JP2018117300A - Control device, control program and control method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow an operator to compensate for deviation of node selection with the saving operation amount.SOLUTION: In a storage part 1a, a plurality of numerical values are stored regarding the number of times in which each of nodes 2, 3, 4 is selected. The plurality of numerical values include a first numerical value corresponding to a node 2 and a second numerical value corresponding to a node 4. When the node 2 is selected as destination of first data, a processing part 1b decreases the first numerical value and increases the second numerical value corresponding to the node 4 not selected. When the second numerical value becomes equal to or more than a specified value, the processing part 1b selects the node 4 as the destination of second data.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a control device, a control program, and a control method.

  In a system including a plurality of devices, each device is connected to a network and communicates with each other. The network may include a plurality of relay nodes that relay communication between a data transmission source device and a destination device. In the network, when the amount of traffic increases, the load on the relay node also increases and communication may be delayed. In addition, when a failure occurs in a certain relay node, there is a possibility that communication via the corresponding relay node cannot be performed. Therefore, at certain points on the network (for example, where traffic is likely to concentrate), multiple relay nodes are made redundant and data transfer is distributed to reduce communication delays and improve the availability of communication functions. May be planned.

  For example, a system is considered in which packet transfer is distributed using a plurality of switch devices. In this proposal, each transmission port of a switch device belonging to the upper switch group is connected to a plurality of switch devices belonging to the lower switch group. The switch device belonging to the upper switch group selects a transmission port for transmitting a packet according to the hash value calculated from the header of the packet.

  Further, when the relay device that relays communication between the terminal device and the server device measures the communication amount with the terminal device for each server device and receives connection request information from the terminal device, the communication amount measured in a predetermined period. There is also a proposal for transmitting the connection request information to the server device having the smallest number.

  There is also a proposal for a data distribution device that distributes data stored in the first storage device group equally to two storage devices belonging to the second storage device group for each data type. The data distribution device includes a counter group for storing the cumulative number of data output to two storage devices belonging to the second storage device group for each type of the data, and based on the contents of the counter group, The bias for each type of data transferred to the second storage device group is determined.

JP 2006-82333 A JP 2010-56898 A JP-A-4-127271

  When data transfer is distributed by a plurality of redundant relay nodes, it is conceivable to provide a control device that distributes data to be transferred to each relay node. For example, if the processing capacity of each relay node is approximately the same, the control device can appropriately distribute the load of each relay node by distributing data to each relay node to some extent equally according to a predetermined rule. . However, in actual operation, the control device is not always able to distribute data equally to each relay node, and there may be a bias in the selection of the relay node to which data is distributed. Therefore, considering the occurrence of bias in node selection, a mechanism for correcting the bias becomes a problem.

  For example, the control device stores the cumulative number of distributions for each relay node, detects the presence of a relay node with a relatively large cumulative number, detects a bias in node selection, and selects nodes that are biased A method of transferring the assignment destination to a node other than is possible. Specifically, the control device compares the cumulative number of allocations for a certain relay node with the cumulative number of allocations for another relay node by changing the combination of relay nodes, so that the cumulative number The presence of a relay node having a relatively large number can be detected. However, with this method, if the number of relay nodes increases, the amount of computation required for comparison may become excessive.

  In one aspect, an object of the present invention is to make it possible to correct a bias in node selection with a small amount of computation.

  In one aspect, a control device is provided that selects and transfers one of a plurality of nodes when transferring data. The control device includes a storage unit and a processing unit. The storage unit is a plurality of numerical values relating to the number of times each of the plurality of nodes is selected, and includes a plurality of numerical values including a first numerical value corresponding to the first node and a second numerical value corresponding to the second node. Remember. When the first node is selected as the transfer destination of the first data, the processing unit decreases the first numerical value, increases the second numerical value corresponding to the second node not selected, When the numerical value exceeds a predetermined value, the second node is selected as the transfer destination of the second data.

  In one aspect, the bias in node selection can be corrected with a small amount of computation.

It is a figure which shows the control apparatus of 1st Embodiment. It is a figure which shows the example of the information processing system of 2nd Embodiment. It is a figure which shows the hardware example of the distribution server of 2nd Embodiment. It is a figure which shows the function example of the distribution server of 2nd Embodiment. It is a figure which shows the example of the packet of 2nd Embodiment. It is a figure which shows the example of the round robin table of 2nd Embodiment. It is a figure which shows the example of the unselect counter of 2nd Embodiment. It is a figure which shows the example of the USC update rule of 2nd Embodiment. It is a flowchart which shows the packet transfer example of 2nd Embodiment. It is a figure which shows the specific example of the transfer destination transfer of 2nd Embodiment. It is a figure which shows the comparative example (the 1) with respect to 2nd Embodiment. It is a figure which shows the comparative example (the 2) with respect to 2nd Embodiment. It is a figure which shows the example of the packet of 3rd Embodiment. It is a figure which shows the example of the session management table of 3rd Embodiment. It is a flowchart which shows the packet transfer example of 3rd Embodiment. 15 is a flowchart illustrating an example of canceling transfer destination transfer according to the third embodiment. It is a figure which shows the specific example of the transfer destination transfer of 3rd Embodiment. It is a figure which shows the comparative example with respect to 3rd Embodiment. It is a figure which shows the other example (the 1) of transfer destination transfer. It is a figure which shows the other example (the 2) of transfer destination transfer. It is a figure which shows the example of a distribution accumulation counter. It is a figure which shows the comparative example of a distribution accumulation counter and USC.

Hereinafter, the present embodiment will be described with reference to the drawings.
[First Embodiment]
FIG. 1 is a diagram illustrating a control device according to the first embodiment. The system according to the first embodiment includes a control device 1, nodes 2, 3, 4 and destination devices 5, 6, 7. The control device 1 is connected to the nodes 2, 3, and 4. Nodes 2, 3, and 4 are connected to the network N1. The destination devices 5, 6, and 7 are connected to the network N1. Furthermore, the control device 1 is connected to the network N2. Terminal devices 8 and 9 are also connected to the network N2. However, the control device 1 may be connected to the nodes 2, 3, and 4 via a switch device (not shown).

  Here, the control device 1 selects and transfers one of the nodes 2, 3 and 4 when transferring the data. That is, when transferring data, the control device 1 selects one of the nodes 2, 3, and 4 as a transfer destination, and transfers the data to the selected transfer destination node.

  The destinations of the data transmitted by the terminal devices 8 and 9 are the destination devices 5, 6, and 7, respectively. Data may be transmitted in units called packets. Data transmitted by each of the terminal devices 8 and 9 can pass through a plurality of routes before reaching the destination device 5, 6, or 7. The first route is a route that passes through the control device 1 and the node 2. The second route is a route that passes through the control device 1 and the node 3. The third route is a route that passes through the control device 1 and the node 4. Which of the nodes 2, 3, 4 is routed is controlled by the control device 1. Therefore, the nodes 2, 3, and 4 are candidates for data transfer destinations by the control device 1. The “data transfer destination candidate” may be expressed as “data distribution destination candidate”.

  The nodes 2, 3, and 4 transfer data to the network N1. The network N1 transfers data to the destination devices 5, 6, and 7. It can be said that the nodes 2, 3, and 4 are relay nodes that relay data. It is assumed that the capacity of data transfer processing by the nodes 2, 3, and 4 is substantially the same.

  The control device 1 includes a storage unit 1a and a processing unit 1b. The storage unit 1a may be a volatile storage device such as a RAM (Random Access Memory) or a non-volatile storage device such as an HDD (Hard Disk Drive) or a flash memory. The processing unit 1b may include a CPU (Central Processing Unit), a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), and the like. The processing unit 1b may be a processor that executes a program. Here, the “processor” may include a set of multiple processors (multiprocessor).

  The storage unit 1a stores a plurality of numerical values related to the number of times each of the nodes 2, 3, and 4 is selected. The plurality of numerical values include a numerical value corresponding to the node 2, a numerical value corresponding to the node 3, and a numerical value corresponding to the node 4.

  For example, the identification information (node #) of node 2 is “1”. The identification information of the node 3 is “2”. The identification information of the node 4 is “3”. The storage unit 1a stores a table T1 in which node identification information (node #) is associated with a counter. For example, at a certain timing, a record of node # “1” and counter “−1” is registered in the table T1. The numerical value registered in the counter item is a numerical value related to the number of times the corresponding node is selected as a data transfer destination by the processing unit 1b (a specific counting method will be described later). Similarly, a record of node # “2” and counter “−2” is registered in the table T1. Further, records of node # “3” and counter “3” are registered in the table T1. Note that the initial value of the counter of each of the nodes 2, 3, and 4 is, for example, “0”.

  The processing unit 1b determines to which of the nodes 2, 3 and 4 the data to be transferred is distributed (to which data is transferred) according to a predetermined rule. The following example can be considered as the predetermined rule used by the processing unit 1b.

  As a first example, a rule for distributing data to nodes 2, 3, and 4 with equal probability for each data destination (for example, a rule for determining a transfer destination using a round robin or a random number) can be considered. . When round robin is performed for each data destination, the processing unit 1b distributes the data destined for the destination device 5 to the nodes 2, 3, and 4 in order (node 2 is the next node 4). Further, the processing unit 1b distributes the data destined for the destination device 6 to the nodes 2, 3, and 4 in order (the node 2 is next to the node 4). Further, the processing unit 1b distributes the data destined for the destination device 7 to the nodes 2, 3, and 4 in order (the node 2 is the node next to the node 4). As a result, access from the terminal devices 8 and 9 can be distributed to each redundant path for each destination.

