by the slot optimization processing unit 15 when adding a node in the method (2) Full Justification method will be described.
The slot optimization processing unit 15 first selects a node with the smallest amount of data movement as a starting node (starting node selection processing), and then uses the selected starting node as a reference, similarly to the process at the time of removal. The node IDs of the nodes 1 other than the starting node are determined so that the size of the assigned area of each node 1 (slot size) is equally divided (node ID update processing).

FIG. 9 is a diagram for explaining a starting node selection process when the method (2) Full Justification method is added. The slot optimization processing unit 15 selects the starting node so that the data movement amount of each node 1 is minimized.
Therefore, the slot optimization processing unit 15 calculates the average load (node average load μ) of each node after expansion. Then, as shown in Expression (4), the node “N” having the smallest Euclidean distance total d _N from the node average load μ is selected.

  As shown in FIG. 9A, when three nodes 1 (nodes “A”, “B”, and “D”) are arranged in the ID space, a case where the node “C” is added is considered. In FIG. 9, for ease of explanation, the load of each node is indicated by data (circle “◯” in the figure), and the load of each data “◯” is assumed to be the same “1”.

First, the slot optimization processing unit 15 refers to the distribution ID table 200 (FIG. 5), extracts the node 1 with the largest size (slot size) of the assigned area of each node 1 before the addition, and extracts the node 1 The node 1 to be newly added is assigned to a position where the assigned area of the node 1 is divided. When the size of the assigned area of each node 1 (slot size) is the same, the position of the node 1 to be newly added is randomly assigned.
In FIG. 9A, a description will be given assuming that a new node “C” is added to the area in charge of the node “D”.

  Next, the slot optimization processing unit 15 calculates the node average load μ after adding the node “C”. Here, the node average load μ = 1/12/4 = 3. Here, N (number of nodes) is “4”, and the number of data is “12”.

At this time, as shown in FIG. 9B, the size (slot size) of the assigned area of each node 1 (nodes “A”, “B”, “C”, and “D”) starting from the node “A” , Euclidean distance total d _A is summed in the order of node “A”, node “B”, node “C”, and node “D”,
d _A = | 3-3 | + | 4-3 | + | 2-3 | + | 3-3 | = 0 = 0 + 1 + 1 + 0 = 2
It becomes.

Further, as shown in FIG. 9C, the size (slot size) of the assigned area of each node 1 (nodes “A”, “B”, “C”, and “D”) starts from the node “B”. in case of equally dividing the Euclidean distance sum d _B, the node "a", the node "B", the node "C", and the sum in the order of node "D",
d _B = | 3-3 | + | 3-3 | + | 3-3 | + | 3-3 | = 0 + 0 + 0 + 0 = 0
It becomes.

Also, as shown in FIG. 9D, the size (slot size) of the assigned area of each node 1 (nodes “A”, “B”, “C”, and “D”) starting from the node “D” is set. In the case of equal division, the Euclidean distance total d _D is summed in the order of the node “A”, the node “B”, the node “C”, and the node “D”,
d _D = | 3-3 | + | 4-3 | + | 3-3 | + | 2-3 | = 0 + 1 + 0 + 1 = 2
It becomes.

As described above, in the case shown in FIG. 9, the slot optimization processing unit 15 starts the node “B” (d _B = 0) having the smallest Euclidean distance total d _N from the node average load μ (= 3). Choose as.

  Then, the slot optimization processing unit 15 assigns the node ID of each node 1 to the above-described formula (5) so that the size of the assigned area (slot size) is equally divided with reference to the selected starting node. Calculate using. Here, N = the number of nodes (after expansion).

In this way, when adding the node 1, the slot optimization processing unit 15 selects the starting node with the smallest amount of data movement, and the size of the area in charge of each node 1 is based on the starting node. The node IDs of the nodes 1 other than the starting node are calculated so that (slot size) is equally divided. Then, the slot optimization processing unit 15 outputs the node ID of each node 1 other than the calculated starting node to the node identifier management unit 11 to update the distribution ID table 200. Thereby, it is possible to prevent the load on each node from being biased.
In the above description, the average load (node average load) of each node 1 is “μ”, and the Euclidean distance total d _N is calculated by Equation (4). However, the present invention is not limited to this, and the slot optimization processing unit 15 calculates the Euclidean distance total d _N using the average slot size of each node 1 and the average number of data of each node, and selects the starting node. Also good.

