JP2003152729A - Load distribution method, contents distribution system, and load distribution device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a load distribution method capable of solving problems of a conventional method that cannot have efficiently utilized a plurality of mirror servers because a server side cannot uniformly share connections from clients to a plurality of the mirror servers resulting in unable to uniformize a load imposed on a plurality of the mirror servers. SOLUTION: The contents distribution system of this invention is configured such that a connection distributor 30 installed between an edge router 21 and a contents server 40 properly distributes the connection from a client 10 to a plurality of the mirror servers 40-1 to 40-4 being components of the contents server 40.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a load distribution method, a content distribution system and a load distribution device for distributing the load on each mirror server by appropriately distributing the connection from the client via the Internet to each mirror server. Regarding

[0002]

2. Description of the Related Art With the spread of the Internet today,
It is generally performed that a user accesses a server from a client via the Internet and acquires various contents from the server. Some servers are very popular and crowded with many client accesses. In such servers, in order to eliminate congestion, by installing multiple so-called mirror servers that retain the same functions and information as the original server,
The load on the server is distributed.

However, the frequency of access from a client to a plurality of mirror servers differs for each mirror server, and the plurality of mirror servers are not evenly loaded. Therefore, a plurality of mirror servers may not be efficiently utilized, and as a result, a user may not be able to obtain contents from the mirror server at high speed. Therefore, a method in which a user selects an optimum mirror server from a plurality of mirror servers is disclosed in, for example, Japanese Patent Application Laid-Open No.
No. 0-29813 and Japanese Patent Application Laid-Open No. 2001-256208.

The method (server selection system) disclosed in Japanese Patent Laid-Open No. 2000-29813 recognizes a plurality of mirror servers prepared as a primary server, checks the states of the plurality of mirror servers, and checks data with a client. A mirror server suitable for sending and receiving is determined and connected.

Further, the method disclosed in Japanese Patent Laid-Open No. 2001-256208 (method for balancing load between mirror servers) is such that when a web page is accessed by a client, a predetermined script is provided in addition to the web page on the client. In order to establish a connection with each of multiple mirror servers and to measure their response time, the client automatically executes a script to select the mirror server with the shortest response time.

[0006]

However, since the above-mentioned method of selecting a mirror server is a method of selecting an optimum mirror server which is not overloaded on the client side, means for realizing the method in the client (hereinafter , Server selection means) must be set or installed (installed). Therefore, the client that does not set or install the server selection means cannot select the optimum mirror server, and it takes time and effort for the user to set or install the server selection means in the client.

On the other hand, conventionally, on the server side, the connection from the client to the mirror server could not be evenly distributed to the plurality of mirror servers. When multiple mirror servers are installed, an IP address is set for each mirror server. in this case,
The connection from the client is established only to the mirror server corresponding to the IP address specified by the client. That is, on the server side, for a mirror server (mirror server with a different IP address) that does not correspond to the IP address specified by the client,
Does not transfer (route) the packet for establishing the connection from the client.

As described above, since the connection from the client cannot be distributed evenly to the plurality of mirror servers on the server side, the load on the plurality of mirror servers cannot be made uniform and the plurality of mirror servers cannot be distributed. Could not be utilized efficiently. As a result, even if a plurality of mirror servers are installed, the user may not be able to quickly acquire the content from the mirror server because the specific mirror server is congested. .

[0009] For example, if a client uses a certain URL (Un
files existing in iform Resource Locator)
Even if you try to acquire a file that exists at the URL from another client during the acquisition (GET) using the P (Hyper Text Transfer Protocol) protocol, the mirror server of the URL is congested (downloading). Therefore, you will be kept waiting to get the file.

Since the file transfer performance of the mirror server depends on the hardware such as the external storage device of the mirror server, no matter how the line speed is increased, the file acquisition speed (transfer processing time) of the client is Limited.

By using the cache function of the edge router or proxy server on the client side, information about frequently used sites can be temporarily stored,
Although it has been attempted to reduce the traffic, this is not an essential solution because the caching capacity is limited and the load on multiple mirror servers is not equalized on the server side.