  As a second example, a rule for selecting a data transfer destination according to a flow identified by information of a header included in a packet can be considered. More specifically, the processing unit 1b identifies a flow from 5 tuples (protocol number, transmission source / destination IP address and transmission source / destination port number) included in an IP (Internet Protocol) packet, and calculates by the 5 tuples. The transfer destination may be selected according to the hash value to be processed. Thereby, the access from the terminal devices 8 and 9 can be distributed to each redundant path for each flow.

  The processing unit 1b selects the first node as the transfer destination for the first data according to the predetermined rule. Then, the processing unit 1b decreases the counter value (first numerical value) corresponding to the first node, and increases the counter value (second numerical value) corresponding to the second node that is not selected. The amount to be reduced and the amount to be increased can be arbitrarily determined. In the example of the first embodiment, the decrease amount is set to “−2” and the increase amount is set to “1” so that the total increase / decrease amount becomes “0”.

  For example, it is assumed that the processing unit 1b selects the node 2 as the transfer destination for the first data. In this case, the processing unit 1b decreases the numerical value of the counter of the node # “1” in the table T1 by “2”. On the other hand, the processing unit 1b increases the numerical value of the counter of the node # “2” and the numerical value of the counter of the node # “3” by “1”, respectively.

  More specifically, the processing unit 1b updates the numerical value “−1” registered for the node # “1” in the table T1 to −1-2 = −3. In the table T1, the processing unit 1b updates the numerical value “−2” registered for the node # “2” to −2 + 1 = −1. In the table T1, the processing unit 1b updates the numerical value “3” registered for the node # “3” to 3 + 1 = 4. The table T1a shows after the results of these updates are reflected. It can also be said that the processing unit 1b counts the number of times that the nodes 2, 3, and 4 are not selected as data transfer destinations.

  Then, when the second numerical value is equal to or greater than the predetermined value, the processing unit 1b selects the second node associated with the second numerical value that is equal to or greater than the predetermined value as the transfer destination of the second data. . For example, the storage unit 1a stores in advance the counter threshold value = 4 as the predetermined value. In the example of the table T1a, the value of the counter corresponding to the node 4 (node # “3”) is “4”. Therefore, the processing unit 1b detects that the numerical value “4” corresponding to the node 4 is equal to or greater than the predetermined value “4”. Therefore, the processing unit 1b selects the node 4 as the transfer destination of the second data regardless of the predetermined rule described above. The processing unit 1b transfers the second data to the selected node 4.

  When the processing unit 1b selects the node 4, the processing unit 1b updates the table T1a. Specifically, the processing unit 1b updates the numerical value of the counter of the node # “1” to −3 + 1 = −2. Further, the processing unit 1b updates the numerical value of the counter of the node # “2” to −1 + 1 = 0. Further, the processing unit 1b updates the numerical value of the counter of the node # “3” to 4-2 = 2. However, in FIG. 1, illustration of the updated table is omitted.

  By the way, for example, when the processing unit 1b performs load distribution in a round robin manner for each destination, if the transfer destination selected by the round robin is synchronized with respect to each destination of a plurality of data, traffic to a specific node is May increase in bursts. In this case, data transfer waiting for the corresponding node occurs, and in the worst case, data may be discarded due to buffer overflow.

  Further, even when the processing unit 1b selects a transfer destination node so as to be unique for each flow, if the transfer destinations of a plurality of flows are concentrated on a specific node, traffic to the specific node is bursty. And similar problems can arise.

Therefore, considering the occurrence of bias in node selection, a mechanism for correcting the bias becomes a problem.
For example, a device that distributes data to a plurality of nodes stores the selected cumulative number for each node, and detects the bias of node selection by detecting the presence of a node with a relatively large cumulative number. A method of transferring the distribution destination to a node other than the biased node is also conceivable. Specifically, the allocation device performs a brute force comparison of the cumulative number of allocations for a certain relay node and the cumulative number of allocations for other relay nodes while changing the combination of relay nodes. The presence of a relay node having a relatively large cumulative number can be detected. However, with this method, if the number of nodes increases, the amount of computation required for comparison may become excessive.

  More specifically, as one comparative example, a combination of two nodes is compared brute force, and the minimum value of the cumulative number of selections and an index indicating the degree of bias (for example, the difference between the two cumulative times) And calculating the average value or the maximum value). For example, when an index indicating the degree of bias exceeds a predetermined threshold, it can be determined that bias has occurred in node selection. However, in this method, the amount of calculation increases linearly as the number of transfer destination candidate nodes increases. In addition, an operation for detecting the occurrence of bias by an index representing the difference in the number of times of accumulation, and an operation for selecting a transfer destination node (for example, selecting a node having the selected accumulation number having the minimum value). ) Will be performed separately. For this reason, the method of the comparative example cannot sufficiently speed up the data transfer destination selection process.

  On the other hand, the control device 1 counts down the counter of the node selected as the data transfer destination, and counts up the counter of the node not selected. For this reason, the control device 1 can efficiently identify a node with a relatively small number of times selected as a transfer destination based on the counter. In the example of FIG. 1, the control device 1 can determine that there is a bias in node selection by detecting that the counter value of the node 4 is equal to or greater than a predetermined threshold. This is because the relatively large number of times that node 4 has not been selected means that the number of times that another node (for example, both of nodes 2 and / or 3) has been selected is relatively large. Because. Furthermore, since a node that has not been selected a relatively large number of times has a relatively small amount of distributed data, it is considered that resources (including resources used for data transfer) have a relatively large margin. Therefore, the control device 1 can determine that it is better to preferentially select the node 4 when attempting to distribute the data to the nodes 2, 3, 4 evenly to some extent.

  In this way, the control device 1 can efficiently determine both the occurrence of bias in node selection and the nodes to be selected as transfer destinations to correct the bias based on the counter values in the table T1a. . In particular, as in the above-described comparative example, the amount of calculation can be reduced as compared with the method of detecting the bias by comparing the cumulative number of selections between nodes in a brute force manner.

  Thus, according to the control device 1, it is possible to correct the bias in node selection with a small amount of calculation. As a result, the calculation accompanying the selection of the data transfer destination of the control device 1 can be speeded up, which contributes to a reduction in communication delay between the terminal devices 8, 9 and the destination devices 5, 6, 7.

Below, a more specific information processing system is illustrated and the function of the control apparatus 1 is demonstrated in detail.
[Second Embodiment]
FIG. 2 is a diagram illustrating an example of an information processing system according to the second embodiment. The information processing system according to the second embodiment includes distribution servers 100, 100a, 100b, relay servers 200, 200a, 200b, business servers 300, 300a, 300b,..., 300m, and a switch (SW: SWitch) 400. including.

  The distribution servers 100, 100a, 100b and the relay servers 200, 200a, 200b are connected to the SW 400. Relay servers 200, 200a, 200b and business servers 300, 300a, 300b,..., 300m are connected to the network 10.

  The distribution servers 100, 100 a, 100 b are connected to the network 20. Furthermore, clients 500, 500a, 500b,..., 500n are connected to the network 20. The networks 10 and 20 are, for example, LANs (Local Area Networks). However, the networks 10 and 20 may be a WAN (Wide Area Network).

  The information processing system according to the second embodiment relays access to the business servers 300, 300a, 300b,..., 300m by the clients 500, 500a, 500b,. Communication between the two.

  Here, in the information processing system according to the second embodiment, Ethernet (registered trademark) is used as a communication protocol in layer 2 of the OSI (Open Systems Interconnection) reference model. Further, IP (Internet Protocol) is used as a communication protocol in the layer 3. Further, UDP (User Datagram Protocol) or TCP (Transmission Control Protocol) is used as a communication protocol in the layer 4.

  The distribution servers 100, 100a, 100b are server computers that receive packets transmitted by the clients 500, 500a, 500b,..., 500n and distribute them to the relay servers 200, 200a, 200b. The distribution servers 100, 100a, 100b distribute packets to each relay server by designating the MAC (Media Access Control) addresses of the relay servers 200, 200a, 200b that are the distribution destinations. Here, the “allocation destination” of the packet by the allocation servers 100, 100a, 100b corresponds to the “forwarding destination” of the packet by the allocation servers 100, 100a, 100b. The distribution servers 100, 100a, and 100b are an example of the control device 1 according to the first embodiment.

  The relay servers 200, 200 a, and 200 b are server computers that accept packets distributed by the distribution servers 100, 100 a, and 100 b and transfer them to the network 10. The packet transfer processing capabilities of the relay servers 200, 200a, and 200b are almost the same. Routing from the relay server 200, 200a, 200b to the destination business server (any of the business servers 300, 300a, 300b,..., 300m) is performed using IP or the like. For example, the relay servers 200, 200a, and 200b transfer the packet to the next hop belonging to the network 10 according to the routing table held by each of them. It can also be said that the relay servers 200, 200a, and 200b form one virtual router (virtual router). The relay servers 200, 200a, and 200b are examples of the nodes 2, 3, and 4 according to the first embodiment.

  In the information processing system according to the second embodiment, distribution servers 100, 100a, and 100b are provided at the edges that accept access from each client. Then, ECMP (Equal Cost Multi Path) routing is realized for each destination IP address of the packet by load distribution by the distribution servers 100, 100a, and 100b. ECMP routing is a technique in which a plurality of paths are provided between a transmission source device and a destination device, and packets are transferred using each path equally.

  The business servers 300, 300a, 300b,..., 300m are server computers that provide predetermined business services to the clients 500, 500a, 500b,. The business service is, for example, a name resolution service based on DNS (Domain Name System). DNS uses UDP. However, the business service may be a Web service using TCP, for example. The business servers 300, 300a, 300b,..., 300m are examples of the destination devices 5, 6, and 7 according to the first embodiment.

