JP4683345B2 - Network load balancing apparatus, network load balancing method, and program - Google Patents

Description

  The present invention relates to a network load distribution apparatus, a network load distribution method, and a program for distributing a load by distributing access from a client to a plurality of servers.

In a network system that accesses a server from a client (terminal) on a network, a network load balancer (load balancer) that distributes load by distributing many accesses from the client to a plurality of servers is installed. . In this type of network load balancer, a NAT (Network Address Translation) method and a MAT (MAC Address) are used as load balancing algorithms for selecting an optimum server from among available servers.
Translation: MAC address conversion) is widely used.

In the NAT system, the load balancer performs network address conversion between a virtual server that is a data destination of the client and the selected server. However, in general, network address translation uses NAPT (Network Address Port Translation) which also performs port number translation, which complicates the apparatus, and ICMP (Internet)
Control Message Protocol: Internet control notification protocol) packet or Raw (raw) IP packet cannot be transferred, and there is a problem that communication from the server side to which the load is distributed cannot be transmitted to the client side.

  On the other hand, the MAT method is widely used for multihoming (router load balancing) of Internet connection and firewall load balancing, and selects a device to be connected by converting the destination MAC address without changing the destination IP address. However, in the case of this MAT method, it is necessary to register the IP address of the virtual server and the MAC address of the corresponding server group in advance in the load balancer, and ARP (Address Resolution Protocol) from the client. It is necessary to make a setting for each server so as not to return an ARP response to the request.

  FIG. 7 shows a general configuration of a load distribution apparatus using the MAT method. In the MAT method, the server group 61 does not return an ARP response to the ARP request from the client 60 as shown in FIG. 7, and instead the MAC address of the server group set in advance by the load balancer 62 Select one of them and return it. Therefore, in this method, it is necessary to prevent the server from returning an ARP response, and it is necessary to register the MAC address of the server in advance in the load balancer.

  A related technique of a load distribution device (load balancer) based on the NAT method is disclosed in Japanese Patent Laid-Open No. 2004-118622 (Patent Document 1).

JP 2004-118622 A

  As described above, the load distribution device described in Patent Document 1 and the like has the following problems.

  (1) Since the NAT method uses NAPT, complicated processing such as conversion of IP addresses and port numbers, session management, etc. by complicated hardware and software is required for all packets. In the NAT method, since it is necessary to perform address conversion on all packets, the processing capacity decreases as the number of packets increases.

  (2) Since the NAT method performs address conversion based on a TCP or UDP port number, it cannot process a packet without a port number, such as an ICMP packet or a raw IP packet. In addition, since the NAT method uses NAPT, communication from the server side to the client side cannot be performed.

  (5) The NAT method requires special settings such as setting the IP address of the virtual server, the IP address of the interface connecting the server group, the IP address of the server group, etc. to the load balancer. The MAT method also needs to be set not to respond to ARP responses to all servers that perform load balancing, and the MAC address of the server must be registered in advance in the load balancing device. . Changing the server settings is complicated because the number of servers is large and the setting method differs depending on the server OS. In addition, the load balancer setting must be changed each time a new server is added or deleted.

  The present invention has been made in view of the above-described problems. A network load distribution apparatus, a network load distribution method, and a program, which are not complicated and do not require any special setting in a load distribution apparatus or a server group. The purpose is to provide.

The network load distribution apparatus according to the present invention transfers an ARP response from a server selected based on a connection status between each server and the client among a plurality of servers having an ARP response to the ARP request from the client to the client. a network load balancer that, to produce a counter table containing the number of connections with the client for each server as a count value, and means for managing the connection status between the server and client, a small load on the basis of the counter table Server selection means for selecting a server .

The load distribution method according to the present invention transfers an ARP response from a server selected based on a connection status between each server and the client among a plurality of servers having an ARP response to the ARP request from the client to the client. A load balancing method for a network load balancing apparatus, comprising: generating a counter table including the number of connections with clients for each server as a count value; managing a connection status between each server and clients; and based on the counter table And a server selection step of selecting a server with a low load .