The present invention has been made in order to solve the above problems, and by distributing the load on a plurality of mirror servers on the server side, the content transfer processing time on the client side is shortened. An object of the present invention is to provide a load balancing method, a content distribution system, and a load balancing device that can perform the above.

[0013]

In order to achieve the above object, a load balancing method of the present invention is a system in which a client receives a content transfer from any of a plurality of mirror servers via a communication network and a network connection device. In the load balancing method in claim 1, a right to which a connection from a client is allocated is assigned to one mirror server of a plurality of mirror servers, and the one mirror server to which the right is assigned to a network connection device A request for making the MAC address the MAC address of one mirror server is sent out.

In the load balancing method of the present invention, the information of the ARP cache of the network connection device, which is transmitted from one mirror server when the right is assigned, is automatically set to the IP address and MAC address of the one mirror server. Forwards the "irrelevant ARP request" to be updated by the network connection device, and is transferred from the network connection device.
When a packet for a new connection from a client is received, the packet is transferred to one mirror server that holds the right, and a connection is established between the one client and one mirror server to execute application communication. I was allowed to.

Further, in the load balancing method of the present invention, the right is returned from one mirror server when a connection is established, and when the returned right is received, the right is assigned to the other mirror server and the other mirror server The "unfounded ARP request" sent from the server is transferred to the network connection device, and transferred from the network connection device.
When a packet for a new connection from another client is received, the packet is transferred to the other mirror server that currently owns the right, and a connection is established between the other client and the other mirror server to enable the application. I tried to execute communication.

Further, the content distribution system of the present invention is a content distribution system in which a client receives a content transfer from any one of a plurality of mirror servers via a communication network and a network connection device, and the network connection device and the plurality of mirror connection servers A connection distribution device is provided between the mirror server and the mirror server. When the connection distribution device allocates the right to allocate a connection, the mirror distribution server sends an "irrelevant ARP request", and the connection distribution device has a plurality of mirrors. The right is assigned to one mirror server among the servers, and a new connection from the client is assigned to the one mirror server that holds the right.

Further, the content distribution system of the present invention includes a right allocation processing unit for allocating a right to a connection distribution device to one mirror server among a plurality of mirror servers, and one mirror server among the plurality of mirror servers. I of a client that has already established a connection with
A table that stores the correspondence between the P address and port number and the MAC address of the one mirror server to which the connection is established, and constantly monitors the connection and disconnection of the connection between the client and the one mirror server,
A connection monitoring processing unit that adds a correspondence relationship to a table when a connection is connected and deletes the correspondence relationship from the table when the connection is disconnected, and a client to one mirror server transferred from a network connection device. Destination MAC of packet sent to
From a MAC address conversion processing unit that converts the address from the MAC address of another mirror server that currently holds the right to the MAC address of one mirror server, and a switching processing unit that switches based on the destination MAC address of the packet. Configured.

Further, in the content distribution system of the present invention, the switching processing unit does not transfer the "irrelevant ARP request" received from one mirror server to the other mirror server, but to the network connection device. Configured.

Further, in the content distribution system of the present invention, one mirror server that owns the right returns the right when it sends a response packet to the client.

Further, the load balancing apparatus of the present invention is provided between a network connection device and a plurality of mirror servers in a system in which a client receives content transfer from any of a plurality of mirror servers via a communication network and a network connection device. The load balancer is assigned to one mirror server of the plurality of mirror servers and a new mirror server from the client assigns the right to the one mirror server that holds the right. Configured to allocate different connections.

Further, the load distribution apparatus of the present invention has already established between the right allocation processing section for allocating the right to one mirror server among the plurality of mirror servers and the one mirror server among the plurality of mirror servers. A table that stores the correspondence relationship between the IP address and port number of the client that has the connection and the MAC address of the one mirror server that has the connection, and the connection of the connection between the client and the one mirror server And connection disconnection processing unit that constantly monitors the disconnection and adds the correspondence relationship to the table when the connection is connected, and deletes the correspondence relationship from the table when the connection is disconnected, and is transferred from the network connection device. Destination MAC of the packet sent from the client to one mirror server From a MAC address conversion processing unit that converts the address from the MAC address of another mirror server that currently owns the right to the MAC address of one mirror server, and a switching processing unit that switches based on the destination MAC address of the packet. Configured.