The SW 400 is a switch device that aggregates connections of the distribution servers 100, 100a, 100b and the relay servers 200, 200a, 200b.
The clients 500, 500a, 500b,..., 500n are client computers used by users. The clients 500, 500a, 500b,..., 500n are examples of the terminal devices 8 and 9 according to the first embodiment.

  Here, NFV (Network Functions Virtualization) technology for realizing a network function by software on a server computer is known. For example, by using the NFV technology, there are merits such as cost reduction and flexible introduction according to operation. Further, by providing a plurality of such servers (for example, relay servers 200, 200a, and 200b) and executing the data transfer process in a distributed manner, the data transfer processing performance can be improved. At this time, by providing a plurality of distribution servers that distribute data to the relay servers 200, 200a, and 200b (for example, the distribution servers 100, 100a, and 100b), the data distribution processing performance can be improved.

  In the example of the second embodiment, the distribution servers 100, 100a, and 100b select a packet distribution destination (transfer destination) relay server by round robin for each packet destination IP address. However, with this method, at a certain timing, the same relay server may be selected as a distribution destination for packets with a plurality of destination IP addresses, and traffic may concentrate on a specific relay server. Such traffic concentration causes communication delay. Therefore, the distribution servers 100, 100a, and 100b provide a function of suppressing the bias of the relay servers selected as distribution destinations in order to suppress the concentration of traffic.

  FIG. 3 illustrates a hardware example of the distribution server according to the second embodiment. The distribution server 100 includes a processor 101, a RAM 102, an HDD 103, an image signal processing unit 104, an input signal processing unit 105, a medium reader 106, and a communication interface 107. Each hardware is connected to the distribution server 100 bus.

  The processor 101 is hardware that controls information processing of the distribution server 100. The processor 101 may be a multiprocessor. The processor 101 is, for example, a CPU, DSP, ASIC, or FPGA. The processor 101 may be a combination of two or more elements of CPU, DSP, ASIC, FPGA, and the like.

  The RAM 102 is a main storage device of the distribution server 100. The RAM 102 temporarily stores at least part of an OS (Operating System) program and application programs to be executed by the processor 101. The RAM 102 stores various data used for processing by the processor 101.

  The HDD 103 is an auxiliary storage device of the distribution server 100. The HDD 103 magnetically writes and reads data to and from the built-in magnetic disk. The HDD 103 stores an OS program, application programs, and various data. The distribution server 100 may include other types of auxiliary storage devices such as flash memory and SSD (Solid State Drive), or may include a plurality of auxiliary storage devices.

  The image signal processing unit 104 outputs an image to the display 11 connected to the distribution server 100 in accordance with an instruction from the processor 101. As the display 11, a CRT (Cathode Ray Tube) display, a liquid crystal display, or the like can be used.

  The input signal processing unit 105 acquires an input signal from the input device 12 connected to the distribution server 100 and outputs the input signal to the processor 101. As the input device 12, for example, a pointing device such as a mouse or a touch panel, a keyboard, or the like can be used.

  The medium reader 106 is a device that reads programs and data recorded on the recording medium 13. As the recording medium 13, for example, a magnetic disk such as a flexible disk (FD) or an HDD, an optical disk such as a CD (Compact Disc) or a DVD (Digital Versatile Disc), or a magneto-optical disk (MO) is used. Can be used. Further, as the recording medium 13, for example, a non-volatile semiconductor memory such as a flash memory card can be used. For example, the medium reader 106 stores the program and data read from the recording medium 13 in the RAM 102 or the HDD 103 in accordance with an instruction from the processor 101.

  The communication interface 107 communicates with other devices via the network 10. The communication interface 107 may be a wired communication interface or a wireless communication interface.

  The distribution servers 100a, 100b, the relay servers 200, 200a, 200b, the business servers 300, 300a, 300b,..., 300m and the clients 500, 500a, 500b,. It can be realized using hardware.

  FIG. 4 is a diagram illustrating a function example of the distribution server according to the second embodiment. The distribution server 100 includes a storage unit 110, an identification unit 120, a bias determination unit 130, and a distribution determination unit 140. For example, the storage unit 110 is realized using a storage area secured in the RAM 102 or the HDD 103. For example, the processor 101 exhibits the functions of the identification unit 120, the bias determination unit 130, and the distribution determination unit 140 by executing a program stored in the RAM 102.

  The storage unit 110 stores control information used for packet transfer control. The control information includes a round robin table for managing a distribution destination for each destination by round robin, and a counter regarding the number of times the relay server is selected as the distribution destination. A counter is provided for each of the relay servers 200, 200a, and 200b. The counter is used to count the number of times that it has not been selected as a distribution destination.

  The identification unit 120 accepts packets transmitted by the clients 500, 500a, 500b,..., 500n, and identifies the destination (destination IP address) of the packet with reference to the header information of the packet.

  The bias determination unit 130 refers to the counter for each relay server stored in the storage unit 110 and determines whether there is a bias in selection of the distribution destination. Specifically, the bias determination unit 130 compares the counter value for each relay server with a predetermined bias threshold. Then, the bias determination unit 130 determines that there is a bias in selection of the distribution destination when the value of the counter corresponding to at least one relay server is equal to or greater than the bias threshold. The bias determination unit 130 determines that there is no bias in the selection of the distribution destination when the counter values corresponding to all the relay servers are less than the bias threshold.

  The distribution determination unit 140 determines the distribution destination for the destination IP address identified by the identification unit 120 based on the round robin table stored in the storage unit 110 when the bias determination unit 130 determines that there is no bias. decide. The distribution determination unit 140 transfers the packet to the determined relay node.

  On the other hand, when the bias determination unit 130 determines that there is a bias, the distribution determination unit 140 determines a packet allocation destination based on the value of the counter stored in the storage unit 110. The distribution determination unit 140 transfers the packet to the determined relay node.

  FIG. 5 is a diagram illustrating an example of a packet according to the second embodiment. The packet P1 includes a payload f1, a TCP / UDP header f2, an IP header f3, and a MAC header f4. The main body of data to be transmitted is registered in the payload f1. TCP or UDP header information is registered in the TCP / UDP header f2. IP header information is registered in the IP header f3. The IP header f3 includes a destination IP address (D-IP: Destination-IP address). Ethernet header information is registered in the MAC header f4. The MAC header f4 includes a destination MAC address (D-MAC: Destination-MAC address).

  For example, the distribution determination unit 140 rewrites the destination MAC address included in the packet P1 with the MAC address of the relay server that is the distribution destination, thereby designating the relay server that is the distribution destination of the packet P1 to the SW 400. When SW400 receives packet P1 from distribution server 100, SW400 transmits packet P1 from the port corresponding to the destination MAC address (corresponding to the port to which the specified relay server is connected).

  A layer 3 PDU (Protocol Data Unit) that includes the payload f1, the TCP / UDP header f2, and the IP header f3 but does not include the MAC header f4 may be referred to as an “IP packet”. In this case, the layer 2 PDU including the MAC header f4 may be referred to as a “frame” (the packet P1 may be referred to as a “frame”).

  FIG. 6 is a diagram illustrating an example of a round robin table according to the second embodiment. The round robin table 111 is stored in the storage unit 110. The round robin table 111 includes items of destination # and Svr (Server) #.

In the destination # item, business server identification information is registered. In the item of Svr #, the identification information of the relay server selected as the current distribution destination is registered.
The identification information of the business server is associated with the IP address of the business server. In the example of the round robin table 111, “A”, “B”, “C”, “D”, “E”, “F”, “G”, “H” are indicated as the identification information of the business server. Yes.

  The identification information of the relay server 200 is “1”. When referring to the relay server 200, “Svr1” may be used. The identification information of the relay server 200a is “2”. When referring to the relay server 200a, it may be expressed as “Svr2”. The identification information of the relay server 200b is “3”. When referring to the relay server 200b, it may be expressed as “Svr3”.

  Here, in the round robin table 111, Svr # is set to “1 → 2 → 3” for each destination # so that it is easy to understand that the relay servers 200, 200a, and 200b are sequentially selected according to the rule of the round robin. ". In addition, the setting value of the Svr # item is represented by enclosing the identification information of the relay server selected this time with a dotted circle. After Svr # “3”, the process returns to Svr # “1”.

  For example, in the round robin table 111, a record in which the destination # is “A” and the Svr # is “1” is registered. This record indicates that the relay server 200 has been selected as the latest distribution destination for the packet of destination A. This indicates that the next distribution destination for the packet of destination A is the relay server 200a.

  Further, for example, in the round robin table 111, a record in which the destination # is “B” and the Svr # is “2” is registered. This record indicates that the relay server 200a is selected as the latest distribution destination for the packet of destination B. This indicates that the next distribution destination for the packet of destination B is the relay server 200b.

  Further, for example, in the round robin table 111, a record in which the destination # is “C” and the Svr # is “3” is registered. This record indicates that the relay server 200b is selected as the latest distribution destination for the packet of destination C. This indicates that the next distribution destination is the relay server 200 for the packet of the destination C.

Further, in the round robin table 111, a record indicating the latest selection result of the distribution destination is similarly registered for other destinations #.
FIG. 7 is a diagram illustrating an example of an unselect counter according to the second embodiment. An unselect counter (USC: UnSelectCounter) 112 is stored in the storage unit 110. The USC 112 includes items of Svr1, Svr2, and Svr3.

  In the item of Svr1, a count value corresponding to the relay server 200 is registered. The count value corresponding to the relay server 200a is registered in the item of Svr2. In the item of Svr3, a count value corresponding to the relay server 200b is registered.

