JP2014096113A - Load balancer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform load balancing with the throughput and memory usage of a server taken into consideration.SOLUTION: A domain classification unit calculates hash values of key information indicative of processing contents of execution requests for processing which may be received from a client, and classifies the execution requests for processing into a prescribed number of domains on the basis of the calculated hash values. A request processing server assignment unit divides each of the domains into sub-domains in which request frequencies of execution requests are equal to or lower than a prescribed threshold, and assigns sub-domains of each domain to a plurality of request processing servers in such a manner that the number of sub-domains assigned to each of the request processing server is equal to or smaller than a prescribed number. When receiving an execution request for processing from the client, a request processing server determination unit calculates a hash value indicative of processing contents of the execution request and specifies a domain to which the received execution request is classified, on the basis of the calculated hash value and determines request processing servers to which sub-divided domains of the specified domain are assigned, as transfer destinations of the execution request.

Description

  The present invention relates to a load distribution apparatus.

  Conventionally, in a server system that processes requests from clients, when a large amount of requests cannot be processed by a single server, a load balancer is installed between the client and the server system, and the load balancer sends requests to multiple servers. It is widely done to distribute. Here, several methods are known as load distribution methods in the load distribution apparatus. For example, the round robin method widely used for load distribution of a DNS server is the simplest load distribution method and distributes requests from clients to servers in order. Here, since an equal number of requests can be allocated to each server in the round robin method, if the processing load per request is approximately the same, the processing load of each server can be expected to be approximately equal.

  In addition, for example, in a key-value store that is a kind of data management system, a load distribution method called a consistent hashing method (Consistent Hashing) may be used (for example, Non-Patent Document 1). In the load distribution method using the consistent hash method, the load distribution device calculates the hash value of the key information included in the request, and sends the request based on the correspondence relationship between the server and the range of the hash value stored in advance. Sort to the server corresponding to the hash value.

  The consistent hash method is a method in which an approximately equal number of key information is assigned to each server due to the randomness of the hash function. Unlike the round robin method, the consistent hash method is different from the key information included in the request and the server to which the request is distributed. Correspondence is fixed. Therefore, when each server processes the request, in a situation where data corresponding to the key information included in the request (for example, value in the key value store) is required, each server holds all the key information data. There is no need to store only key information data distributed to the own server.

  This characteristic is useful when the data size for all key information is large. For example, when the server needs to store data corresponding to key information in memory in order to process requests at high speed. The round robin method, in which each server needs to store all key information data in memory, may not be applicable due to memory capacity limitations. Even in such cases, the consistent hash method can be applied. It is.

  On the other hand, the consistent hash method has a problem that the processing load of the server is not uniform when the request frequency for each key information is largely biased. For example, if there is key information whose request frequency is very high compared to the average key information request frequency, and the key information request frequency exceeds the processing performance of a single server, the consistent hash method is applied in the first place. Can not. Even if the processing performance of a single server is not exceeded, there is a large bias in the processing load between servers that are assigned key information with a very high request frequency and servers that are not so requested. Processing performance cannot be used effectively.

  In order to solve such a server processing load bias in the consistent hash method, a technology called a virtual node is expanded and the number of key information allocated to each server is adjusted, thereby reducing the server processing load bias. A mitigation method has been proposed (for example, Patent Document 1).

JP 2012-48424 A

D. Karger, E. Lehman, T. Leighton, M. Levine, D. Lewin, and R. Panigrahy, "Consistent Hashing and Random Trees: Distributed Caching Protocols for Relieving Hot Spots on the World Wide Web," in Proceedings of the 29th ACM Symposium on Theory of Computing (STOC'97), pp.654-663, May 1997.

  However, with the above-described conventional technology, it is difficult to perform load distribution in consideration of server processing volume and memory usage. For example, in the above-described prior art, if the number of key information allocated to each server differs for each server, even if the processing load bias is reduced, the data size held by the server is biased. Therefore, in the above-described prior art, the memory usage amount of each server is biased, and it is difficult to perform load distribution in consideration of the server processing amount and the memory usage amount. As a result, with the above-described conventional technology, the memory usage cannot be effectively utilized as the entire system.

  Accordingly, the technology according to the present application has been made in view of the above-described problems of the prior art, and provides a load distribution device that can perform load distribution in consideration of server processing amount and memory usage. For the purpose.

  In order to solve the above-described problems and achieve the object, the apparatus of the present application is a load distribution apparatus that transfers a process execution request received from a client apparatus to one of a plurality of request processing servers, and includes a classification unit And an assigning unit and a determining unit. The classifying unit calculates a hash value of key information indicating processing contents of a processing execution request that will be received from the client device, and manages a predetermined number of management requests based on the calculated hash value. Classify into units. The allocation unit divides each of the management units into partial management units in which the request frequency of the execution request is equal to or less than a predetermined threshold, and allocates the partial management unit in each management unit to each of the plurality of request processing servers. The number is assigned so that it is less than or equal to the predetermined number. When the determination unit receives a process execution request from the client device, the determination unit calculates a hash value indicating the processing content of the execution request, and manages the received execution request classified based on the calculated hash value A unit is specified, and a request processing server to which a partial management unit of the specified management unit is assigned is determined as a transfer destination of the execution request.

  The apparatus of the present application makes it possible to perform load distribution in consideration of the server processing amount and memory usage.