[0022]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. Embodiment 1. FIG. 1 is a diagram showing a network configuration to which the content distribution system of the present invention is connected. The content distribution system of the present invention includes a connection distribution device 30 and a content server 40 that realize a load distribution device.
Composed of. The content server 40 is composed of a plurality of mirror servers 40-1 to 40-4.

In FIG. 1, the client 10 is, for example, a personal computer provided by the user on the side to which the content is provided from the content server 40. The client 10 is connected to the Internet 100 via, for example, a modem or a terminal adapter. It is assumed that “IP1” is assigned as the IP address to the client 10.

The edge router 21 as a network connecting device is provided with an external Internet 100 and an internal LAN.
It is provided at a connection point with (Ethernet) and performs packet routing (route selection) based on the IP address in the IP header of the packet. Edge router 21
Includes an ARP cache 22 that is a table that stores the correspondence between MAC addresses and IP addresses.

Generally, the terminal connected to the LAN recognizes the correspondence between the MAC address and the IP address and realizes the LAN communication, like the edge router 21.
It has an ARP cache. ARP (Address Reso
For example, when the MAC address is unknown because there is no information in the ARP cache at the time of the first communication (such as after completion of setup after new purchase), the IP address is inquired from all terminals and Is a protocol for acquiring MAC address information in the reply frame. A for this ARP
There are RP requests and ARP responses.

It is assumed that "MAC1" is assigned to the edge router 21 as a MAC address which is a physical address of a terminal required for LAN communication.

The connection distribution device 30 is installed between the edge router 21 and the content server 40, and each of the mirror servers 40-1 to 40-1 constituting the content server 40.
It is a device that appropriately distributes the connection from the client 10 to 40-4.

The connection distributor 30 sends to the edge router 21 an "unfounded ARP request" which is a kind of ARP (IEF (Internet Engineeri) for ARP.
ng Task Force) Mirror server 40-1 which can notify its own MAC address using document RFC826)
~ 40-4, that is, the mirror server 40-that the connection from the client 10 is allocated
Token (Token), which is the right of 1,40-2,
One mirror server (for example, mirror server 40-
It has a function to be assigned to 1). Further, the connection distribution device 30 has a function of allocating a new connection from the client 10 to the mirror server (for example, the mirror server 40-1) holding the token.

Furthermore, the connection distribution device 30 terminates the connection between the client 10 and the mirror server at the same mirror server, that is, maintains the connection already established between the client 10 and the mirror server. Destination MAC of packet transmitted from 10 to mirror server 40-1, for example
It also has a function of converting an address into a valid MAC address of the mirror server 40-1.

Incidentally, the connection distribution device 30 realizes to appropriately distribute the connection from the client 10 to each of the mirror servers 40-1 to 40-4 by adding the above-mentioned functions to a normal MAC switch. is doing. In addition, the connection distribution device 30 uses "MAC" as a MAC address used when the token is transferred between the mirror servers 40-1 to 40-4.
6 ”is assigned. Connection distributor 30
The specific configuration of the above will be described later (see FIG. 2).

The content server 40 is the client 1
It is a server that provides contents in response to requests from 0. As described above, the content server 40 is composed of the plurality of mirror servers 40-1 to 40-4.
Each mirror server 40-1 to 40-4 has the same content with the same IP address "IP2". Therefore, the edge routers 21 connected to the WAN side are not aware of the mirror servers 40-1 to 40-4,
It is recognized as if there is only one content server 40 to which the global address is assigned.

"MAC2" is assigned to the mirror server 40-1 as a MAC address, and the mirror server 4-1
“MAC3” is assigned to 0-2, “MAC4” is assigned to the mirror server 40-3, and “MAC5” is assigned to the mirror server 40-4. Is assigned.

Incidentally, each mirror server 40-1 to 40-4
Have the same IP address and the same content, but may have the same hard disk image (dead copy). The specific configuration of each of the mirror servers 40-1 to 40-4 will be described later (see FIG. 4).