  For example, in the USC 112, numerical values “−1” for Svr1, “−1” for Svr2, and “2” for Svr3 are registered. This is because the count value corresponding to the relay server 200 at a certain timing is “−1”, the count value corresponding to the relay server 200 a is “−1”, and the count value corresponding to the relay server 200 b is “2”. Indicates that

  FIG. 8 is a diagram illustrating an example of the USC update rule according to the second embodiment. The USC update rule 113 is stored in the storage unit 110 in advance. The USC update rule 113 includes items of category and added value.

  In the category item, a category indicating that it has been selected as a distribution destination (denoted as “selected”) or not selected (denoted as “non-selected”) is registered. In the addition value item, an addition value of USC 112 corresponding to the category is registered.

  For example, in the USC update rule 113, a record with the classification “non-selected” and the added value “+1” is registered. This record indicates that the addition value “+1” is added to the count value of the corresponding relay server in the USC 112 when it is not selected as a distribution destination.

  Further, for example, in the USC update rule 113, a record having a classification “selection” and an addition value “− (X−1)” is registered. This record indicates that, when selected as a distribution destination, an addition value “− (X−1)” is added to the count value of the corresponding relay server in the USC 112. This corresponds to “subtracting“ (X−1) ”from the count value”. Here, X is the number of relay servers that are candidates for packet distribution. In the example of the second embodiment, X = 3.

  For example, in the USC 112, consider a case where the distribution determination unit 140 distributes a packet to the relay server 200 when the count values of Svr1, Svr2, and Svr3 are all “0”. In this case, the relay server 200 is selected as a distribution destination (the relay server 200 is classified as “selected”). On the other hand, the relay servers 200a and 200b are not selected as distribution destinations (the relay servers 200a and 200b are classified as “non-selected”).

  Therefore, the distribution determination unit 140 subtracts “3-1 = 2” from the count value of Svr1 in the USC 112 in accordance with the USC update rule 113. Similarly, the distribution determination unit 140 adds “+1” to the count value of Svr2 in the USC 112. Similarly, the distribution determination unit 140 adds “+1” to the count value of Svr3 in the USC 112.

However, the addition value and the subtraction value exemplified in the USC update rule 113 are examples, and values other than those described above may be used.
Next, the processing procedure of the distribution servers 100, 100a, 100b will be described. In the following, although the procedure of the distribution server 100 will be mainly described, the distribution servers 100a and 100b also execute the same procedure.

FIG. 9 is a flowchart illustrating an example of packet transfer according to the second embodiment. In the following, the process illustrated in FIG. 9 will be described in order of step number.
(S11) The identification unit 120 receives the packet P1 transmitted by any client. The identification unit 120 refers to the IP header f3 of the packet P1 and identifies the business server identification information corresponding to the destination IP address of the packet P1.

  (S12) The bias determination unit 130 refers to the USC 112 stored in the storage unit 110, and determines whether there is a relay server in which the count value of the USC 112 is equal to or greater than the bias threshold. If there is a relay server whose USC 112 count value is greater than or equal to the bias threshold value, the bias determination unit 130 proceeds to step S13. If there is no relay server having a count value of USC 112 equal to or greater than the bias threshold, the bias determination unit 130 proceeds to step S14.

  (S13) The distribution determination unit 140 determines, as the distribution destination, the relay server that has been determined in step S12 that the count value of the USC 112 is equal to or greater than the bias threshold. Then, the distribution determination unit 140 proceeds to step S15.

  (S14) The distribution determination unit 140 refers to the round robin table 111 stored in the storage unit 110, and determines the relay server that is the distribution destination for the business server identification information specified in step S11. .

  (S15) The distribution determination unit 140 updates the USC 112. Specifically, the distribution determination unit 140 updates the count value of each relay server in the USC 112 in accordance with the USC update rule 113 stored in the storage unit 110. In the example of the second embodiment, the distribution determination unit 140 decreases the count value of the relay server selected as the current distribution destination by “2”, and the count value of the relay server that is not selected as the current distribution destination. Is increased by "1".

  (S16) The distribution determination unit 140 updates the round robin table 111. Specifically, the distribution determination unit 140 registers the identification information of the relay server selected at this time in the round robin table 111 with respect to the identification information of the business server specified in step S11.

(S17) The distribution determining unit 140 transfers the received packet P1 to the relay server determined as the distribution destination in step S13 or step S14.
FIG. 10 is a diagram illustrating a specific example of transfer destination transfer according to the second embodiment. As an example, a plurality of packets with destination numbers “A”, “B”, “C”, “D”, “E”, “F”, “G”, “H” are sorted in this order. Assume that the server 100 arrives.

  It is assumed that the count values of the relay servers of the USC 112 before the arrival of the packet whose destination # is “A” are all “0”. 10 is updated every time each packet is transferred by the distribution server 100 (the direction from the upper side to the lower side is the direction in which the count value changes due to the update). Further, as an example, it is assumed that the bias threshold is “4”.

  First, the distribution server 100 receives a packet of destination # “A”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 refers to the round robin table 111 and distributes the packet of the destination # “A” to the relay server 200 (Svr1). The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “0-2 = −2”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “0 + 1 = 1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “0 + 1 = 1”.

  Next, the distribution server 100 receives the packet of the destination # “B”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. Therefore, the distribution server 100 refers to the round robin table 111 and distributes the packet of destination # “B” to the relay server 200a (Svr2). The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−2 + 1 = −1”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “1-2 = −1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “1 + 1 = 2”.

  Next, the distribution server 100 receives the packet of destination # “C”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 refers to the round robin table 111 and distributes the packet of destination # “C” to the relay server 200b (Svr3). The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−1 + 1 = 0”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “−1 + 1 = 0”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “2-2 = 0”.

  Next, the distribution server 100 receives the packet of the destination # “D”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. Therefore, the distribution server 100 refers to the round robin table 111 and distributes the packet of destination # “D” to the relay server 200 (Svr1). The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “0-2 = −2”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “0 + 1 = 1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “0 + 1 = 1”.

  Next, the distribution server 100 receives the packet of destination # “E”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. Therefore, the distribution server 100 refers to the round robin table 111 and distributes the packet of destination # “E” to the relay server 200a (Svr2). The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−2 + 1 = −1”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “1-2 = −1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “1 + 1 = 2”.

  Next, the distribution server 100 receives the packet of destination # “F”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 refers to the round robin table 111 and distributes the packet of destination # “F” to the relay server 200 (Svr1). The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−1-2 = −3”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “−1 + 1 = 0”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “2 + 1 = 3”.

  Next, the distribution server 100 receives the packet of destination # “G”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. Therefore, the distribution server 100 refers to the round robin table 111 and distributes the packet of destination # “G” to the relay server 200a (Svr2). The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−3 + 1 = −2”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “0-2 = −2”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “3 + 1 = 4”.

  Next, distribution server 100 receives the packet of destination # “H”. The distribution server 100 refers to the USC 112 and detects that the count value “4” of the relay server 200b is equal to the bias threshold “4”. For this reason, the distribution server 100 distributes the packet of destination # “H” to the relay server 200b (Svr3) without using the round robin table 111. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−2 + 1 = −1”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “−2 + 1 = −1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “4-2 = 2”.

  Here, it is assumed that the distribution destination by the round robin table 111 is the relay server 200 (“Svr1”) for the current packet of the destination # “H”. As a result, the distribution server 100 transfers the distribution destination of the packet of destination # “H” from the relay server 200 determined based on the round robin table 111 to the relay server 200b determined based on the USC 112. .

  In this way, the distribution server 100 counts down the count value of the USC 112 of the relay server selected as the packet transfer destination, and counts up the count value of the USC 112 of the relay server that has not been selected. For this reason, the distribution server 100 can efficiently identify a node with a relatively small number of times selected as a transfer destination based on the USC 112. The distribution server 100 can determine that there is a bias in the selection of the packet transfer destination by detecting that the count value of the relay server 200b is equal to or greater than the bias threshold. This is because the relatively large number of times that the relay server 200b has not been selected is presumed that the number of times that either or one of the relay servers 200 and 200a has been selected is relatively large. Furthermore, it is considered that the relay server 200b having a relatively large number of times of not being selected has a smaller amount of distributed traffic than the relay servers 200 and 200a. For this reason, it is considered that the resources of the interface connected to the relay server 200b of the SW 400 (for example, the buffer for storing the transfer waiting packet) and the resources of the relay server 200b are relatively free. Accordingly, the distribution server 100 can determine that it is better to preferentially select the relay server 200b when attempting to distribute data to the relay servers 200, 200a, and 200b evenly to some extent.

  As described above, the distribution server 100 can efficiently determine both the fact that the selection of the relay server is biased and the relay server to be selected as the transfer destination in order to correct the bias by the USC 112. In this way, the selection bias of the relay server can be corrected with a small amount of calculation. As a result, the computation associated with the data transfer of the distribution server 100 can be accelerated, and communication between the clients 500, 500a, 500b,..., 500n and the business servers 300, 300a, 300b,. This contributes to a reduction in delay.

  FIG. 11 is a diagram illustrating a comparative example (part 1) with respect to the second embodiment. FIG. 11 shows an example in which distribution servers 600, 600a, 600b and relay servers 700, 700a,..., 700m are connected to SW 400. However, in FIG. 11, “relay server” represents identification information “Svr1”, “Svr2”,..., “Svr10” of the relay servers 700, 700a,. However, it may be written similarly).

  The distribution servers 600, 600a, and 600b use the round robin tables 601, 601a, and 601b to determine packet distribution destinations, but the distribution servers 100, 100a do not hold information corresponding to the USC 112. , 100b.

  For example, when the distribution server 600 determines a packet distribution destination using the round robin table 601, the distribution destination for each destination may be synchronized. In the example of FIG. 11, the distribution destination of packets for each destination is biased toward the relay server 700 (Svr1). Thus, in the operation of the distribution server 600 alone, traffic may be concentrated on a specific relay server 700.