FIG. 1 is a diagram illustrating an example of a configuration of an information processing system according to the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of the load distribution apparatus according to the first embodiment. FIG. 3 is a diagram illustrating an example of information stored by the domain start hash value table according to the first embodiment. FIG. 4 is a diagram illustrating an example of information stored by the domain request frequency table according to the first embodiment. FIG. 5 is a diagram illustrating an example of information stored by the request processing server allocation table according to the first embodiment. FIG. 6 is a diagram illustrating an example of information stored by the request processing server allocation state management table according to the first embodiment. FIG. 7 is a diagram for explaining an example of domain division processing by the request processing server allocation unit according to the first embodiment. FIG. 8 is a diagram for explaining an example of allocation processing by the request processing server allocation unit according to the first embodiment. FIG. 9 is a sequence diagram for explaining a procedure of split domain assignment processing in the information processing system according to the first embodiment. FIG. 10 is a flowchart for explaining a procedure of request processing server allocation processing by the load distribution apparatus according to the first embodiment. FIG. 11 is a flowchart for explaining a procedure of request processing server allocation processing by the load balancer according to the first embodiment. FIG. 12 is a sequence diagram for explaining the procedure of the service request response process in the information processing system according to the first embodiment. FIG. 13 is a flowchart for explaining a procedure of request processing server determination processing by the load distribution apparatus according to the first embodiment. FIG. 14 is a flowchart for explaining a procedure of service response creation processing by the request processing server according to the first embodiment. FIG. 15 is a sequence diagram for explaining the procedure of allocation change processing when the number of domain divisions increases in the information processing system according to the first embodiment. FIG. 16 is a flowchart for explaining a procedure of request processing server allocation change processing when the number of domain divisions by the load distribution apparatus according to the first embodiment increases. FIG. 17 is a sequence diagram for explaining the procedure of the allocation change process when the number of domain divisions decreases in the information processing system according to the first embodiment. FIG. 18 is a flowchart for explaining the procedure of a request processing server allocation change process when the number of domain divisions by the load distribution apparatus according to the first embodiment decreases.

  Hereinafter, an embodiment of a load distribution apparatus according to the present application will be described in detail with reference to the accompanying drawings. In addition, the load distribution apparatus which concerns on this application is not limited by the following embodiment.

[First Embodiment]
[Configuration of Information Processing System According to First Embodiment]
In the following, first, the overall configuration of the information processing system including the load distribution apparatus according to the first embodiment will be described, and then the detailed configuration of the load distribution apparatus will be described. FIG. 1 is a diagram illustrating an example of a configuration of an information processing system 1 according to the first embodiment.

  As illustrated in FIG. 1, the information processing system 1 according to the first embodiment includes clients 11 to 13, a load distribution device 30, request processing servers 41 to 45, and a storage device 51. The clients 11 to 13 and the load balancer 30 are connected by the IP network 21, and the load balancer 30 and the request processing servers 41 to 45 are connected by the IP network 22, and the request processing servers 41 to 45 and the storage are connected. The device 51 is connected via the IP network 23. The IP networks 21 to 23 are IP (Internet Protocol) networks respectively configured by providers that provide Internet services.

  Here, only three clients 11 to 13 are shown, but actually, a larger number of clients are connected to the IP network 21. Further, only five request processing servers 41 to 45 are shown in the figure, but actually, a larger number of request processing servers are connected to the IP network 22 and the IP network 23. Since a plurality of clients and a plurality of request processing servers have the same function, in the following description, a plurality of clients are collectively referred to as a client 10, and a plurality of request processing servers are collectively referred to as a request processing server 40. explain.

  The client 10 generates a service request including key information (service specification value) necessary for execution of processing to be executed by any of the request processing servers 40. Then, the client 10 issues the generated service request to the load balancer 30. Thereafter, the client 10 acquires the execution result of the processing by any of the request processing servers 40 via the load balancer 30.

  The request processing server 40 has a memory that can be accessed at high speed, and is allocated to itself based on the allocation domain information notification transmitted from the load balancer 30 among all the control data held in the storage device 51. The control data (attribute value information) necessary for executing the service request processing of the domain is acquired from the storage device 51 and stored in the memory. When the request processing server 40 receives a service request from the load balancer 30, the request processing server 40 searches the memory for control data associated with the key information included in the received service request, and retrieves the searched control data. The response is transmitted to the load balancer 30 as a response. The allocation domain information notification will be described later in detail.

  The storage device 51 holds control data (attribute value information) corresponding to all key information (service specified values) on the disk, and stores the control data (attribute value information) in the domain included in the attribute value information request transmitted from the request processing server 40. All key information and attribute value information pairs belonging to the disk are acquired from the disk and transmitted to the request processing server 40 as an attribute value information response. Further, the storage apparatus 51 calculates the key information belonging to the domain from the domain included in the attribute value information request, so that the content of the domain start hash value table is notified from the load balancer 30, and the same contents as the table are retained. To do. The domain start hash value table will be described in detail later.

  The load balancer 30 determines a request processing server 40 that processes the service request based on key information (service designation value) included in the service request transmitted from the client, and sends the service request to the request processing server 40. Send. When receiving a service response corresponding to the service request from the request processing server 40, the load distribution apparatus 30 transmits the service response to the client. The load balancer 30 holds information on unused request processing servers 40 as server pool information in preparation for an increase in the service request amount. When the service request amount increases, By acquiring information of the unused request processing server 40 and transmitting an allocation domain information notification to the request processing server 40, processing of the service request can be started. On the contrary, the request processing server 40 in which the service request amount is reduced and the allocated domain is lost is recorded in the server pool information. In FIG. 1, the load balancer 30 is one, but a plurality of load balancers 30 can be installed in parallel for redundancy. In this case, the configuration of the load balancer 30 is changed. Absent. Hereinafter, the details of the load balancer 30 will be described with reference to FIG.

  FIG. 2 is a diagram illustrating an example of the configuration of the load distribution apparatus 30 according to the first embodiment. For example, as illustrated in FIG. 2, the load distribution apparatus 30 includes a communication control I / F unit 31, an input unit 32, a display unit 33, a storage unit 34, and a control unit 35.

  The communication control I / F unit 31 controls communication related to various information exchanged between the client 10 and the control unit 35. Further, the communication control I / F unit 31 controls communication related to various information exchanged between the request processing server 40 and the control unit 35. For example, the communication control I / F unit 31 controls transmission / reception of service requests and service responses. The communication control I / F unit 31 controls the exchange of various information between the input unit 32 and the display unit 33 and the control unit 35.