Each of the mirror servers 40-1 to 40-4 has a function of receiving a token transmitted from the connection distribution device 30 and acquiring it, and then transmits an "irrelevant ARP request" to the edge router 21. ing.

Here, the "unfounded ARP request" notifies the IP address and MAC address of the own terminal (own host) to another terminal (other host), and the IP address and MAC address of the ARP cache of the other terminal. This is the frame to be updated. The other terminal that has received the "unfounded ARP request" must update the information of the IP address and the MAC address stored in the ARP cache according to the content of the ARP request. Note that unless another terminal uses the IP address assigned to its own terminal, it does not respond to this ARP request (if there is a response, it is said that the IP addresses are assigned in duplicate. It is an obstacle and usually impossible.

Further, each mirror server 40-1 to 40-4
Has a function of transmitting a token owned by itself to the connection distribution device 30 and returning it when a connection is established with the client 10, that is, when application communication is started.

In FIG. 1, the content server 4
0 is composed of four mirror servers 40-1 to 40-4, but is not limited to four and may be composed of a plurality of mirror servers 40-1 to 40-n.

FIG. 2 is a block diagram showing the configuration of the connection distribution device. In FIG. 1, four mirror servers 40-1 to 40-4 are connected to the connection distribution device 30, but in FIG. 2, in order to simplify the description,
The connection distribution device 30 has two mirror servers 40-
1, 40-2 are connected.

As shown in FIG. 2, the connection distribution device 30 is provided with a physical port 31 for connecting the edge router 21 on the WAN side and a mirror server 40-on the LAN side.
Physical ports 32 and 33 for connecting 1 and 40-2, respectively
Is provided. In FIG. 2, the connection distribution device 30 has one physical port 31 on the WAN side
The configuration is such that the two physical ports 32 and 33 are accommodated on the side, but the configuration is not limited to this, and N on the WAN side.
It is also possible to accommodate N physical ports on the LAN side.

The token allocation processing unit 34 selects an appropriate mirror server that is not overloaded from the plurality of mirror servers 40-1 and 40-2, and transmits a token to the selected mirror server to allocate it. The connection monitoring processor 35 includes a client 10 and a mirror server 40.
-1, 40-2 are constantly monitored for connection and disconnection, and when the connection is established, an entry is added to the connection-to-MAC address correspondence table 36,
Connection to MAC when the connection is broken
The entry is deleted from the address correspondence table 36.

Connection-to-MAC address correspondence table 36
Is a plurality of clients 10 and a mirror server 40-1,
In the communication of the application 40-2, the connection in each communication and the mirror server 40-1,
It is a table which memorize | stores the information of the correspondence with the MAC address of 40-2.

FIG. 3 is a diagram showing an example of the connection-to-MAC address correspondence table. In FIG. 3, when the client 10 and the mirror server application communicate with each other, the packet I on the client 10 side is transmitted.
A set of the P address and the port number on the client 10 side is called "connection". Also, the MA of the mirror server communicating with the connection and the application.
The information on the correspondence relationship between the C address (the destination MAC address of the downlink frame) is called an “entry”. Entry is MA
The C address conversion processing unit 37 makes the M based on the connection.
Connection-to-MAC in associative memory format with connection as a key so that AC addresses can be searched efficiently
It is stored in the address correspondence table 36.

In the MAC address conversion processing unit 37, the connection in the packet (downstream frame) input from the physical port 31 is the connection-to-MAC address correspondence table 36.
Whether or not the connection exists in the connection-to-MAC address correspondence table 36,
According to the entry of the connection-to-MAC address correspondence table 36, the destination MAC address of a packet
Convert (rewrite) to C address. Further, the MAC address conversion processing unit 37 searches whether or not the connection in the packet (uplink frame) input from the physical port 32 or 33 exists in the connection-to-MAC address correspondence table 36, and the connection is the connection pair M.
If it exists in the AC address correspondence table 36, the source MAC address of the packet is converted into the MAC address of the entry according to the entry of the connection-to-MAC address correspondence table 36.