  Furthermore, the distribution destinations by the distribution servers 600, 600a, and 600b may be synchronized. In the example of FIG. 11, the distribution destinations of packets for each destination by the distribution servers 600a and 600b are also biased toward the relay server 700 (Svr1). In this way, traffic may concentrate on the relay server 700 in a burst manner.

  In this case, among the communication interfaces provided in the SW 400, the load on the communication interface that communicates with the relay server 700 is particularly excessive. Then, a delay due to transfer waiting using a buffer included in the communication interface increases, or packet discard due to buffer overflow occurs.

  On the other hand, a method in which round robin is not performed for each destination can be considered. For example, regardless of the destination, it is also possible to distribute the packets in round robin order in the arrival order of the packets.

  FIG. 12 is a diagram illustrating a comparative example (No. 2) with respect to the second embodiment. The distribution servers 610, 610a, and 610b are different from the distribution servers 600, 600a, and 600b in that the packet distribution destinations are determined by round robin in the arrival order of the packets without distinguishing the packet destinations. . In this case, bursty traffic concentration with respect to the relay servers 700, 700a, 700b,..., 700m subsequent to the distribution servers 610, 610a, 610b is improved. On the other hand, since ECMP cannot be performed for each destination, when a plurality of hops by virtual routers (a set of relay servers) are stacked, a packet of a specific destination may be biased to a route via a specific relay server.

  For example, in the example of FIG. 12, three virtual routers are shown. The first virtual router is a set of relay servers to which the relay servers 700, 700a, 700b,. The second virtual router is a set of relay servers to which the relay servers 800, 800a, 800b,. The third virtual router is a set of relay servers to which the relay servers 900, 900a, 900b,. The relay server identification information is unique within one virtual router, and each virtual router is relayed with identification information “Svr1”, “Svr2”, “Svr3”,..., “Svr10”. Includes servers. A certain relay server transfers a packet to a subsequent relay server having the same identification information as that of the relay server. The third virtual router is a virtual router subsequent to the first virtual router and the second virtual router.

  In this case, since the ECMP for each destination cannot be performed in the method of the comparative example in FIG. 12, there is a possibility that the transfer path of the packet of the specific destination is biased to the route via the relay server 900, for example. As a result, when the load on the relay server 900 increases or when a failure or the like occurs in the relay server 900, the influence on communication targeting the corresponding destination becomes excessive. As described above, it is not preferable that ECMP cannot be performed for each destination.

  On the other hand, according to the distribution servers 100, 100a, and 100b of the second embodiment, the individual distribution servers 100, 100a, and 100b correct the distribution destination bias. Therefore, the specific relay illustrated in FIG. The occurrence of bursty traffic concentration on the server can be suppressed. In addition, the distribution servers 100, 100a, and 100b can implement ECMP for each destination by performing round robin for each destination, and can suppress the deviation of the packet transfer path for each destination illustrated in FIG.

  Moreover, in order to detect the selection bias of the distribution destination, each of the distribution servers 100, 100a, and 100b does not need to communicate with other servers. That is, each of the distribution servers 100, 100a, and 100b can detect the distribution destination bias alone.

  In the second embodiment, round robin is exemplified as a rule for selecting an allocation destination with equal probability, but other methods may be used. For example, the distribution determination unit 140 may determine the distribution destination by a predetermined calculation using random numbers.

  In the second embodiment, the distribution servers 100, 100a, 100b and the relay servers 200, 200a, 200b are realized by a physical computer (physical machine). On the other hand, you may implement | achieve both or any one of distribution server 100,100a, 100b and relay server 200,200a, 200b with the virtual computer (virtual machine) which operate | moves on a physical machine.

[Third Embodiment]
Hereinafter, a third embodiment will be described. Items that differ from the second embodiment described above will be mainly described, and descriptions of common items will be omitted.

  Here, the information processing system and hardware of the third embodiment are the same as the information processing system and hardware of the second embodiment illustrated in FIGS. The function of the distribution server according to the third embodiment is the same as the function of the distribution server 100 according to the second embodiment illustrated in FIG. For this reason, in the third embodiment, each element is indicated by using the same name and symbol as those in the second embodiment. In addition, the communication protocol used in the information processing system of the third embodiment is the same as the communication protocol exemplified in the second embodiment.

  However, in the third embodiment, the contents of the business service provided by the business servers 300, 300a, 300b,..., 300m are different from those in the second embodiment. Specifically, in the third embodiment, the business servers 300, 300a, 300b,..., 300m provide a stateful (communication mode for maintaining a session state) network function such as a firewall or a proxy. In this case, a relay server that relays communication between a certain client and a certain business server is uniquely associated with one communication session. Therefore, in the third embodiment, the distribution determination unit 140 of the distribution server 100 determines the relay server that is the distribution destination of the packet based on the hash value calculated using the header information of the packet. . This point is different from the second embodiment.

  FIG. 13 is a diagram illustrating an example of a packet according to the third embodiment. The packet P1 includes fields of a payload f1, a TCP / UDP header f2, an IP header f3, and a MAC header f4. The TCP / UDP header f2 includes a destination port number (D-Port: Destination Port) and a source port number (S-Port: Source Port). The IP header f3 includes a destination IP address (D-IP), a source IP address (S-IP), and a protocol number (Protocol).

  The destination port number, source port number, destination IP address, source IP address, and protocol number may be collectively referred to as a 5-tuple. It can also be said that the 5-tuple is information used for flow identification.

  The distribution determination unit 140 performs a hash operation using a 5-tuple value. The distribution determination unit 140 determines the distribution destination of the packet P1 based on the hash value obtained by the hash calculation. For example, the obtained hash value may match the identification information (identification number) of any relay server. In that case, the distribution determination unit 140 can determine the relay server corresponding to the obtained hash value as the distribution destination. The distribution determination unit 140 sets the MAC address of the relay server that is the distribution destination to the D-MAC (not shown in FIG. 13) of the MAC header f4 and transmits it to the SW 400, so that the corresponding relay server Packet P1 is forwarded to.

  Here, in the third embodiment, the distribution server 100 stores a session management table instead of the round robin table 111. The session management table is information for managing a flow and a relay server as a distribution destination.

  FIG. 14 illustrates an example of a session management table according to the third embodiment. The session management table 114 is stored in the storage unit 110. The session management table 114 includes items of flow #, Svr #, and flag.

  Flow identification information is registered in the flow # item. The flow identification information may be information representing the 5-tuple value set itself in the packet header information, or may be predetermined information associated with the 5-tuple value set. In the item of Svr #, the identification information of the relay server that is the distribution destination of the packet of the corresponding flow is registered. In the flag item, a flag indicating whether or not the USC 112 has determined the distribution destination for the packet of the corresponding flow is registered. When the assignment destination is determined by the USC 112, the flag is “True”. When the allocation destination is not determined by the USC 112, the flag is “False”.

  For example, in the session management table 114, a record in which the flow # is “a”, the Svr # is “1”, and the flag is “False” is registered. This indicates that the flow distribution destination of the flow # “a” is the relay server 200 (Svr1), and the distribution destination is determined based on the hash value of the 5-tuple of the packet.

  Further, for example, in the session management table 114, a record having a flow # of “h”, Svr # of “3”, and a flag of “True” is registered. This indicates that the flow distribution destination of the flow # “h” is the relay server 200b (Svr3), and the distribution destination is determined based on the USC 112.

In the session management table 114, the identification information and flags of the relay servers that are the distribution destinations are similarly set for the other flows.
Next, the processing procedure of the distribution server 100 according to the third embodiment will be described (the same procedure can be executed by the distribution servers 100a and 100b).

FIG. 15 is a flowchart illustrating an example of packet transfer according to the third embodiment. In the following, the process illustrated in FIG. 15 will be described in order of step number.
(S21) The identification unit 120 receives the packet P1 transmitted by any client.

  (S22) The identification unit 120 identifies a flow based on the 5-tuple information set of the packet P1, and acquires flow identification information (flow #) for the packet P1. Further, the identification unit 120 performs a hash operation using information of five tuples and calculates a hash value.

  (S23) The bias determination unit 130 refers to the USC 112 stored in the storage unit 110 and determines whether there is a relay server in which the count value of the USC 112 is equal to or greater than the bias threshold. If there is a relay server whose USC 112 count value is greater than or equal to the bias threshold, the bias determination unit 130 proceeds to step S24. If there is no relay server with a count value of USC 112 equal to or greater than the bias threshold, the bias determination unit 130 proceeds to step S27.

  (S24) The distribution determination unit 140 determines whether or not the current packet P1 is a packet for a new session. If the packet is a new session, the distribution determination unit 140 proceeds to step S25. If it is not a packet of a new session, the distribution determination unit 140 proceeds to step S27.

  Here, the distribution determination unit 140 refers to the session management table 114 stored in the storage unit 110 to perform the determination in step S24. Specifically, the distribution determining unit 140 determines whether or not a record corresponding to the flow identification information acquired in step S22 has been registered in the session management table 114. If registered, the distribution determining unit 140 determines that the session is not a new session (that is, an existing session). If not registered, the distribution determination unit 140 determines that the session is a new session.

  (S25) The distribution determination unit 140 determines, as the distribution destination, the relay server that has been determined in step S23 that the count value of the USC 112 is equal to or greater than the bias threshold. The process of step S25 is equivalent to performing transfer of the distribution destination.