  The input unit 32 is, for example, a keyboard or a mouse, and accepts various information input processes by the user. For example, the input unit 32 receives threshold value input processing used for domain classification. Note that thresholds used for domain classification will be described later. The display unit 33 is, for example, a display and displays and outputs the processing result to the user.

  As shown in FIG. 2, the storage unit 34 includes a domain start hash value table 34a, a domain request frequency table 34b, a request processing server allocation table 34c, and a request processing server allocation state management table 34d. The storage unit 34 is, for example, a storage device such as a hard disk or an optical disk, or a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, and stores various programs executed by the load balancer 30. Remember.

  The domain start hash value table 34a stores a start hash value indicating a management unit (domain) range for classifying service requests received from the client 10 based on key information. Specifically, the domain start hash value table 34a stores information for identifying domains classified according to the hash value calculated from the key information of the service request by the control unit 35 described later. FIG. 3 is a diagram illustrating an example of information stored by the domain start hash value table 34a according to the first embodiment.

  For example, the domain start hash value table 34a stores information associating the start hash value with the domain, as shown in FIG. Here, the “start hash value” in FIG. 3 means a hash value that is the start of the range of each domain. The “domain” shown in FIG. 3 means a management unit for classifying and managing service requests received from clients by hash values of key information.

  For example, as shown in FIG. 3, the domain start hash value table 34a includes “start hash value: 0013, domain: domain D”, “start hash value: 1156, domain: domain A”, “start hash value”. : 2632, domain: domain H "and the like. These pieces of information mean, for example, that a service request corresponding to key information having hash values “0013” to “1155” belongs to “domain D”. Similarly, for example, it means that a service request corresponding to key information having hash values “1156” to “2631” belongs to “domain A”. That is, the domain start hash value table 34a stores a range of hash values for each domain.

  Returning to FIG. 2, the domain request frequency table 34 b stores the request frequency of service requests for each domain and the number of domain divisions according to the request frequency. FIG. 4 is a diagram illustrating an example of information stored in the domain request frequency table 34b according to the first embodiment. For example, as shown in FIG. 4, the domain request frequency table 34b stores information in which the request frequency (/ second) is associated with the number of domain divisions for each domain.

  For example, as shown in FIG. 4, the domain request frequency table 34b stores information in which “domain A” is associated with “request frequency (/ second): 280” and “domain division number: 1”. To do. This information means that service requests belonging to “domain A” are received 280 times per second, and the number of domain divisions is “1”. Details of the number of domain divisions will be described later. Similarly, the domain request frequency table 34b stores request frequencies from domain B to domain H and the number of domain divisions.

  Returning to FIG. 2, the request processing server allocation table 34c stores information on the request processing server 40 that is a transfer destination of service requests belonging to each domain. Specifically, the request processing server allocation table 34c stores information on the number of request processing servers that execute service requests determined for each domain by the control unit 35, which will be described later, and transfer destination information. FIG. 5 is a diagram illustrating an example of information stored by the request processing server allocation table 34c according to the first embodiment.

  For example, the request processing server allocation table 34c stores information in which a request processing server is associated with a domain, as shown in FIG. As an example, information such as “Domain A: Server 41”, “Domain B: Server 42, Server 43, Server 44, Server 45”, “Domain C: Server 46, Server 41” is stored. In these pieces of information, for example, the transfer destination of a service request belonging to “Domain A” is “Request Processing Server 41”, and the transfer destination of a service request belonging to “Domain B” is “Request Processing Server 42”, “Request Processing”. Server 43 ”,“ request processing server 44 ”, or“ request processing server 45 ”, and the transfer destination of the service request belonging to“ domain C ”is“ request processing server 46 ”or“ request processing server 41 ”. Means that. Similarly, the request processing server allocation table 34c stores information on the transfer destination request processing server 40 to which the service request is transferred for all domains.

  Returning to FIG. 2, the request processing server allocation state management table 34 d stores information on the domain allocated as the transfer destination for each request processing server 40. FIG. 6 is a diagram illustrating an example of information stored in the request processing server allocation state management table 34d according to the first embodiment. For example, as shown in FIG. 6, the request processing server allocation state management table 34d stores information in which an allocation domain is associated with a request processing server.

  For example, as shown in FIG. 6, the request processing server allocation state management table 34d includes “server 41: domain A, domain C, domain F” and “server 42: domain B, domain D, domain G”. Such information is memorized. That is, these pieces of information include that “request processing server 41” is a transfer destination of service requests belonging to “domain A”, “domain C”, and “domain F”, and “request processing server 42” is “domain B”. This means that it is a transfer destination of service requests belonging to “Domain D” and “Domain G”. Similarly, the request processing server allocation state management table 34d stores information on domains allocated as transfer destinations for other request processing servers.

  Returning to FIG. 2, the control unit 35 includes a domain classification unit 35a, a request processing server allocation unit 35b, an allocation domain information notification unit 35c, a request processing server determination unit 35d, and a request frequency update unit 35e. The control unit 35 is, for example, an electronic circuit such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit), an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array), and a load distribution device. 30 overall control is executed.

  The domain classification unit 35a classifies the execution request for processing received from the client 10 into a predetermined number of management units (domains) according to the hash value of the key information (service specification value) indicating the processing content of the execution request. Specifically, the domain classification unit 35a divides the range of values that can be taken by the hash value of the key information included in the service request into a predetermined number of domains, and stores the information in the domain start hash value table 34a.

  For example, the domain classification unit 35a calculates the random hash value as the start hash value of each domain, assuming that the number of domains is “8”, the range of hash values is “0000” to “9999”. And the domain classification | category part 35a stores the information as shown in FIG. 3, for example in the domain start hash value table 34a.

  The request processing server allocation unit 35b divides, for each domain, a divided domain in which the request frequency of execution requests is a predetermined threshold value or less, and divides the divided domains in each domain into a plurality of request processing servers 40 with a predetermined allocation number or less. Assign each. Specifically, the request processing server allocation unit 35b divides the domain into divided domains using a value obtained by dividing the processing performance of the request processing server by a predetermined allocation number as a predetermined threshold. FIG. 7 is a diagram for explaining an example of domain division processing by the request processing server allocation unit 35b according to the first embodiment.