The switching processing unit 38 uses the physical port 3
Based on the destination MAC address of the packet input from 1, the packet is switched to the corresponding physical port 32 or 33. In addition, the switching processing unit 38
Switches the packet toward the physical port 31 based on the destination MAC address of the packet input from the physical port 32 or 33. The switching processing unit 38 switches the packet input from the physical port 32 or 33 only toward the physical port 31. That is, the switching processing unit 38 does not switch from the physical port 32 to the physical port 33 and from the physical port 33 to the physical port 32.

FIG. 4 is a block diagram showing the structure of the mirror server. Since the mirror server 40-1 and the mirror server 40-2 have exactly the same configuration as described above, only the internal configuration of the mirror server 40-1 is shown, and the illustration of the mirror server 40-2 is omitted. ing.

In FIG. 4, the communication unit 41 is connected to the physical port 32 of the connection distribution device 30 and transmits / receives packets (frames) to / from the connection distribution device 30. The storage unit 42 includes a control program 42a for controlling various processes of the mirror server 40-1 and an HTT.
Content 42b such as a file transferred by an application such as P or FTP (File Transfer Protocol)
Etc. are stored (saved). ARP cache 43 is M
It is a table which stores the correspondence between AC addresses and IP addresses.

The control unit 44 controls the operation of providing the contents 42b to the client 10 in accordance with the control program 42a stored in the storage unit 42, and also implements the following load balancing method of the present invention. Such processing is also performed. That is, when the communication unit 41 receives the token transmitted from the connection distribution device 30, the control unit 44 follows the control program 42 a of the storage unit 42 and
The process of transmitting the "unfounded ARP request" to the edge router 21 is executed. In addition, the control unit 44, when a connection is established with the client 10,
According to the control program 42a, the processing for transmitting and returning the held token to the connection distribution device 30 is executed.

FIG. 5 is a diagram showing a protocol stack used by the mirror server for communication. In FIG. 5, MA
C / LLC (Media Access Control / Logical Link Con)
trol) is a protocol corresponding to the data link layer (Layer 2) and is used to frame and transmit / receive data, and perform flow control, transmission error control, etc. thereof.

IP (Internet Protocol) is a protocol compatible with the network layer (layer 3), and identifies a communication partner with an IP address to realize end-to-end communication. ARP is also a protocol compatible with the network layer, and as described above, it obtains the MAC address from the IP address of the communication partner. The token processing unit is a new protocol used for passing (sending and receiving) tokens, and is newly defined in the network layer.

TCP / UDP (Transmission Control P
The rotocol / User Datagram Protocol) is a protocol compatible with the transport layer (layer 4) and establishes / releases a connection and transfers data. The application corresponds to the session layer (layer 5) to the application layer (layer 7), and a protocol such as FTP or HTTP is used.

As shown in FIG. 5, the mirror server 40-
A token processing unit is provided in a frame transmitted / received between the connection units 1 and 40-2 and the connection distribution device 30. The token processing unit causes the control unit 44 of the mirror servers 40-1 and 40-2 to transfer the token. Upon receiving the token transmitted from the connection distribution device 30, the control unit 44 immediately after that receives the ARP of the frame.
Is set to "Unfounded ARP request" and is sent. Further, the control unit 44 returns a token to the connection distribution device 30 when communication is started from the client 10 to the application and application data is transmitted to the client 10.

FIG. 6 is a diagram showing a frame structure (Ethernet frame structure) of the "unfounded ARP request". As shown in FIG. 6, the “destination address” is a broadcast address (FF: FF: FF: FF: F
F: FF) is stored, and as the "source address",
The MAC address given to the own interface is stored. The frame type is ARP.

The "sender's Ethernet address" stores the MAC address assigned to its own interface, and the "sender's IP address" stores the IP address assigned to its own host.
The "target Ethernet address" is a broadcast address (FF: FF: FF: FF: F
F: FF) is stored, and the IP address assigned to the host is stored as the "target IP address".

The "unfounded ARP request" is different from the ordinary "ARP request" in that the "target IP address" is the IP assigned to the host (self).
It's just an address.