(S26) The distribution determination unit 140 sets a flag corresponding to the flow identification information acquired this time to “True”. Then, the distribution determining unit 140 proceeds to step S30.
(S27) The distribution determination unit 140 refers to the session management table 114 and determines whether or not the flag of the record corresponding to the flow identification information acquired in step S22 is “True”. In the case of “True”, the distribution determination unit 140 proceeds with the process to step S28. If it is not “True”, the distribution determination unit 140 proceeds to step S29. If Step S23 is No, a record corresponding to the identification information of the corresponding flow may not be registered in the session management table 114 (when the current packet P1 is a packet of a new session). In this case, the distribution determining unit 140 determines that “True” is not satisfied in step S27.

  (S28) The distribution determination unit 140 determines a distribution destination based on the flag. Specifically, the distribution determining unit 140 acquires Svr # set in the record of the session management table 114 referred to in step S27. The distribution determination unit 140 determines the relay server corresponding to the Svr # as the distribution destination of the current packet P1. Then, the distribution determining unit 140 proceeds to step S30. The process of step S28 is equivalent to continuing the transfer of the distribution destination.

(S29) The distribution determination unit 140 determines the relay server corresponding to the hash value obtained in step S22 as the distribution destination of the current packet P1.
(S30) The distribution determination unit 140 updates the USC 112. Specifically, the distribution determination unit 140 updates the count value of each relay server in the USC 112 in accordance with the USC update rule 113 stored in the storage unit 110. In the example of the third embodiment, the distribution determination unit 140 decreases the count value of the relay server selected as the current distribution destination by “2”, and the count value of the relay server that is not selected as the current distribution destination. Is increased by "1".

(S31) The distribution determination unit 140 updates the session management table 114. Specifically, the distribution determination unit 140 performs the following processing.
When step S <b> 26 is executed for the current packet P <b> 1, the distribution determination unit 140 adds a record relating to the current flow of the new session to the session management table 114. The set value of the flow # of the record to be added becomes the flow identification information acquired in step S22. The set value of Svr # of the record becomes the identification information of the relay server that is the distribution destination determined in step S25. The set value of the flag of the record is “True”.

  When step S <b> 29 is executed for the current packet P <b> 1, the distribution determination unit 140 registers a record relating to the current flow in the session management table 114. At this time, the distribution determination unit 140 may (1) newly add a record related to the current flow, or (2) update the Svr # setting value of an existing record related to the current flow. is there.

  Case (1) is a case where step S23 No, step S27 No, and the current packet P1 is a packet of a new session. In this case, the setting value of the flow # of the record added to the session management table 114 becomes the flow identification information acquired in step S22. The set value of Svr # of the record becomes the identification information of the relay server that is the distribution destination determined in step S29. The set value of the flag of the record is “False”.

  Case (2) is a case where step S23 No, step S27 No, and the current packet P1 is not a packet of a new session. In this case, the allocating unit 140 uses the existing record of the session management table 114 corresponding to the flow identification information acquired in step S22 to obtain the identification information of the relay server that is the distribution destination determined in step S29. Set to Svr #.

When step S28 is executed for the current packet P1, the allocation determination unit 140 does not need to update the session management table 114.
(S32) The distribution determination unit 140 transfers the received packet P1 to the relay server determined as the distribution destination in step S25 or step S29.

  Further, the distribution determination unit 140 monitors the communication idle time (the time during which the packet of the corresponding flow is not transmitted / received) in the flow in which the flag “True” is set in the session management table 114, and the transfer destination The process of canceling the transfer is also performed.

FIG. 16 is a flowchart illustrating an example of canceling transfer destination transfer according to the third embodiment. In the following, the process illustrated in FIG. 16 will be described in order of step number.
(S41) In the session management table 114 stored in the storage unit 110, the distribution determination unit 140 monitors the idle time of communication related to the flow for which the flag = True is set.

  (S42) The distribution determination unit 140 determines whether or not the idle time of communication related to the flow has reached a predetermined time. When the idle time has reached the predetermined time, the distribution determination unit 140 proceeds to step S43. If the idle time has not reached the predetermined time, the distribution determination unit 140 proceeds to step S41 (continues monitoring). The length of the predetermined time used for the determination in step S42 can be arbitrarily determined (for example, it may be 1 minute or about several minutes).

(S43) In the session management table 114, the distribution determination unit 140 sets a flag for the corresponding flow to “False”.
As described above, when the distribution determination unit 140 selects a relay server that is a distribution destination (transfer destination) for a specific flow regardless of the hash value, the distribution determination unit 140 allocates a packet corresponding to the specific flow for a predetermined period. The destination is the selected relay server. As a result, stateful business services such as firewalls and proxies can be appropriately provided to clients.

  In the example of FIG. 16, the idle time of communication is monitored, but other methods are also conceivable. For example, the distribution determination unit 140 may set the flag for the corresponding flow to “False” when detecting the disconnection of the session by the FIN packet or the RST packet in TCP for the flow of the flag “True”.

  FIG. 17 is a diagram illustrating a specific example of transfer destination transfer according to the third embodiment. As an example, a plurality of packets having flow numbers “a”, “b”, “c”, “d”, “e”, “f”, “g”, and “h” are sorted in this order. Assume that the server 100 arrives.

  It is assumed that the count values of the relay servers of the USC 112 before the arrival of the packet whose destination # is “a” are all “0”. 17 is updated every time each packet is transferred by the distribution server 100 (the direction from the upper side to the lower side is the direction in which the count value changes due to the update). Further, as an example, it is assumed that the bias threshold is “4”.

  First, the distribution server 100 receives the packet “a1” of the flow # “a”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 distributes the packet “a1” to the relay server 200 (Svr1) based on the hash value of the header information of the packet “a1”. The distribution server 100 sets the flag of the record of the flow # “a” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “0-2 = −2”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “0 + 1 = 1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “0 + 1 = 1”.

  Next, the distribution server 100 receives the packet “b1” of the flow # “b”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 distributes the packet “b1” to the relay server 200a (Svr2) based on the hash value of the header information of the packet “b1”. The distribution server 100 sets the flag of the record of the flow # “b” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−2 + 1 = −1”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “1-2 = −1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “1 + 1 = 2”.

  Next, the distribution server 100 receives the packet “c1” of the flow # “c”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 distributes the packet “c1” to the relay server 200b (Svr3) based on the hash value of the header information of the packet “c1”. The distribution server 100 sets the flag of the record of the flow # “c” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−1 + 1 = 0”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “−1 + 1 = 0”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “2-2 = 0”.

  Next, the distribution server 100 receives the packet “d1” of the flow # “d”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 distributes the packet “d1” to the relay server 200 (Svr1) based on the hash value of the header information of the packet “d1”. The distribution server 100 sets the flag of the record of the flow # “d” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “0-2 = −2”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “0 + 1 = 1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “0 + 1 = 1”.

  Next, the distribution server 100 receives the packet “e1” of the flow # “e”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 distributes the packet “e1” to the relay server 200a (Svr2) based on the hash value of the header information of the packet “e1”. The distribution server 100 sets the flag of the record of the flow # “e” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−2 + 1 = −1”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “1-2 = −1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “1 + 1 = 2”.

  Next, the distribution server 100 receives the packet “f1” of the flow # “f”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 distributes the packet “f1” to the relay server 200 (Svr1) based on the hash value of the header information of the packet “f1”. The distribution server 100 sets the flag of the record of the flow # “f” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−1-2 = −3”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “−1 + 1 = 0”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “2 + 1 = 3”.

  Next, the distribution server 100 receives the packet “g1” of the flow # “g”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. For this reason, the distribution server 100 distributes the packet “g1” to the relay server 200a (Svr2) based on the hash value of the header information of the packet “g1”. The distribution server 100 sets the flag of the record of the flow # “g” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−3 + 1 = −2”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “0-2 = −2”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “3 + 1 = 4”.

  Next, the distribution server 100 receives the packet “a2” of the flow # “a”. The distribution server 100 refers to the USC 112 and confirms that there is a count value that has reached the bias threshold “4”. On the other hand, the packet “a2” is a packet of an existing session, and the flag of the flow # “a” in the session management table 114 is “False”. Therefore, the distribution server 100 distributes the packet “a2” to the relay server 200 (Svr1) based on the hash value of the header information of the packet “a2”. The distribution server 100 sets the flag of the record of the flow # “a” in the session management table 114 to “False”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−2-2 = −4”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “−2 + 1 = −1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “4 + 1 = 5”.

  Next, the distribution server 100 receives the packet “h1” of the flow # “h”. The distribution server 100 refers to the USC 112 and confirms that there is a count value that has reached the bias threshold “4”. The packet “h1” is a packet for a new session. Therefore, the distribution server 100 distributes the packet “h1” to the relay server 200b whose count value reaches the bias threshold “4” based on the USC 112. The distribution server 100 sets Svr # of the record of the flow # “h” in the session management table 114 to “3” and the flag to “True”. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−4 + 1 = −3”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “−1 + 1 = 0”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “5-2 = 3”.

  Here, it is assumed that the distribution destination based on the hash value of the header information of the packet “h1” is the relay server 200 (“Svr1”). Then, the distribution server 100 transfers the distribution destination of the packet “h1” from the relay server 200 determined based on the hash value to the relay server 200b determined based on the USC 112.

  Next, the distribution server 100 receives the packet “h2” of the flow # “h”. The distribution server 100 refers to the USC 112 and confirms that there is no count value that has reached the bias threshold “4”. On the other hand, the packet “h2” is a packet of an existing session, and the flag of the flow # “h” in the session management table 114 is “True”. Therefore, the distribution server 100 distributes the packet “h2” to the relay server 200b based on the session management table 114. The distribution server 100 updates the count value of the relay server 200 in the USC 112 to “−3 + 1 = −2”. The distribution server 100 updates the count value of the relay server 200a in the USC 112 to “0 + 1 = 1”. The distribution server 100 updates the count value of the relay server 200b in the USC 112 to “3-2 = 1”.