  Here, FIG. 7 shows processing when the service request is classified into eight domains by the domain classification unit 35a as shown in FIG. For example, as illustrated in FIG. 7, the request processing server allocation unit 35b acquires the request frequency of each domain, and divides each domain into divided domains according to the acquired request frequency. Here, when the request frequency for each key information (service specification value) is known, the request processing server allocation unit 35b refers to the information stored in the domain start hash value table 34a, and Calculate the request frequency.

  On the other hand, when the request frequency for each key information (service specification value) is not known, the request processing server allocation unit 35b calculates a value obtained by dividing the total service request amount received from the client 10 by the number of domains. Set as a provisional value for the request frequency. In such a case, the request processing server allocation unit 35b updates the provisional value to the actual measurement value according to the actual request frequency for each domain that is found in the subsequent processing.

  As described above, the request processing server allocation unit 35b acquires the request frequency of each domain and divides it into divided domains according to the request frequency. For example, as shown in FIG. 7, the request processing server allocation unit 35b divides the processing performance “900 (/ second)” of the request processing server 40 by a predetermined allocation number “3” and a value “300”. And Then, the request processing server allocation unit 35b divides the request frequency of each domain by a predetermined threshold value “300”, and determines a value obtained by rounding up the value after the decimal point as an integer as the domain division number.

  The request processing server allocating unit 35b stores the acquired request frequency of each domain and the determined number of domain divisions in the domain request frequency table 34b, and, as shown in the right diagram of FIG. Divide by the number of divisions. For example, the request processing server allocation unit 35b uses “4” obtained by rounding up the decimal point of a value “3.5” obtained by dividing “request frequency: 1050” of “domain B” by a predetermined threshold “300”. “Domain B” is divided into “split domain B1”, “split domain B2”, “split domain B3”, and “split domain B4”. Similarly, the request processing server allocation unit 35b divides each domain into divided domains by the number of domain divisions.

  Further, the request processing server allocation unit 35b determines the number of request processing servers 40 to which the divided domains are allocated by dividing the total number of the divided domains by a predetermined allocation number, and assigns the number of request processing servers 40 to the determined number of request processing servers 40. Assign. Here, the request processing server allocation unit 35b allocates a divided domain in which the number of allocated divided domains is the smallest among the plurality of request processing servers 40 and the divided source domain is the same as the scheduled divided domain. The request processing server 40 having the smallest number is determined as the allocation destination of the divided domain to be allocated.

  FIG. 8 is a diagram for explaining an example of allocation processing by the request processing server allocation unit 35b according to the first embodiment. Here, FIG. 8 shows processing after the division processing shown in FIG. For example, as shown in the left diagram of FIG. 8, the request processing server allocation unit 35b divides the number “18” of the divided domains “A1” to “H3” by the predetermined allocation number “3” “6”. Is the number of request processing servers 40. Then, the request processing server allocation unit 35b determines the six servers of the request processing servers 41 to 46 as allocation target servers, and allocates a divided domain to each request processing server.

  At this time, the request processing server allocation unit 35b allocates the divided domain by the above-described method. For example, when assigning the divided domains B1 to B4 after assigning the divided domain A1 to the request processing server 41, the request processing server assigning unit 35b first extracts the request processing servers 42 to 46 having the smallest number of assigned divided domains. (The request processing server 41 is not the minimum because the divided domain A1 is allocated). Then, the request processing server allocation unit 35b allocates the request processing servers in order from the divided domain B1. Here, the request processing server allocation unit 35b has the smallest allocation number of the divided domains, and the request processing server 40 having the smallest allocation number of the divided domains in which the split source domain is the same as the scheduled split domain. Is determined as the assignment destination of the split domain to be assigned.

  That is, if the request processing server allocation unit 35b has four request processing servers (from the servers 41 to 44) in the right side of FIG. 8, the request processing server allocation unit 35b divides the divided domain B4 with the same dividing source. Control is performed so as to assign to the server 41 instead of the servers 42 to 44 to which the domains B1 to B3 are assigned. Similarly, the request processing server allocation unit 35b sequentially allocates the divided domains in the domains C to H to the request processing servers 41 to 46, as shown in the diagram on the right side of FIG. The request processing server allocation unit 35b stores the allocated information in the request processing server allocation table 34c and the request processing server allocation state management table 34d.

  In the load balancer 30 according to the first embodiment, as described above, the service requests are classified into a plurality of domains based on the hash value of the key information (service specification value), thereby eliminating the randomness of the hash function. The number of pieces of key information in each domain can be made approximately equal, and the memory usage can be made approximately equal. Further, the load balancer 30 divides the request frequency of each domain by using 1 / N of the processing performance of the request processing server 40 as a threshold, and assigns N to each server. Can be even.

  Further, when the request frequency of each domain is updated by a request frequency update unit 35e described later, the request processing server allocation unit 35b re-divides the domain into divided domains in which the updated request frequency is equal to or less than a predetermined threshold. The allocation to the plurality of request processing servers is changed according to the increase / decrease of the divided domains after the re-division. This process will be described in detail later.

  Returning to FIG. 2, when the request processing server allocation state management table 34d is updated by the request processing server allocation unit 35b, the allocation domain information notification unit 35c responds to the request processing server 40 whose allocation of the divided domains has been changed. The assigned domain information is notified. For example, the allocation domain information notifying unit 35c refers to the information in the request processing server allocation state management table 34d shown in FIG. 6 and the domain A, domain C, and domain F are allocated to the request processing server 41. Notify that. That is, the assigned domain information notification unit 35c notifies the request processing server 41 to execute a response to the service requests of the domain A, the domain C, and the domain F.