Next, the operation of the load balancer of the present invention will be described. FIG. 7 is a sequence diagram for explaining the token allocation operation in the content distribution system of the present invention.

As a premise, it is assumed that the connection distribution device 30 is in a state of holding a token. This state is an initial state in which the power supply of the connection distribution device 30 is turned on, or a state in which tokens are returned from the mirror servers 40-1 and 40-2 (step ST4 in FIG. 9).
0).

In the connection distribution device 30, the token allocation processing section 34 determines the mirror servers 40-1 and 40-2 to which tokens should be allocated (given) by an arbitrary algorithm (step ST1). This algorithm is, for example, each mirror server 40-1, 40.
-2 is given an order and the connection destination of the connection from the client 10 is sequentially selected for each connection, or the mirror servers 40-1 and 40-2.
There is a priority order method in which a predetermined criterion is weighted and the priority is determined in consideration of the weighting. In addition, in this Embodiment, it demonstrates as what uses a round robin method.

The token allocation processing unit 34 uses the mirror server (here, the mirror server 4) determined by the round robin method.
A token is transmitted via the physical port 32 (step ST2). At this stage, the connection distribution device 30 has the mirror server 40-
M that carries the token because it does not know the MAC address of 1
A broadcast address is used as the destination MAC address of the AC frame, and a MAC address (MAC6) assigned to the connection distribution device 30 is used as the source MAC address.

In the mirror server 40-1, the control unit 44
Receives the token transmitted from the connection distribution device 30 via the communication unit 41 (step ST3),
According to the control program 42a, control is performed so as to output the "unfounded ARP request". That is, the control unit 44
6, as shown in FIG. 6, assigns only one IP address (IP2) provided as the content server 40 to the sender IP address and the target IP address of the ARP frame (ARP packet), and sets it as such. The "unfounded ARP request" is transmitted via the communication unit 41 (step ST4).

In the connection distribution device 30, when the token allocation processing unit 34 receives the "irrelevant ARP request" transmitted from the mirror server 40-1 via the physical port 32, the "unreliable ARP request" is returned. The source address (MAC2) is taken out and stored in the connection-to-MAC address correspondence table 36, and is notified to the MAC address conversion processing unit 37 (step ST5).

When the switching processing unit 38 receives the "irrelevant ARP request" transmitted from the mirror server 40-1 via the physical port 32, the "irrelevant ARP request" is transmitted to the physical port on the WAN side. It switches to 31 as it is. As a result, "Unfounded A
The RP request ”is sent to the edge router 2 via the physical port 31.
1 (step ST6).

Incidentally, as shown in FIG.
Since the broadcast address is stored in the destination address of the "RP request", the "baseless A
The “RP request” is transferred to all terminals connected to the LAN. However, the switching processing unit 38 uses the LAN
Since the physical ports 32 and 33 on the side are configured not to switch, the "unfounded ARP request" transmitted from the mirror server 40-1 is not switched to the physical port 33 on the LAN side. Therefore, the "irrelevant ARP request" is not transferred to the mirror server 40-2.

When the edge router 21 receives the "irrelevant ARP request" transferred from the connection distributor 30, the edge router 21 determines the IP address (IP2) of the mirror server 40-1 according to the content of the "irrelevant ARP request". MAC
The address (MAC2) is stored in the ARP cache 22 and updated (step ST7). As a result, the connection destination of the connection from the client 10 is the mirror server 4
It will be automatically determined to 0-1.

FIG. 8 is a sequence diagram for explaining the connection allocating operation in the content distribution system of the present invention.

When the user wants to acquire the content provided on the Internet 100 by the content server 40, the user uses the content server 4 at the client 10.
The IP address (IP2) of 0 is designated, the packet is transmitted to the content server 40 for access (step S
T10).

When the edge router 21 receives a packet for accessing the content server 40 from the client 10, the edge router 21 refers to the information stored in the ARP cache 22 to transfer the packet.
The MAC address (destination MAC address) corresponding to the P address (IP2) is immediately resolved. At this time, ARP
The MAC address stored in the cache 22 is "M
Since it is "AC2", the edge router 21 transfers the packet from the client 10 to the mirror server 40-1 of MAC2 which currently holds the token (step ST11).