<Process flow>
Next, for the flow of load distribution processing (processing for reducing load imbalance) executed by the node 1 according to the present embodiment, a method (1) Minimum Adjustment method, a method ( 2) Each of the Full Justification methods will be described. Note that the load distribution process described below is a process performed by a privileged node among the nodes 1 constituting the distributed processing system 1000.

≪Method (1) Minimum Adjustment method≫
FIG. 10 is a flowchart illustrating a process flow when the node 1 according to the present embodiment executes the technique (1) Minimum Adjustment method as the load distribution process.

  First, the slot optimization processing unit 15 of the node 1 obtains load information (for example, memory usage, CPU usage rate, number of data, etc.) of each node from each node 1 constituting the distributed processing system 1000 according to a predetermined value. Receive at time intervals (step S1).

  Subsequently, the slot optimization processing unit 15 determines whether or not the node 1 constituting the distributed processing system 1000 has been removed (Step S2). Here, in the slot optimization processing unit 15, the node identifier management unit 11 of the node 1 receives an instruction to remove a specific node from the management apparatus of the distributed processing system 1000, or monitors whether each node 1 is active or not. By detecting that the node 1 has been removed, it is possible to recognize the removal of the node 1 (removed node). If the node 1 has not been removed (step S2 → No), the slot optimization processing unit 15 proceeds to step S7. On the other hand, when the node 1 is removed (step S2 → Yes), the process proceeds to the next step S3.

In step S3, the slot optimization processing unit 15 determines a node 1 whose node ID is to be changed among the nodes 1 adjacent to the reduced node.
Specifically, the slot optimization processing unit 15 includes a node 1 (second node) positioned second on the clockwise side of the removed node, and a node 1 (previous node) positioned on the counterclockwise side of the removed node. The node is compared with the size of the assigned area (slot size), the load size, or the number of data. Then, if the value of the next node is smaller, the slot optimization processing unit 15 determines the next node as the node 1 whose node ID is to be changed. If the value of the previous node is smaller, the slot optimization processing unit 15 sets the previous node as the node ID. The node 1 to be changed is determined.

  If the node 1 whose node ID is to be changed is determined as the next node (step S4 → Yes), the slot optimization processing unit 15 proceeds to the next step S5. On the other hand, when the node 1 whose node ID is to be changed is determined as the previous node (step S4 → No), the slot optimization processing unit 15 proceeds to the next step S6.

  When node 1 whose node ID is to be changed is determined as the next node (step S4 → Yes), in step S5, the slot optimization processing unit 15 calculates the node ID of the next node using the above-described equation (1). To do. Then, the process proceeds to step S10.

  On the other hand, when the node 1 whose node ID is to be changed is determined as the previous node (step S4 → No), in step S6, the slot optimization processing unit 15 uses the above-described equation (2) to determine the ID of the previous node. calculate. Then, the process proceeds to step S10.

  If it is determined in step S2 that the node 1 is not removed (step S2 → No), the slot optimization processing unit 15 determines whether or not to add the node 1 constituting the distributed processing system 1000. (Step S7). Here, the slot optimization processing unit 15 determines whether or not to add the node 1 because the node identifier management unit 11 of the node 1 has received an instruction to add the node 1 from the management apparatus of the distributed processing system 1000. Can be determined. If the slot optimization processing unit 15 determines that the node 1 is not added (no extension instruction has been received) (step S7 → No), the slot optimization processing unit 15 returns to step S1 and adds a new one from each node at predetermined time intervals. Receive load information and continue processing. On the other hand, if it is determined that the node 1 is to be added (step S7 → Yes), the slot optimization processing unit 15 proceeds to the next step S8.

  In step S8, the slot optimization processing unit 15 refers to the distribution ID table 200, compares the size of the assigned area (slot size) of each node 1 before expansion, and determines the node 1 with the largest assigned area. The node 1 to be newly added is assigned to the position where the extracted area in charge of the node 1 is divided. When the size of the assigned area of each node 1 (slot size) is the same, the position of the node 1 to be newly added is randomly assigned. Then, the process proceeds to next Step S9.

  In step S9, the slot optimization processing unit 15 calculates the node ID of the extension node using the above-described equation (3). Then, the process proceeds to step S10.