The load distribution program according to the present invention operates on a computer, and an ARP from a server selected based on a connection status between each server and the client among a plurality of servers having an ARP response to the ARP request from the client. A load balancing program for transferring a response to a client, and generating a counter table including the number of connections with the client for each server as a count value on a computer, processing and counter for managing the connection status between each server and the client A server selection process for selecting a server with a low load based on the table is executed.

  According to the present invention, complicated processing such as IP address and port number conversion and session management by complicated hardware and software is not required for all packets. Further, it is not necessary to perform address conversion for all packets, and processing capacity does not decrease as the number of packets increases.

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

(Configuration of the embodiment)
FIG. 1 is a diagram showing a network configuration example including a network load distribution apparatus 10 according to an embodiment of the present invention. In the figure, the function of the network load distribution apparatus 10 will be schematically described. When the client 20 accesses a server having an IP address of IP1, the network load distribution apparatus 10 has servers 31, 32, Select one from 33 and distribute.

  Specifically, first, the client 20 broadcasts an ARP request for the IP address IP1 in order to acquire a MAC address corresponding to the IP address IP1. The network load distribution apparatus 10 transfers the ARP request from the client 20 to the port connected to the server group. Receiving the ARP request from the client 20, the servers 31, 32, and 33 return their own MAC addresses as ARP responses.

  Here, the network load distribution apparatus 10 that has received ARP responses from the servers 31, 32, and 33 does not return all ARP responses to the client 20, but selects only one of them and transfers it to the client 20. The client 20 communicates with only one server in the server group as a server having an IP address of IP1.

  Thus, for example, when there are 100 clients and 5 servers, load distribution is performed so that 20 clients are connected to each server.

  FIG. 2 is a block diagram illustrating a configuration of the network load distribution apparatus 10. In the figure, the network load distribution apparatus 10 includes an input / output port unit 100, a layer 2 switch processing unit 11, a packet analysis unit 12, a server selection unit 13, a counter table unit 14, and a timer unit 15.

  The input / output port unit 100 has a plurality of input / output ports 101, 102, 103,... For connecting a network cable. These input / output ports are connected to another network switch, router, or the like via the network cable. Clients and servers are connected.

  The layer 2 switch processing unit 11 transfers the packet received from the input / output port to an appropriate input / output port.

  The packet analysis unit 12 receives a packet received from the input / output port via the layer 2 switch processing unit 11, analyzes the ARP packet, and performs counter table 200-1, 200-2 in the counter table unit 14. 200-3, ... are updated.

  The server selection unit 13 refers to the counter tables 200-1, 200-2, 200-3,... In the counter table unit 14, and sends the ARP response of the MAC address with the smallest counter to the layer 2 switch processing unit 11. From the input / output port.

  The counter table unit 14 includes counter tables 200-1, 200-2, 200-3,... For each IP address. The counter table unit 14 is formed in a recording device composed of a nonvolatile memory such as a magnetic disk or a semiconductor memory, for example.

  FIG. 4 shows an example of a counter table held by the counter table unit 14. As shown in FIG. 4, the counter tables 200-1 and 200-2 record, for each IP address, a MAC address of each server, a counter indicating the number of client connections for each MAC address, and a flag of a reception mark. Updated. The communication status of each server, that is, the number of clients connected to each server is managed by the counter value in the counter table. Further, it is determined by the reception mark whether or not the target server is in a state where communication with the client is possible.

  In the timer unit 15, timers (timer (1), timer (2), timer (3),...) That are times of a predetermined length are set for each IP address. The timer unit 15 is activated after the packet analysis unit 12 receives an ARP request, and is used to complete acceptance of an ARP response when a set timer has elapsed.