  Here, the distribution destination based on the hash value of the header information of the packet “h2” is the relay server 200 (“Svr1”), similarly to the packet “h1”. For this reason, the distribution server 100 transfers the distribution destination of the packet “h2” from the relay server 200 determined based on the hash value to the relay server 200b determined based on the USC 112.

  In this way, the distribution server 100 counts down the count value of the USC 112 of the relay server selected as the packet transfer destination, and counts up the count value of the USC 112 of the relay server that has not been selected. For this reason, the distribution server 100 can efficiently identify a node with a relatively small number of times selected as a transfer destination based on the USC 112. The distribution server 100 can determine that there is a bias in the selection of the packet transfer destination by detecting that the count value of the relay server 200b is equal to or greater than the bias threshold. This is because the relatively large number of times that the relay server 200b has not been selected is presumed that the number of times that either or one of the relay servers 200 and 200a has been selected is relatively large. Furthermore, it is considered that the relay server 200b having a relatively large number of times of not being selected has a smaller amount of distributed traffic than the relay servers 200 and 200a. For this reason, it is considered that the resources of the interface connected to the relay server 200b of the SW 400 (for example, the buffer for storing the transfer waiting packet) and the resources of the relay server 200b are relatively free. Accordingly, the distribution server 100 can determine that it is better to preferentially select the relay server 200b when attempting to distribute data to the relay servers 200, 200a, and 200b evenly to some extent.

  As described above, the distribution server 100 can efficiently determine both the fact that the selection of the relay server is biased and the relay server to be selected as the transfer destination in order to correct the bias by the USC 112. In this way, the selection bias of the relay server can be corrected with a small amount of calculation. As a result, the computation associated with the data transfer of the distribution server 100 can be accelerated, and communication between the clients 500, 500a, 500b,..., 500n and the business servers 300, 300a, 300b,. This contributes to a reduction in delay.

  FIG. 18 is a diagram illustrating a comparative example with respect to the third embodiment. FIG. 18 shows an example in which distribution servers 620, 620a, 620b and relay servers 700, 700a,..., 700m are connected to the SW 400. However, in FIG. 18, “relay server” represents identification information “Svr1”, “Svr2”,..., “Svr10” of the relay servers 700, 700a,.

  The distribution servers 600, 600a, and 600b determine packet distribution destinations using the session management tables 621, 621a, and 621b, respectively. The session management tables 621, 621a, and 621b include items of flow # and Svr #, but are different from the session management table 114 in that they do not include a flag item. The distribution servers 620, 620a, and 620b are different from the distribution servers 100, 100a, and 100b in that they do not hold information corresponding to the USC 112.

  For example, the distribution server 620 may receive a plurality of flow packets passing through the relay server 700 during a relatively short period. That is, packet distribution destinations for a plurality of flows may be biased toward the relay server 700 (Svr1).

  Furthermore, the distribution destinations by the distribution servers 620, 620a, and 620b may be synchronized. In the example of FIG. 18, the distribution destination of packets for each flow by the distribution servers 620a and 620b is also biased toward the relay server 700 (Svr1). In this way, traffic may concentrate on the relay server 700 in a burst manner.

  In this case, among the communication interfaces provided in the SW 400, the load on the communication interface that communicates with the relay server 700 is particularly excessive. Then, a delay due to transfer waiting using a buffer included in the communication interface increases, or packet discard due to buffer overflow occurs.

  On the other hand, according to the distribution servers 100, 100a, and 100b of the third embodiment, the individual distribution servers 100, 100a, and 100b correct the distribution destination bias. Therefore, the specific relay illustrated in FIG. The occurrence of bursty traffic concentration on the server can be suppressed.

  Moreover, in order to detect the selection bias of the distribution destination, each of the distribution servers 100, 100a, and 100b does not need to communicate with other servers. That is, each of the distribution servers 100, 100a, and 100b can detect the distribution destination bias alone.

  In the third embodiment, the distribution server 100, 100a, 100b and / or the relay server 200, 200a, 200b are realized by a virtual computer (virtual machine) that operates on a physical machine. May be.

  By the way, when using a plurality of distribution servers such as the distribution servers 100, 100a, 100b, there may be a plurality of relay servers in which the count value of the USC has reached the bias threshold in two or more distribution servers. In this case, if the two or more distribution servers all select the same relay server as the transfer destination, the load of resources for transferring the packet to the transfer destination relay server may be excessive. For this reason, each distribution server may control the selection of the transfer destination so that it is different from other distribution servers. For example, the following method can be considered.

  FIG. 19 is a diagram illustrating another example (part 1) of transfer destination transfer. The distribution server 100a stores the USC 112a. The direction from the upper side to the lower side in FIG. 19 represents the direction of update transition of USCs 112 and 112a. Note that the initial value of the count value for each relay server of USC 112 and 112a is “0”. In FIG. 19, as an example, it is assumed that the distribution destination of packets from the first time to the fourth time by the distribution servers 100 and 100 a is the relay server 200. Further, the bias threshold is “4”.

  In this case, in the fourth distribution, the count value of the relay server 200 in the USCs 112 and 112a is “−8”, the count value of the relay server 200a is “4”, and the count value of the relay server 200b is “4”. At this stage, the count values of both relay servers 200a and 200b reach the bias threshold “4”.

  At this time, the distribution server 100 preferentially selects the smaller Svr # as the distribution destination (priority number priority). On the other hand, the distribution server 100a preferentially selects the one having a larger Svr # as the distribution destination (priority number priority). In the above example, the distribution server 100 sets the distribution destination of the next packet as the relay server 200a (Svr2). On the other hand, the distribution server 100a sets the distribution destination of the next packet as the relay server 200b (Svr3). Note that the distribution server that selects the allocation destination with the priority of the younger number and the distribution server that selects the allocation destination with the priority of the older number are determined in advance by the system administrator. Each distribution server may cooperate with the distribution servers to determine different priorities.

  As described above, when there are a plurality of relay servers whose count values reach the bias threshold, the distribution servers 100 and 100a may select a distribution destination based on a predetermined priority for each of the plurality of relay servers. . At this time, the priority of the relay server used by a certain distribution server and the priority of the relay server used by another distribution server are set to have different priorities. Thereby, when both of the distribution servers 100 and 100a perform distribution destination transfer, it is possible to suppress the concentration of traffic on either the relay server 200a or the relay server 200b.

  Alternatively, each of the distribution servers 100 and 100a may change the added value for updating the count values of the relay servers 200, 200a, and 200b for each relay server. Specifically, it is as follows.

  FIG. 20 is a diagram illustrating another example (part 2) of transfer destination transfer. The direction from the upper side to the lower side in FIG. 20 represents the direction of update transition of USCs 112 and 112a. Note that the initial value of the count value for each relay server of USC 112 and 112a is “0”. In FIG. 20, as an example, it is assumed that the distribution destination of packets from the first time to the fourth time by the distribution servers 100 and 100 a is the relay server 200. Further, the bias threshold is “4”.

  Here, the distribution server 100 sets “1.1 (= 1.0 + 0.1)” as the addition value of the count value of the relay server 200a in the USC 112. On the other hand, the distribution server 100 sets the added value of the count value of the relay server 200b in the USC 112 to “0.9 (= 1.0−0.1)”.

  Further, the distribution server 100a sets the added value of the count value of the relay server 200a in the USC 112a to “0.9 (= 1.0−0.1)”. On the other hand, the distribution server 100a sets the added value of the count value of the relay server 200b in the USC 112a to “1.1 (= 1.0 + 0.1)”.

  Then, in the fourth distribution, the count value of the relay server 200 in the USC 112 is “−8”, the count value of the relay server 200a is “4.4”, and the count value of the relay server 200b is “3.6”. At this stage, the count value of the relay server 200a reaches the bias threshold “4”.

  On the other hand, in the fourth distribution, the count value of the relay server 200 in the USC 112a is “−8”, the count value of the relay server 200a is “3.6”, and the count value of the relay server 200b is “4.4”. At this stage, the count value of the relay server 200b reaches the bias threshold “4”.

  Therefore, the distribution server 100 sets the next packet distribution destination as the relay server 200a (Svr2). On the other hand, the distribution server 100a sets the distribution destination of the next packet as the relay server 200b (Svr3). The added value for each relay server is determined in advance by the system administrator. Each distribution server may cooperate with the distribution servers and determine different addition values.

  As described above, the distribution servers 100 and 100a may set different addition values for the count value of each relay server for each relay server. For example, if the weight of the addition value of the first relay server by the distribution server 100 is increased and the weight of the addition value of the second relay server is decreased, the addition value of the first relay server by the distribution server 100a And the weight of the added value of the second relay server is increased. Thereby, when both of the distribution servers 100 and 100a perform distribution destination transfer, it is possible to suppress the concentration of traffic on either the relay server 200a or the relay server 200b.

  Note that the methods of FIGS. 19 and 20 can be applied to both the second and third embodiments. It is also conceivable to apply both the methods of FIGS. 19 and 20 to the second and third embodiments.

  By the way, in 2nd, 3rd embodiment, the frequency | count which was not selected as a distribution destination was counted using USC112. On the other hand, as a simple method of detecting the distribution destination bias, it may be possible to count the number of times of accumulation selected as the distribution destination for each relay server. Thus, for example, consider that the distribution server 100 determines a packet distribution destination using a distribution accumulation counter instead of the USC 112.

  FIG. 21 is a diagram illustrating an example of a distribution accumulation counter. The distribution accumulation counter 115 is a counter that counts the number of accumulations selected as a distribution destination for each relay server. For example, the distribution server 100 uses the distribution accumulation counter 115 to increment the count value of the relay server selected as the packet distribution destination by one each time it is selected. At this time, the distribution server 100 does not update the count value of the relay server that is not selected as the distribution destination.