In step S10, the node identifier management unit 11 acquires the changed node ID or the node ID of the additional node calculated in step S5, step S6, or step S9 from the slot optimization processing unit 15, and assigns the distribution ID table. 200 and the node identifier management table 100 are updated.
Specifically, when the node identifier 1 is removed, the node identifier management unit 11 deletes the node identifier 201 of the removed node in the distribution ID table 200, and the next node or the previous node of the removed node. The node ID is changed (the area in charge of the ID space 202 is changed). Further, the node identifier management unit 11 deletes the record (node identifier 101 and address 102) of the reduced node from the node identifier management table 100.
On the other hand, when adding the node 1, the node identifier management unit 11 adds a record of the node 1 to be added to the distribution ID table 200 based on the node ID of the extension node. Further, the node identifier management unit 11 adds a record of the node 1 to be added in the node identifier management table 100.

  Subsequently, the node identifier management unit 11 transmits the updated distribution ID table 200 and node identifier management table 100 to each node 1 configuring the distributed processing system 1000 (step S11). Thereby, in each node 1, the distribution ID table 200 and the node identifier management table 100 are updated, and synchronization can be achieved as the entire system. Then, the node 1 ends the load distribution process.

≪Method (2) Full Justification method≫
FIG. 11 is a flowchart showing the flow of processing when the node 1 according to the present embodiment executes the technique (2) Full Justification method as load distribution processing (processing for reducing load bias).

  First, the slot optimization processing unit 15 of the node 1 obtains load information (for example, memory usage, CPU usage rate, number of data, etc.) of each node from each node 1 constituting the distributed processing system 1000 according to a predetermined value. Receive at time intervals (step S20).

  Subsequently, the slot optimization processing unit 15 determines whether or not the node 1 constituting the distributed processing system 1000 has been removed (step S21). If the slot optimization processing unit 15 determines that the node 1 constituting the distributed processing system 1000 has been removed (step S21 → Yes), the slot optimization processing unit 15 proceeds to step S24. On the other hand, if the slot optimization processing unit 15 does not determine that it has been removed (step S21 → No), the process proceeds to the next step S22.

  In step S <b> 22, the slot optimization processing unit 15 determines whether or not to add the node 1 configuring the distributed processing system 1000. Here, the slot optimization processing unit 15 determines whether or not to add the node 1 because the node identifier management unit 11 of the node 1 has received an instruction to add the node 1 from the management apparatus of the distributed processing system 1000. Can be determined. If the slot optimization processing unit 15 determines that the node 1 is not added (no extension instruction has been received) (step S22 → No), the slot optimization processing unit 15 returns to step S20 and starts a new operation from each node at a predetermined time interval. Receive load information and continue processing. On the other hand, if it is determined that the node 1 is to be added (step S22 → Yes), the slot optimization processing unit 15 proceeds to the next step S23.

  In step S23, the slot optimization processor 15 refers to the distribution ID table 200, compares the size of the assigned area (slot size) of each node 1 before expansion, and determines the node 1 with the largest assigned area. The node 1 to be newly added is assigned to the position where the extracted area in charge of the node 1 is divided. When the size of the assigned area of each node 1 (slot size) is the same, the position of the node 1 to be newly added is randomly assigned. Then, the process proceeds to next Step S24.

  In step S24, the slot optimization processing unit 15 calculates the node average load μ using the number of nodes after reduction (or addition) and the load information of each node 1.

Subsequently, the slot optimization processing unit 15 calculates the total Euclidean distance d _N with each node 1 as a starting point using the above-described equation (4) (step S25).
The slot optimization processing unit 15, among the Euclidean distance sum d _N for each node 1 calculated, the smallest node 1 that Euclidean distance sum d _N, is selected as starting node (step S26).

  Next, the slot optimization processing unit 15 sets the node ID of each node 1 to the above equation (5) so that the size of the assigned area (slot size) is equally divided with reference to the selected starting node. ) To calculate (step S27).

Subsequently, the node identifier management unit 11 acquires the changed node ID calculated in step S27 from the slot optimization processing unit 15, and updates the distribution ID table 200 and the node identifier management table 100 (step S28).
Specifically, in the distribution ID table 200, the node identifier management unit 11 deletes the record (node identifier 101 and address 102) of the reduced node and changes the node ID when deleting the node 1. For node 1, the area in charge of the ID space 202 is changed. Further, when adding node 1 in the distribution ID table 200, the node identifier management unit 11 adds a record of the node to be added. Further, in the node identifier management table 100, the node identifier management unit 11 deletes the record of the reduced node in the case of reduction, and adds the record of the extension node in the case of expansion.

  Subsequently, the node identifier management unit 11 transmits the updated distribution ID table 200 and node identifier management table 100 to each node 1 constituting the distributed processing system 1000 (step S29). Thereby, in each node 1, the distribution ID table 200 and the node identifier management table 100 are updated, and synchronization can be achieved as the entire system. Then, the node 1 ends the load distribution process.