  When the request processing server determination unit 35d receives a processing execution request from the client 10, the request processing server determination unit 35d calculates a hash value indicating the processing content of the execution request, and a divided domain of the domain corresponding to the calculated hash value is assigned. The request processing server 40 is determined as the transfer destination of the execution request. Specifically, the request processing server determination unit 35d executes an execution request belonging to a domain to each request processing server 40 when divided domains divided from the same domain are assigned to different request processing servers 40, respectively. The service request is transferred so that the number of times of transfer is substantially equal.

  For example, when the request processing server determination unit 35d receives a plurality of service requests belonging to the domain B, the request processing server determination unit 35d determines the transfer destination so that the number of transfer times of the service requests belonging to the domain B to the request processing servers 42 to 45 becomes equal. And transfer the service request.

  The request frequency update unit 35e monitors the request frequency of execution requests for each domain and updates the request frequency. Specifically, the request frequency update unit 35e counts the frequency at which the service request is transmitted from the client 10 for each domain classified by the domain classification unit 35a, and periodically updates the domain request frequency table. The request frequency update unit 35e notifies the request processing server allocation unit 35b to change the allocation of the request processing server 40 when the request frequency increases or decreases beyond a predetermined threshold.

  For example, when the number of domain divisions for each domain stored in the domain request frequency table 34b increases or decreases, the request frequency update unit 35e changes the allocation of the increased or decreased divided domains.

  That is, when the number of divided domains increases, the request processing server allocation unit 35b allocates the increased number of divided domains to the request processing server 40 having the smallest number of allocated divided domains with the same source domain. On the other hand, when the number of divided domains decreases, the request processing server allocation unit 35b cancels the allocation of the reduced divided domains from the request processing server 40 having the largest number of allocation of the divided domains with the same source domain. .

[Procedure of processing by the load balancer according to the first embodiment]
Next, a processing procedure by the load distribution apparatus 30 according to the first embodiment will be described with reference to FIGS. In the following, each procedure of divided domain allocation processing, service request response processing, and allocation change processing will be described in sequence using the sequence diagram in the information processing system 1 and the flowchart in the load distribution apparatus 30.

  First, split domain assignment processing will be described with reference to FIGS. FIG. 9 is a sequence diagram for explaining the procedure of split domain assignment processing in the information processing system 1 according to the first embodiment. As shown in FIG. 9, in the load balancer 30, the domain classification unit 35a executes domain initialization processing (step S101), and stores the start hash value of each domain in the domain start hash value table 34a.

  When the domain start hash value table 34a is stored, the load distribution apparatus 30 notifies the storage apparatus 51 of the domain start hash value (step S102). When the storage device 51 receives the notification of the domain start hash value, the storage device 51 stores the content in its own memory and transmits a domain start hash value response to the load balancer 30 (step S103).

  Subsequently, in the load balancer 30, the request processing server allocation unit 35b executes a request frequency initialization process (step S104), and stores the request frequency and the number of domain divisions of each domain in the domain request frequency table 34b. . Then, the request processing server allocation unit 35b executes a request processing server allocation initialization process (step S105), determines the number of request processing servers 40, and allocates a divided domain to each request processing server 40.

When the request processing server allocation unit 35b allocates a divided domain to each request processing server 40, the allocation domain information notification unit 35c refers to the information stored in the request processing server allocation state management table 34d and refers to each request processing server. The allocation domain information is notified to 40 (step S106). When the allocation domain information is notified, the request processing server 40 transmits an attribute value information request for acquiring the attribute value information of the allocated domain to the storage apparatus 51 (step S107). The storage apparatus 51 transmits an attribute value information response corresponding to the received attribute value information request to each request processing server 40 (step S108).

  When receiving the attribute value information response from the storage device 51, the request processing server 40 transmits an allocation domain information response, which is a response to the notification of the allocation domain information, to the load balancer 30 (step S109).

  Next, details of the request processing server allocation initialization process shown in step S105 of FIG. 9 will be described. FIG. 10 and FIG. 11 are flowcharts for explaining the procedure of the request processing server allocation process by the load balancer 30 according to the first embodiment. Here, the process shown in FIG. 11 shows the details of the process of step S202 of FIG.

  As shown in FIG. 10, in the load balancer 30, when the domain request frequency and the number of domain divisions are stored in the domain request frequency table 34b, the request processing server allocation unit 35b adds the total number of divided domains of all domains, The required number of request processing servers is determined from the number of server divisions (predetermined allocation number), the determined number of request processing servers 40 are acquired from the server pool information, and each request processing is stored in the request processing server allocation state management table 34d. A record of the server 40 is added (step S201).

  Subsequently, the request processing server allocation unit 35b executes a request processing server allocation initialization process for each divided domain (step S202). This process is repeatedly executed for all divided domains. In such processing, as shown in FIG. 11, the request processing server allocation unit 35b requests the request processing server allocation state management table 34d with the minimum number of allocated domains and the minimum number of allocated domains. One of the processing servers is determined as a request processing server of the divided domain to be allocated (step S301).

  Then, the request processing server allocation unit 35b adds the information of the determined request processing server to the record of the domain that is the division source of the division domain in the request processing server allocation table 34c (step S302). Further, the request processing server allocating unit 35b adds information on the divided domain (domain to be divided) to be allocated to the determined request processing server record in the request processing server allocation state management table 34d (step S303). The request processing server allocation unit 35b repeatedly executes the above-described processing for all the divided domains.

  Next, service request response processing will be described with reference to FIGS. FIG. 12 is a sequence diagram for explaining the procedure of the service request response process in the information processing system 1 according to the first embodiment. As shown in FIG. 12, in the load distribution apparatus 30, when a service request is received from the client 10 (step S401), the request processing server determining unit 35d determines a request processing server to which the received service request is transferred (step S402). ). The service request received from the client 10 includes a single service specification value (key information).