(Operation of the embodiment)
Next, the operation of the network load distribution apparatus 10 according to the embodiment of the present invention will be described with reference to FIG. FIG. 3 is a flowchart for explaining the operation of the network load distribution apparatus 10 shown in FIG.

  First, the packet received from the input / output port in step S301 is transferred to the input / output port to be transferred by the layer 2 switch processing unit 11 (step S302). Specifically, if the received packet is a unicast packet, the packet is transferred to an input / output port to which a device having a destination MAC address is connected. If the received packet is a broadcast packet, the packet is transferred to all input / output ports belonging to the same VLAN (Virtual LAN) as the input / output port that received the packet.

  Further, the packet analysis unit 12 checks whether or not the received packet is an ARP request (step S303). If the packet is an ARP request, the counter of the IP address requested by the ARP request is stored in the counter table unit 14. It is confirmed whether the tables 200-1, 200-2, 200-3,... Exist (step S304).

  If the counter table does not exist, the packet analysis unit 12 newly creates a counter table for the IP address (see step S305, FIG. 4). On the other hand, if the counter table already exists, all the reception marks in the counter table are set to off (step S306).

  Thereafter, the packet analysis unit 12 waits for a predetermined time t1 until an ARP response to the ARP request is received from the input / output port (step S307). The predetermined time t1 is set as a timer in the timer unit 15 for each IP address requested by the ARP request, and is counted by the timer unit 15 that is activated after the packet analysis unit 12 receives the ARP request.

  When the ARP response is received within the predetermined time t1 (step S308), the response is made with the ARP response to the counter tables 200-1, 200-2, 200-3,... Corresponding to the IP address requested by the ARP request. It is confirmed whether the MAC address of the server that has been registered is registered (step S309). If there is no corresponding MAC address in the counter table, the MAC address is newly registered in the counter table (step S310), the counter value is set to 0, and the reception mark is turned on (step S311). If there is a corresponding MAC address in the counter table, the reception mark corresponding to the MAC address is turned on (step S311).

  The reception of the ARP response and the updating of the counter tables 200-1, 200-2, 200-3,... Are performed for a predetermined time t1.

  In step S307, when the predetermined time t1 elapses, the server selection unit 13 refers to the counter tables 200-1, 200-2, 200-3,... Corresponding to the IP address requested by the ARP request, and receives them. After deleting the MAC address marked off from the counter table (step S312), the MAC address with the smallest counter value is selected from the counter table (step S313). If there are a plurality of MAC addresses having the smallest counter value, any of them may be selected.

  When selecting the MAC address to be returned, the server selection unit 13 increments the counter corresponding to the MAC address by “1”, and transfers the ARP response including the selected MAC address to the input / output port that has received the ARP request. (Step S314).

  When a new server is added, an ARP response to the ARP request is newly received from the new server. Therefore, a new server is added to the counter tables 200-1, 200-2, 200-3,. Are added by the counter 0, and are automatically subject to load distribution. Also, if the server that has been connected so far is removed, the ARP response to the ARP request will not be returned, so the reception mark in the counter table will remain off, it will be deleted from the counter table, and it will automatically be removed from the load distribution target. Will come off.

(Effect of embodiment)
Hereinafter, the effects of the present embodiment will be described.

  First, for all the packets, complicated processing such as complicated hardware and software conversion of IP address and port number, session management and the like is not required. That is, since the NAT method uses NAPT, complicated processing such as IP address and port number conversion and session management is required for all packets. In this embodiment, the existing layer 2 switch It is only necessary to add processing related to ARP according to the present embodiment.

  Second, the process is simple. In the NAT method, it is necessary to perform address conversion on all packets. In this embodiment, only ARP processing performed at the beginning of communication requires special processing. However, after ARP resolution is completed, Since the operation is a normal layer 2 switch operation, the processing is simple, and the processing capability does not decrease as the number of packets increases.

  Third, non-TCP and non-UDP packets such as ICMP packets and raw IP packets can be handled. Since the NAT method performs address conversion based on the TCP or UDP port number, it cannot process a packet without a port number such as an ICMP packet or a raw IP packet. There is no problem because the port number of the packet is not conscious.