In the connection distribution device 30, when the connection monitoring processor 35 receives the packet transferred from the edge router 21 via the physical port 31, the connection monitoring processor 35 of FIG.
As shown in FIG. 5, the connection of the client 10 side IP address (IP1) and the port number (for example, 23) is used as a connection, and the connection and the destination MAC address (MAC2) of the packet (frame) are used as an entry for connection-to-MAC address correspondence It is added to the table 36 (step ST12).

Since there is no entry corresponding to the connection of the packet in the connection-to-MAC address correspondence table 36, the MAC address conversion processing unit 37 sends the packet to the switching processing unit 38 without converting the MAC address of the packet. .

The switching processing unit 38 follows the physical port 3 according to the destination MAC address (MAC2) of the packet.
Switch towards 2. As a result, the packet is transferred to the mirror server 40-1 (step ST
13).

In the mirror server 40-1, the control unit 44
Receives a packet transmitted from the connection distribution device 30 via the communication unit 41 (step ST1
4) Execute application processing according to the control program 42a and send a response packet (step S
T22).

However, the control unit 44 controls the ARP cache 4
If the information is not stored in 3, the response packet cannot be transmitted. Therefore, before transmitting the response packet, the ARP request is transmitted (originated) to resolve the MAC address of the edge router 21 ( Step ST1
5). After that, the control unit 44 executes application processing (step ST16).

In the connection distribution device 30, when the switching processing unit 38 receives the ARP request transmitted from the mirror server 40-1 via the physical port 32, A
The RP request is switched to only the physical port 31. As a result, the ARP request is transferred only to the edge router 21 (step ST17).

When the edge router 21 receives the ARP request transferred from the connection distribution device 30, the AR
An ARP response to the P request is transmitted (step ST1
8). Then, the connection distribution device 30 transfers the ARP response to the mirror server 40-1 by the switching processing unit 38 switching the ARP response toward the physical port 32 (step ST19).

In the mirror server 40-1, the control unit 44
Receives the ARP response transferred from the connection distribution device 30 via the communication unit 41 (step ST2
0), the ARP cache 4 according to the content of the ARP response
The IP address and the MAC address (MAC1) of the edge router 21 are stored in 3 (step ST21). After that, the control unit 44 transmits a response packet for the access from the client 10 to the client 10 (step ST22).

In the connection distribution device 30, when the switching processing unit 38 receives the response packet transmitted from the mirror server 40-1 via the physical port 32,
The response packet is switched to the physical port 31. Therefore, the response packet is the switching processing unit 38.
From the MAC address conversion processing unit 37.

The MAC address conversion processing unit 37 refers to the entry of the connection-to-MAC address correspondence table 36 to refer to the source MAC address (MAC2) of the response packet as the mirror server 4 which holds the token.
It is rewritten to the MAC address (MAC2) of 0-1 (step ST23). At this time, the source MAC of the packet
Mirror server 40 holding address and token-
Since the MAC address of 1 is MAC2 and has the same value, the MAC address is overwritten with the same value.

After the MAC address conversion processing unit 37 performs the MAC address conversion processing, the response packet passes through the connection monitoring processing unit 35 and is transferred to the edge router 21 via the physical port 31 (step ST2
4).

The edge router 21 sends the response packet transferred from the connection distributor 30 to the mirror server 40.
It is received as a response packet from -1. Upon receiving the response packet, the edge router 21 performs routing and transfers the response packet to the client 10 (step ST25).

When the client 10 receives the response packet transmitted from the edge router 21 via the Internet 100, the client 10 transmits the packet corresponding to the response packet (step ST26). When the edge router 21 receives the packet transmitted from the client 10,
The data is routed and transferred to the mirror server 40-1 (step ST27).

In the connection distribution device 30, when the connection monitoring processing unit 35 receives the packet transferred from the edge router 21 via the physical port 31, the same connection as the connection in the packet is stored in the connection-to-MAC address correspondence table 36. Check if it exists. At this time, since the connection monitoring processing unit 35 finds the same connection in the connection-to-MAC address correspondence table 36, no entry is added to the connection-to-MAC address correspondence table 36. The packet that has passed through the connection monitoring processing unit 35 is sent to the MAC address conversion processing unit 37.