  Then, when the request processing server determination unit 35d determines the request processing server 40 to which the service request is transferred, the load distribution apparatus 30 transfers the service request to the determined request processing server 41 (step S403). Upon receiving the service request, the request processing server 41 executes service response creation processing (step S404) and transmits a service response (step S405). The load balancer 30 that has received the service response transfers the received service response to the client that has transmitted the service request (step S406).

  Next, details of the request processing server determination process shown in step S402 in FIG. 12 will be described. FIG. 13 is a flowchart for explaining the procedure of the request processing server determination process by the load balancer 30 according to the first embodiment. As illustrated in FIG. 13, when the load distribution apparatus 30 receives a service request from the client 10, the request processing server determination unit 35d calculates a hash value of a service designation value included in the service request, and calculates the calculated hash value. Is used as a key to search the domain start hash value table 34a to determine the domain of the service specified value (step S501).

  Then, the request processing server determination unit 35d searches the request processing server allocation table 34c using the determined domain as a key, and the request processing server 40 serving as the transfer destination of the service request belonging to the corresponding domain (divided domain). A list is acquired (step S502). Further, the request processing server determination unit 35d randomly selects one from the request processing servers included in the request processing server list and determines it as a request processing server for processing the service request (step S503). In the above-described processing example, the case where the request processing server is selected at random has been described. However, this is merely an example, and the embodiment is not limited thereto. For example, the request processing server may be selected in consideration of the allocation status of the divided domains.

  Next, details of the service response creation process shown in step S404 in FIG. 12 will be described. FIG. 14 is a flowchart for explaining a procedure of service response creation processing by the request processing server 41 according to the first embodiment. As shown in FIG. 14, when the request processing server 41 receives a service request from the load balancer 30, the request processing server 41 searches the memory for an attribute value corresponding to the service specified value using the service specified value included in the service request as a key. (Step S601).

  If there is a corresponding record (Yes at Step S602), the request processing server 41 creates a service response including the corresponding attribute value (Step S603), and the created service response is sent to the load balancer 30. Send to. On the other hand, if there is no corresponding record (No at Step S602), the request processing server 41 creates a service response without including the corresponding attribute value (Step S604), and uses the created service response as the load balancer 30. Send to. In the above-described example, the case where the request processing server 41 executes the service response creation process has been described. However, other request processing servers 40 can execute the same process.

  Next, the allocation change process will be described with reference to FIGS. FIG. 15 is a sequence diagram for explaining the procedure of the allocation change process when the number of domain divisions increases in the information processing system 1 according to the first embodiment. As shown in FIG. 15, in the load balancer 30, the request frequency update unit 35e counts the requests for each domain for the service requests transmitted from the client 10, and periodically updates the domain request frequency table 34b. A request frequency monitoring / update process, which is a process to be executed, is executed (step S701).

  When the domain request frequency table 34b is updated by the request frequency update unit 35e, the request processing server allocation unit 35b executes a request processing server allocation change process (step S702), and the request processing newly allocated as the request processing server An allocation domain information addition notification is transmitted to the server 46 (step S703). Upon receiving the allocation domain information addition notification, the request processing server 46 transmits an attribute value information request to the storage apparatus 51 (step S704), and receives an attribute value information response (step S705). Upon receiving the attribute value information response, the request processing server 46 transmits an allocation domain information addition response to the load balancer 30 (step S706).

  Next, details of the request processing server assignment change process shown in step S702 of FIG. 15 will be described. FIG. 16 is a flowchart for explaining a procedure of request processing server allocation change processing when the number of domain divisions by the load distribution apparatus 30 according to the first embodiment increases. As shown in FIG. 16, when the number of domain divisions increases, the request processing server allocation unit 35b determines whether or not there is a request processing server whose allocation domain number is less than the server division number in the request processing server allocation state management table 34d. Is determined (step S801).

  Here, when it is determined that there is a request processing server having an allocation domain number less than the server division number (Yes in step S801), the request processing server allocation unit 35b further performs a request process in which the allocation domain number is less than the server division number. It is determined whether or not there is a request processing server having the smallest number of assigned domains and the smallest number of assigned domains in the same domain as the divided domain to be assigned (step S802).

  Here, when it is determined that there is a request processing server having the smallest number of assigned domains and the smallest number of assigned domains in the same domain as the divided domain to be assigned (Yes in step S802), the request processing server. The allocating unit 35b determines one of the corresponding request processing servers as one of the request processing servers in the divided domain (step S803). On the other hand, if it is determined that there is no request processing server having the smallest number of assigned domains and the number of assigned domains having the same division source as the divided domain to be assigned (No in step S802), request processing server assignment is performed. The unit 35b plans to allocate one of the request processing servers whose allocation domain is less than the number of server divisions and whose allocation domain is the same as the allocation domain and whose division source is the smallest. It is determined as one of the request processing servers in the divided domain (step S804).

  On the other hand, if it is determined in step S801 that there is no request processing server having an allocation domain number less than the server division number (No in step S801), the request processing server allocation unit 35b determines that an unused request processing server is used from the server pool information. 40 is acquired, and the request processing server 40 is added to the request processing server allocation state management table 34d, and is determined as one of the request processing servers of the divided domain to be allocated (step S805).

Then, the request processing server allocation unit 35b adds the determined request processing server information to the record of the domain in the request processing server allocation table 34c (step S806), and further determines the request processing server allocation state management table 34d. The domain information is added to the request processing server record (step S807). The request processing server allocation unit 35b repeatedly executes the above-described processing for the increase in the number of domain divisions.

  Thereafter, the allocation domain information notification unit 35c transmits an allocation domain information addition notification to the newly allocated request processing server (step S808), and the load distribution apparatus 30 sends an allocation domain information addition response from the request processing server. Receiving (step S809), the process ends.

  FIG. 17 is a sequence diagram for explaining the procedure of the allocation change process when the number of domain divisions decreases in the information processing system 1 according to the first embodiment. As shown in FIG. 17, in the load balancer 30, the request frequency update unit 35e counts requests for each domain for service requests transmitted from the client 10, and periodically updates the domain request frequency table 34b. A request frequency monitoring / update process, which is a process to be executed, is executed (step S901).