  Fourth, the server group can connect to the client. Since the NAT method uses NAPT, it can only be connected from the client. However, in this embodiment, since the NAPT is not used, connection from the server to the client is possible.

  Fifth, no special setting is required. In the NAT system, it is necessary to set an IP address of a virtual server, an IP address of an interface connecting the server group, an IP address of the server group, and the like to the load balancer. In the MAT method, it is necessary to make a setting so as not to respond to an ARP response to all servers that perform load balancing, and it is necessary to register the MAC address of the server in advance in the load balancing device. In this embodiment, no special setting is required for the server, and no manual setting of the MAC address is required for the load balancer.

  Here, the hardware configuration of the network load distribution apparatus 10 according to the above embodiment will be described with reference to FIG.

Referring to FIG. 6, the network load distribution device 10 can be realized by a hardware configuration similar to that of a general computer device, and is a main memory such as a CPU (Central Processing Unit) 401 and a RAM (Random Access Memory). Yes, a main storage unit 402 used for a data work area and a temporary data save area, a communication unit 403 that transmits and receives data to and from other nodes via a network, an input device 405 such as a keyboard and a mouse, a display device An input / output interface unit 404 that transmits and receives data by connecting to the output device 406 and the storage device 407, and a system bus 408 that interconnects the above-described components are provided. The storage device 407 is, for example, a ROM (Read
Only Memory), a hard disk device including a non-volatile memory such as a magnetic disk and a semiconductor memory.

    The network load distribution apparatus 10 according to the present embodiment implements its operation in hardware by mounting circuit components which are hardware components such as LSI (Large Scale Integration) incorporating a program for performing load distribution control. Of course, a program that provides each function such as the packet analysis unit 12 and the server selection unit 13 is stored in the auxiliary storage unit 405, and the program is loaded into the main storage unit 402 and executed by the CPU 401. Therefore, it can be realized by software.

  In the above-described embodiment, as the method for selecting one from a plurality of MAC addresses, the one with the smallest counter is selected. However, the MAC address may be selected by a round robin method, and the ARP response is returned to the beginning. You may select it as a server with less load.

  Further, as shown in FIG. 5, in the case of a network configuration in which clients exist in a plurality of VLANs VL1 and VL2, and each server belongs to each VLAN with a tagged VLAN (tag VLAN) TVL1 and TVL2, If the load balancer 10 is disposed between them, the above-described configuration can be handled without change. In such an environment, it is not necessary to prepare a separate server for each VLAN, so that efficient load distribution is possible.

  Although the present invention has been described with reference to the preferred embodiments, the present invention is not necessarily limited to the above embodiments, and various modifications can be made within the scope of the technical idea. .

It is a figure which shows the network structural example containing the network load distribution apparatus by embodiment of this invention. It is a block diagram which shows the structure of the network load distribution apparatus by embodiment of this invention. 3 is a flowchart illustrating an operation in the network load distribution apparatus illustrated in FIG. 2. It is a figure which shows an example of the counter table which the counter table part of the network load distribution apparatus by embodiment of this invention hold | maintains. It is a block diagram which shows the example of mounting of the network load distribution apparatus in the network structure where a client is connected via several VLAN. It is a block diagram which shows the hardware structural example of the network load distribution apparatus by this Embodiment. It is a figure which shows the general structural aspect of the network load distribution apparatus by a MAT system.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10: Network load distribution apparatus 20: Client 31-33: Server 11: Layer 2 switch process part 12: Packet analysis part 13: Server selection part 14: Counter table part 15: Timer part 100: Input / output port part 101-103: I / O port

Claims (15)