  In this case, in order to determine whether or not the selection of the distribution destination is biased, it is considered to obtain the minimum value of the count value of each relay server and the magnitude of the difference between the minimum value and the maximum value. It is done. In this case, when the difference is larger than a predetermined threshold value, it can be determined that there is a bias in selection of the distribution destination. In addition, when there is a bias, it is conceivable to select the relay server corresponding to the minimum value of the count value as the distribution destination.

  In order to obtain the minimum value and the maximum value of the count value in the distribution cumulative counter 115, it is conceivable to compare the count values of each relay server with respect to a set of two relay servers. However, in this method, the amount of computation for comparison increases linearly as the number of distribution destination candidate relay servers increases. For this reason, as the number of relay servers increases, the amount of delay until the allocation destination is determined may become excessive.

  FIG. 22 is a diagram illustrating a comparative example of the distribution accumulation counter and the USC. FIG. 22A shows the transition of the update of the distribution cumulative counter 115. FIG. 22B shows the transition of the update of the USC 112a when the bias threshold> 7 (denoted as USC 112a to distinguish it from the USC 112 in FIG. 22C). FIG. 22C shows the transition of USC 112 updates when the bias threshold = 4. The direction from the upper side to the lower side in FIG. 22 is the direction of the update transition.

  Here, it is assumed that the relay servers 200, 200a, and 200b are distribution destination candidates. In all of the three cases illustrated in FIG. 22, it is assumed that the count value of each relay server is initially “0”. In all three cases, the relay server selection order is the same from the first to the seventh selection. The first selection from the first time is relay server 200 (Svr1), relay server 200a (Svr2), relay server 200b (Svr3), relay server 200 (Svr1), relay server 200a (Svr2), relay server 200, (Svr1). Assume that the relay server 200a (Svr2) is in this order.

  In the case of FIG. 22A, each time a distribution destination is selected, it is determined based on the distribution accumulation counter 115 whether the selection of the distribution destination is biased. Further, according to the selection result, the distribution cumulative counter 115 counts up the count value of the relay server selected as the distribution destination. For example, in the eighth, ninth, and tenth selections, the relay server 200 (Svr1), the relay server 200a (Svr2), and the relay server 200 (Svr1) are selected, respectively. In the distribution accumulation counter 115 after the tenth selection, the count value of the relay server 200 is “5”. The count value of the relay server 200a is “4”. The count value of the relay server 200b is “1”. For example, when it is determined whether or not there is a bias in the selection of the distribution destination after the 10th selection, the count value of each relay server of the distribution accumulation counter 115 is compared to the round robin so that the maximum number of selections can be obtained. It is determined that there is a difference of 4. In addition, if the difference is greater than the allowable difference, for example, the relay server 200b (Svr3) having the smallest count value of the distribution cumulative counter 115 is selected as the distribution destination.

  In the case of FIG. 22B, each time a distribution destination is selected, it is determined whether there is a bias in the selection of the distribution destination based on the USC 112a. Further, according to the selection result, the count value of each relay server of the USC 112a is updated according to the USC update rule 113. For example, in the eighth, ninth, and tenth selections, the relay server 200 (Svr1), the relay server 200a (Svr2), and the relay server 200 (Svr1) are selected, respectively. In the USC 112a after the tenth selection, the count value of the relay server 200 is “−5”. The count value of the relay server 200a is “−2”. The count value of the relay server 200b is “7”.

  The count value is updated for every relay server every time it is selected. However, the count value of each relay server in the USC 112a can be updated independently, and parallel calculation is possible. For example, when the processor 101 is a multi-core, it is possible to update the count values of all the relay servers in one cycle by updating in parallel with a plurality of cores. Therefore, it can be considered that the processing delay due to the update for all the relay servers is very small.

  In addition, it is only necessary to compare the count value of each relay server with the bias threshold value, without comparing the brute force of the count values of each relay server. That's it. For example, assuming that the number of relay servers is 10, when the distribution cumulative counter 115 is used, 10 cycles of comparison operations are required to identify the relay server with the smallest number of distributions. Therefore, it is not necessary to perform the comparison operation. In this case, as described above, the count value of the USC 112 for 10 relay servers is updated, but the update itself can be performed in parallel (for example, can be updated in one cycle). The same applies to the case of FIG.

  In the case of FIG. 22C as well, each time a distribution destination is selected, it is determined whether there is a bias in selection of the distribution destination based on the USC 112. Further, according to the selection result, the count value of each relay server of the USC 112 is updated according to the USC update rule 113. For example, in the eighth selection, since the count value of the relay server 200b (Svr3) has reached the bias threshold, the relay server 200b is selected as the distribution destination instead of the relay server 200. Subsequently, in the ninth and tenth selections, the relay server 200a (Svr2) and the relay server 200 (Svr1) are selected, respectively. In the USC 112 after the tenth selection, the count value of the relay server 200 is “−2”. The count value of the relay server 200a is “−2”. The count value of the relay server 200b is “4”.

  As described above, the distribution servers 100, 100a, and 100b have both the fact that the selection of the node is biased and the node that should be selected as the transfer destination in order to correct the bias of the USC 112 (or USC 112a). It can be determined efficiently by the count value. In particular, as in the comparative example using the distribution cumulative counter 115, the amount of calculation can be reduced as compared with the method of detecting the bias by comparing the cumulative count values between the relay servers by brute force.

  Thus, according to the distribution servers 100, 100a, and 100b, it is possible to correct the selection bias of the relay server that is the distribution destination with a small amount of calculation. As a result, it is possible to speed up computations associated with data transfer of the distribution servers 100, 100a, 100b, and the clients 500, 500a, 500b,..., 500n and the business servers 300, 300a, 300b,. It contributes to the reduction of communication delays.

  In addition, as a normal distribution rule by the distribution servers 100, 100a, and 100b, a method of selecting a distribution destination by a round robin or a hash value of header information of a packet is illustrated, but a normal distribution rule is Other than that. For example, the distribution server 100 acquires a load on the CPU and memory of each of the relay servers 200, 200a, and 200b as a normal distribution rule, and determines a packet distribution destination according to the load. May be adopted. In this case, it is conceivable that the distribution server 100 determines the relay server with the minimum load as the distribution destination. The same applies to the distribution servers 100a and 100b.

  The information processing according to the first embodiment can be realized by causing the processing unit 1b to execute a program. The information processing according to the second and third embodiments can be realized by causing the processor 101 to execute a program. The program can be recorded on a computer-readable recording medium 13.

  For example, the program can be distributed by distributing the recording medium 13 on which the program is recorded. Alternatively, the program may be stored in another computer and distributed via a network. For example, the computer stores (installs) a program recorded in the recording medium 13 or a program received from another computer in a storage device such as the RAM 102 or the HDD 103, and reads and executes the program from the storage device. Good.

DESCRIPTION OF SYMBOLS 1 Control apparatus 1a Memory | storage part 1b Processing part 2, 3, 4 Node 5, 6, 7 Destination apparatus 8, 9 Terminal apparatus N1, N2 Network T1, T1a Table

Claims (10)

In a control device that selects and transfers one of a plurality of nodes when transferring data,
A plurality of numerical values relating to the number of times each of the plurality of nodes has been selected, the first numerical value corresponding to the first node and the second numerical value corresponding to the second node being stored. A storage unit to
When the first node is selected as the transfer destination of the first data, the first numerical value is decreased, the second numerical value corresponding to the second node not selected is increased, and the second node A processing unit that selects the second node as a transfer destination of the second data when the numerical value of
Control device.   The processing unit selects the first node as a transfer destination of the first data according to a predetermined rule, and when the second numerical value is equal to or greater than the predetermined value, The control device according to claim 1, wherein the second node is selected as a transfer destination of second data.   The control device according to claim 2, wherein the predetermined rule is a rule for selecting a transfer destination with equal probability for each data destination.   The control device according to claim 2, wherein the predetermined rule is a rule for selecting a transfer destination according to a flow identified by header information of a packet used for data transmission.   When the processing unit selects the second node as a transfer destination without depending on the predetermined rule for a specific flow, the processing unit sets the transfer destination of the data corresponding to the specific flow for the predetermined period. The control device according to claim 4, wherein two nodes are used.   6. The processing unit according to claim 1, wherein when the first node is selected as a transfer destination, the processing unit increases the plurality of second numerical values corresponding to the plurality of second nodes that are not selected. The control device according to item 1.   The control according to claim 6, wherein the processing unit selects a transfer destination of the second data based on a priority for each of the plurality of nodes when a plurality of the second numerical values are equal to or greater than the predetermined value. apparatus.   The control device according to claim 1, wherein the processing unit sets a value added to each of the plurality of numerical values to be a different value for each node. In a control program used for processing to select and transfer one of a plurality of nodes when transferring data,
A plurality of numerical values relating to the number of times each of the plurality of nodes has been selected, the first numerical value corresponding to the first node and the second numerical value corresponding to the second node being stored. And
When the first node is selected as the transfer destination of the first data, the first numerical value is decreased, and the second numerical value corresponding to the second node not selected is increased,
When the second numerical value is equal to or greater than a predetermined value, the second node is selected as a second data transfer destination.
A control program that causes a computer to execute processing. A computer that selects and transfers one of a plurality of nodes when transferring data.
A plurality of numerical values relating to the number of times each of the plurality of nodes has been selected, the first numerical value corresponding to the first node and the second numerical value corresponding to the second node being stored. And
When the first node is selected as the transfer destination of the first data, the first numerical value is decreased, and the second numerical value corresponding to the second node not selected is increased,
When the second numerical value is equal to or greater than a predetermined value, the second node is selected as a second data transfer destination.
Control method.

Images