  As described above, according to the node 1 and the slot optimization method according to the present embodiment, the node 1 serving as the privileged node of the distributed processing system 1000 can use the method (1) Minimum Adjustment method and method as a load distribution method. (2) Execute one of the full justification methods. As a result, it is possible to reduce the load imbalance on specific nodes during the reduction / expansion operation of the nodes 1 constituting the distributed processing system 1000.

  Note that the method (1) Minimum Adjustment method has a feature that the processing speed is high because the amount of data movement is small compared to the method (2) Full Justification method. Therefore, this is particularly effective when the frequency of the reduction / expansion of the nodes 1 constituting the distributed processing system 1000 is high. On the other hand, although the method (2) Full Justification method is slower than the method (1) Minimum Adjustment method because the amount of data movement is large, the processing is slow, but the load can be evenly distributed to specific nodes. Therefore, this is particularly effective when the frequency of the node 1 constituting the distributed processing system 1000 is low.

DESCRIPTION OF SYMBOLS 1 Node 2 Client 3 Load balancer 10 Control part 11 Node identifier management part 12 Distribution part 13 Signal processing part 14 Node load measurement part 15 Slot optimization processing part 20 Input / output part 30 Storage part 100 Node identifier management table 200 Distribution ID table ( (Distribution ID information)
300 Data 400 Node load measurement information 500 System load information 1000 Distributed processing system

Claims (6)

A node of a distributed processing system that distributes and processes data by a consistent hash method to each of a plurality of nodes constituting a cluster,
A storage unit for storing distribution ID information in which a slot indicating a responsible area in the ID space is stored in association with each of the node IDs of the plurality of nodes;
A node load measurement unit that measures load information of its own node;
When a reduction or expansion occurs in a plurality of nodes constituting the cluster, a node to change the node ID is determined from among the plurality of nodes, and the load information of the plurality of nodes is used, The node ID of the determined node is changed to a position on the ID space that reduces the load deviation of the plurality of nodes after being reduced or added, and the changed node ID is stored in the distribution ID information. A slot optimization processor;
A node characterized by comprising: The slot optimization processing unit includes:
When a specific node is removed, the node ID of the next node located next in the clockwise direction of the node to be removed in the ID space is set to the next next in the clockwise direction from the node to be removed. The total of the node ID of the next node that is located and the node ID of the previous node that is located one counterclockwise before the node to be removed is equally divided and calculated, and the calculated node ID of the next node The node according to claim 1, wherein the distribution ID information is updated using. The slot optimization processing unit includes:
When a specific node is removed, the node ID of the previous node located one counterclockwise before the node to be removed in the ID space is changed clockwise from the node to be removed. Is calculated by equally dividing the total of the node ID of the next node located at the node ID and the node ID of the previous node located two times counterclockwise from the reduced node. The node according to claim 1, wherein the distribution ID information is updated using a node ID. The slot optimization processing unit includes:
When a specific node is added, referring to the distribution ID information, the node having the largest slot size is extracted from the plurality of nodes, and the slot of the extracted node is divided into positions. Assign to arrange the nodes to be expanded,
The node ID of the node to be added is the node ID of the extracted node into which the slot is divided by the node to be added, and the node ID of the previous node located one counterclockwise from the extracted node. The node according to claim 1, wherein the node ID is calculated by equally dividing the total, and the distribution ID information is updated using the calculated node ID of the node to be added. The slot optimization processing unit includes:
When a specific node is removed or added, the starting node is selected so that the amount of data movement of each node is minimized, and the size of the slot of each node is based on the starting node. 2. The node according to claim 1, wherein a node ID of each node is determined so as to be equal to each other, and the distribution ID information is updated using the determined node ID of each node. A slot optimization method for the node of the distributed processing system in which each of a plurality of nodes constituting a cluster distributes and processes data by a consistent hash method,
The node is
A storage unit that stores distribution ID information in which a slot indicating a responsible area on the ID space is stored in association with each node ID of the plurality of nodes;
Measuring the load information of its own node;
When a reduction or expansion occurs in a plurality of nodes constituting the cluster, a node to change the node ID is determined from among the plurality of nodes, and the load information of the plurality of nodes is used, The node ID of the determined node is changed to a position on the ID space that reduces the load deviation of the plurality of nodes after being reduced or added, and the changed node ID is stored in the distribution ID information. Steps,
The slot optimization method characterized by performing.

Images