A network load distribution device for transferring an ARP response from the server selected based on a connection status between each server and the client among the plurality of servers having an ARP response to the ARP request from the client to the client Because
Means for generating a counter table including the number of connections with the client for each server as a count value, and managing a connection status between the server and the client;
Server selection means for selecting a server with a low load based on the counter table;
Network load balancer, characterized in that it comprises a. Means for transferring an ARP request received from a client to a server group and receiving an ARP response from each server of the server group;
Network load balancer according to claim 1, characterized in that Ru and means for transferring the ARP response from the selected server to the client. In the counter table, each IP address is associated with the MAC address of each server and the count value for each MAC address,
The means for managing the connection status between each server and client registers or deletes a MAC address in the counter table according to the presence or absence of an ARP response from each server in the server group. The network load distribution apparatus according to claim 1 or 2 . Timer means for setting a predetermined time to complete acceptance of the ARP response to the ARP request from the client;
The means for managing the connection status between each server and client performs registration or deletion of the MAC address with respect to the counter table until a predetermined time set in the timer means elapses.
The server selection means, after a predetermined time set in the timer means has elapsed, any of claims 1 to 3, characterized in that selecting a small server load among servers said there is ARP response A network load balancer according to claim 1. Receiving a packet from the client, the network load balancing device according to any one of claims 1 to 4, the packet is characterized in that it comprises an analysis means for analyzing whether containing the ARP request. A network load distribution device for transferring an ARP response from the server selected based on a connection status between each server and the client among the plurality of servers having an ARP response to the ARP request from the client to the client Load balancing method,
Generating a counter table including the number of connections with the client for each server as a count value, and managing the connection status between each server and the client;
And a server selection step of selecting a server with a low load based on the counter table . Transferring the ARP request received from the client to the server group and receiving an ARP response from each server of the server group;
The load balancing method according to claim 6 , further comprising a step of transferring an ARP response from the selected server to the client . In the counter table, each IP address is associated with the MAC address of each server and the count value for each MAC address,
The step of managing a connection status between each server and a client performs registration or deletion of a MAC address with respect to the counter table in accordance with the presence or absence of an ARP response from each server of the server group. The load balancing method according to claim 6 or claim 7 . Set a predetermined time to complete acceptance of the ARP response to the ARP request from the client;
In the step of managing the connection status between each server and the client, until the predetermined time elapses, the MAC address is registered or deleted from the counter table,
The load according to any one of claims 6 to 8 , wherein, in the server selection step, a server having a small load is selected from the servers having the ARP response after the predetermined time has elapsed. Distribution method. Receiving a packet from the client, the load balancing method according to claim 6 to claim 9 in which the packet is characterized in that it comprises an analysis step of analyzing whether containing the ARP request. An ARP response from the server selected on the basis of the connection status between each server and the client among the plurality of servers operating on the computer and receiving an ARP response to the ARP request from the client is sent to the client. A load balancing program to transfer ,
In the computer,
A process for managing a connection status between each server and the client, generating a counter table including the number of connections with the client for each server as a count value;
And a server selection process for selecting a server having a low load based on the counter table . In the computer,
A process of transferring an ARP request received from a client to a server group and receiving an ARP response from each server of the server group;
The load distribution program according to claim 11 , wherein a process for transferring an ARP response from the selected server to a client is executed. In the counter table, each IP address is associated with the MAC address of each server and the count value for each MAC address,
The processing for managing the connection status between each server and client performs registration or deletion of a MAC address with respect to the counter table in accordance with the presence or absence of an ARP response from each server of the server group. The load distribution program according to claim 11 or claim 12 . Set a predetermined time to complete acceptance of the ARP response to the ARP request from the client;
In the process of managing the connection status between each server and client, the MAC address is registered or deleted from the counter table until the predetermined time has elapsed.
The load according to any one of claims 11 to 13 , wherein, in the server selection process, a server having a small load is selected from the servers having the ARP response after the predetermined time has elapsed. Distributed program. The load distribution program according to any one of claims 11 to 14 , further comprising an analysis process for receiving a packet from the client and analyzing whether the packet includes the ARP request.

Images