  When the request frequency update unit 35e updates the domain request frequency table 34b, the request processing server allocation unit 35b executes a request processing server allocation change process (step S902) and deletes the request processing server 46 from the request processing server. A notification of deletion of the allocated domain information is transmitted to (step S903). Upon receiving the allocation domain information deletion notification, the request processing server 46 transmits an allocation domain information deletion response to the load balancer 30 (step S904).

  Next, details of the request processing server allocation change process shown in step S902 of FIG. 17 will be described. FIG. 18 is a flowchart for explaining the procedure of a request processing server allocation change process when the number of domain divisions by the load distribution apparatus according to the first embodiment decreases. As shown in FIG. 18, when the number of domain divisions decreases, the request processing server allocation unit 35b has the largest number of allocated domains in the request processing server allocation state management table 34d, and the division domain and the division source scheduled to be deleted. It is determined whether there is a request processing server with the largest number of allocated divided domains (step S1001).

  If it is determined that there is a request processing server with the largest number of assigned domains and the assigned number of the divided domains having the same division source as the divided domain to be deleted (Yes in step S1001), the request processing is performed. The server allocation unit 35b determines one of the corresponding request processing servers as a request processing server for canceling the allocation of the divided domain (step S1002).

  On the other hand, if it is determined that there is no request processing server having the largest number of assigned domains and the number of assigned divided domains having the same division source as the divided domain to be deleted (No in step S1001), the request processing server The allocating unit 35b determines one of the request processing servers having the largest number of allocations of the divided domains having the same division source as the divided domain to be deleted as a request processing server for canceling the allocation of the divided domains (step S1003).

Then, the request processing server allocation unit 35b deletes the information of the request processing server determined from the record of the domain in the request processing server allocation table 34c (step S1004), and further determines the request processing server allocation state management table 34d. The information of the divided domain is deleted from the record of the request processing server (step S1005). The request processing server allocation unit 35b repeatedly executes the above-described processing for the decrease in the number of domain divisions.

  Thereafter, the allocation domain information notifying unit 35c transmits an allocation domain information change notification to the request processing server that has canceled the allocation of the divided domains (step S1006), and the load distribution apparatus 30 receives the allocation domain information from the request processing server. A change notification response is received (step S1007), and the process ends.

[Effect of the first embodiment]
As described above, according to the first embodiment, in the load balancer 30 that transfers a process execution request received from a client device to one of a plurality of request processing servers, the domain classification unit 35 a The hash value of the key information indicating the processing content of the processing execution request that will be received from the server 10 is calculated, and the processing execution request is classified into a predetermined number of domains based on the calculated hash value. The request processing server allocating unit 35b divides each domain into divided domains in which the request frequency of execution requests is equal to or less than a predetermined threshold, and the divided domains in each domain have an allocated number for each of the plurality of request processing servers 40. Allocation is made so that it is less than a predetermined number. When the request processing server determination unit 35d receives a processing execution request from the client 10, the request processing server determination unit 35d calculates a hash value indicating the processing content of the execution request, and classifies the received execution request based on the calculated hash value. The specified domain is specified, and the request processing server 40 to which the divided domain of the specified domain is assigned is determined as the transfer destination of the execution request. Therefore, the load distribution apparatus 30 according to the first embodiment classifies service requests into a plurality of domains based on the hash value of the key information, and uses the memory so that the number of key information in each domain is approximately equal. In addition to making the amount substantially equal and dividing into divided domains according to the request frequency of each domain and assigning them to each server substantially equally, the processing amount of each request processing server 40 can be made approximately equal. As a result, the load distribution device 30 can perform load distribution in consideration of the server processing amount and memory usage.

  Further, according to the first embodiment, the request processing server allocation unit 35b uses a value obtained by dividing the processing performance of the request processing server 40 by a predetermined number as a predetermined threshold. Therefore, the load distribution apparatus 30 according to the first embodiment assigns the maximum number of divided domains by setting a value obtained by dividing the processing performance by the maximum number of divided domains assigned to the request processing server 40 as a division threshold. However, the processing amount after the allocation does not exceed the processing performance of the request processing server 40, and stable load distribution can be executed.

  In addition, according to the first embodiment, the request processing server allocation unit 35b determines the number of request processing servers 40 to which the divided domains are allocated by dividing the total number of the divided domains by a predetermined number. A divided domain is assigned to the request processing server 40. Therefore, the load distribution apparatus 30 according to the first embodiment can determine the optimal number of request processing servers 40 without excess or deficiency as the allocation server, and can effectively use the server system.

  In addition, according to the first embodiment, the request processing server allocation unit 35b has the smallest number of allocation of divided domains among the plurality of request processing servers 40, and the divided domain whose allocation source domain is scheduled to be allocated The request processing server 40 with the smallest number of allocations of the same divided domain is determined as the allocation destination of the allocation target divided domain. Therefore, the load distribution apparatus 30 according to the first embodiment can suppress allocation of divided domains having the same division source to the same request processing server 40, and the processing amount and memory usage of the request processing server 40 Allows you to make the quantity more even. In addition, since the processing of a specific domain is rarely executed only by the specific request processing server 40, the load distribution apparatus 30 according to the first embodiment has a failure in any of the request processing servers 40. In this case, it is possible to prevent the influence of a failure from being biased to a specific domain.

  For example, when the processing amount of each domain is evenly distributed to N servers (when each divided domain of one domain is assigned to N servers), it is temporarily assigned to one of the servers. Even if a failure occurs, the influence on the domain is only 1 / N. This is more effective for a domain that is frequently processed and is divided into a plurality of divided domains.

  Further, according to the first embodiment, the request processing server determination unit 35d determines that each of the request processing servers 40 is divided when the divided domains divided from the same domain are assigned to different request processing servers 40, respectively. The execution requests are transferred so that the transfer times of the execution requests belonging to the domain are substantially equal. Therefore, the load distribution apparatus 30 according to the first embodiment makes it possible to prevent a specific process from being concentrated on a specific request processing server 40.

  Further, according to the first embodiment, the request frequency update unit 35e monitors the request frequency of the execution request for each domain and updates the request frequency. When the request frequency is updated by the request frequency update unit 35e, the request processing server determination unit 35d re-divides the domain into divided domains in which the updated request frequency is equal to or less than a predetermined threshold, and the divided domains after the re-division The allocation to the plurality of request processing servers 40 is changed according to the increase / decrease. Therefore, the load distribution apparatus 30 according to the first embodiment makes it possible to execute load distribution flexibly according to the request frequency of service requests that change over time.

  Further, according to the first embodiment, when the number of divided domains increases, the request processing server determination unit 35d increases the request processing server that has the smallest number of allocated divided domains with the same split source domain. Assign split domains. Therefore, the load distribution apparatus 30 according to the first embodiment makes it possible to assign the divided domains in a balanced manner even when the divided domains increase.

  Further, according to the first embodiment, when the number of divided domains decreases, the request processing server determination unit 35d decreases from the request processing server 40 having the largest number of allocations of divided domains with the same source domain. Unassign the split domain for the minute. Therefore, the load distribution apparatus 30 according to the first embodiment makes it possible to cancel allocation of divided domains in a balanced manner even when the number of divided domains is reduced.

[Second Embodiment]
Although the first embodiment has been described so far, the embodiment according to the present application is not limited to the first embodiment. That is, these embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made.

  In the first embodiment described above, a case has been described in which control is performed so that the same divided domain as the division source is assigned to the request processing server 40 having the smallest number of assignments already assigned. However, the embodiment is not limited to this, and for example, it may be a case where control is performed so that the same divided domain as the division source is not assigned to the same request processing server 40.

  In such a case, if there is no server with the allocation number “0” already assigned to the split domain in which the split source domain is the same as the scheduled split domain, an unused request processing server is referred to the server pool information. 40 is a server to be assigned. Thereby, it is possible to improve the fault tolerance (≈improve the safety of the entire system when one or more servers fail). Note that the above-described split domain allocation method increases the number of required request processing servers 40, and is therefore preferably used when there are a large number of unused request processing servers 40.

  In addition, for example, the specific form of distribution / integration of each device (for example, the form of FIG. 2) is not limited to that shown in the figure, and all or a part thereof can be arbitrarily set according to various loads or usage conditions. It can be distributed or integrated functionally or physically in units. For example, the request processing server allocation unit 35b and the request frequency update unit 35e may be integrated as one processing unit, while the request processing server allocation unit 35b includes a split domain generation unit that generates a split domain; Alternatively, it may be distributed to an allocation unit that allocates divided domains.

  In addition, the control unit 35 may be connected as an external device of the load distribution device 30 via a network, or each table stored in the storage unit 34 has a separate device and is connected to the network. You may make it implement | achieve the function of the load distribution apparatus 30 mentioned above by cooperating.

  These embodiments and modifications thereof are included in the invention disclosed in the claims and equivalents thereof as well as included in the technology disclosed in the present application.

11, 12, 13 Client 21, 22, 23 IP network 30 Load balancer 34 Storage unit 35 Control unit 35a Domain classification unit 35b Request processing server allocation unit 35c Allocation domain information notification unit 35d Request processing server determination unit 35e Request frequency update unit 41, 42, 43, 44, 45, 46 Request processing server 51 Storage device

Claims (8)

A load balancer that transfers a process execution request received from a client device to one of a plurality of request processing servers,
A hash value of key information indicating processing contents of a processing execution request that will be received from the client device is calculated, and the processing execution request is classified into a predetermined number of management units based on the calculated hash value. A classification section;
Each of the management units is divided into partial management units in which the request frequency of the execution request is equal to or less than a predetermined threshold, and the partial management unit in each management unit is assigned a predetermined number of requests to each of the plurality of request processing servers. An assigning part to be assigned, so that
When a processing execution request is received from the client device, a hash value indicating the processing content of the execution request is calculated, and a management unit into which the received execution request is classified is specified based on the calculated hash value. Determining a request processing server to which a partial management unit of the specified management unit is assigned as a transfer destination of the execution request;
A load balancer comprising:   The load distribution apparatus according to claim 1, wherein the allocation unit uses a value obtained by dividing the processing performance of the request processing server by the predetermined number as the predetermined threshold.   The allocation unit determines the number of request processing servers to which the partial management unit is allocated by dividing the total number of the partial management units by the predetermined number, and allocates the partial management unit to the determined number of request processing servers. The load distribution apparatus according to claim 1 or 2, wherein   The allocation unit has an allocation number of partial management units in which the allocation number of the partial management unit is the smallest among the plurality of request processing servers, and the divisional management unit is the same as the partial management unit to be allocated 4. The load distribution apparatus according to claim 1, wherein a request processing server having a minimum is determined as an allocation destination of the partial management unit scheduled to be allocated. 5.   In the case where the partial management units divided from the same management unit are assigned to different request processing servers, the determination unit is configured to reduce the number of execution requests belonging to the management unit to each request processing server. The load distribution apparatus according to claim 1, wherein the execution requests are transferred so as to be equal. An update unit that monitors the request frequency of the execution request for each management unit and updates the request frequency,
The allocating unit subdivides the management unit into partial management units in which the updated request frequency is equal to or less than a predetermined threshold when the request frequency is updated by the updating unit, and the partial management unit after the subdivision The load distribution apparatus according to claim 1, wherein allocation to the plurality of request processing servers is changed in accordance with increase / decrease.   The allocating unit allocates an increased partial management unit to a request processing server having the smallest number of allocated partial management units with the same division source management unit when the partial management unit increases. The load distribution apparatus according to claim 6.   When the partial management unit is decreased, the allocating unit releases allocation of the reduced partial management unit from the request processing server having the largest allocation number of partial management units having the same divisional management unit. The load distribution apparatus according to claim 6.

Images