When the MAC address conversion processing unit 37 confirms that the same connection as the packet connection exists in the connection-to-MAC address correspondence table 36, the destination MAC address (MAC2) of the packet corresponds to the connection-to-MAC address. The MAC address of the entry in Table 36 is rewritten (step ST28). At this time, since the destination MAC address of the packet and the MAC address of the entry in the connection-to-MAC address correspondence table 36 are both MAC2 and have the same value, the MAC address is overwritten.

The packet has a MAC address conversion processing unit 3
After the MAC address conversion processing is performed at 7, the data is sent to the switching processing unit 38. Switching processing unit 38
Switches the packet towards the physical port 32 according to the destination MAC address of the packet. As a result, the packet is transferred to the mirror server 40-1 (step ST29). The mirror server 40-1 receives the packet transferred from the connection distribution device 30 (step ST30).

After that, the packets exchanged in the application communication between the client 10 and the mirror server 40-1 are the same as those in steps ST22 to S22.
The process of T30 is repeated.

FIG. 9 is a sequence diagram for explaining the token allocation operation after the token is returned in the content distribution system of the present invention.

The control unit 44 of the mirror server 40-1, when a connection is established with the client 10,
That is, immediately after transmitting the response packet in the process of step ST22 of FIG. 8, the token is transmitted to the connection distribution device 30 and returned (step ST40).

The reason why the mirror server 40-1 returns the token is that, when there is a connection from another client 10 'to the content server 40, the connection destination of the connection is handed over to another mirror server.

In the connection distribution device 30, when the token allocation processing section 34 receives the token transmitted from the mirror server 40-1 via the physical port 32 (step ST41), the token allocation processing section 34 allocates the next token according to the round robin method. Decide the mirror server (step ST
42), and transmits the token to the mirror server (step ST43). Here, the token allocation processing unit 34
Mirror server 40-2 as the connection destination of the next connection
It is assumed that the token has been sent.

In the mirror server 40-2, the control unit 44
Receives the token transmitted from the connection distribution device 30 via the communication unit 41 (step ST4
4), "Unfounded ARP request" is transmitted via the communication unit 41 (step ST45).

In the connection distribution device 30, when the token allocation processing unit 34 receives the “irrelevant ARP request” transmitted from the mirror server 40-2 via the physical port 33, the “unreliable ARP request” is returned. The source address (MAC3) is taken out, stored in the connection-to-MAC address correspondence table 36, and notified to the MAC address conversion processing unit 37 (step ST46).

Further, when the switching processing unit 38 receives the "groundless ARP request" transmitted from the mirror server 40-2 via the physical port 33, the "groundless ARP request" is transmitted to the WAN side physical port. Switching towards 31. As a result, the "irrelevant ARP request" is transferred to the edge router 21 via the physical port 31 (step ST47).

When the edge router 21 receives the "unfounded ARP request" transferred from the connection distributor 30, the edge router 21 determines the IP address (IP2) of the mirror server 40-2 in accordance with the contents of the "unfounded ARP request". MAC
The address (MAC3) is stored in the ARP cache 22 and updated (step ST48). As a result, the connection destination of the next connection from the client 10 is automatically determined by the mirror server 40-2.

FIG. 10 is a sequence diagram for explaining another new connection allocation operation in the content distribution system of the present invention.

In FIG. 10, it is assumed that there is a connection from the client 10 'to the content server 40 separately from the client 10. When the user wants to acquire the content from the content server 40, the client 10 'uses the IP of the content server 40
The address (IP2) is specified, and the packet is transmitted to the content server 40 for access (step ST5).
0).

The edge router 21 uses the client 10 '.
When the packet for accessing the content server 40 from is received, the routing is performed by referring to the information stored in the ARP cache 22, and the packet is transferred (step ST51). Here, since the MAC address stored in the ARP cache 22 is “MAC3”, the edge router 21
Transfers the packet from the client 10 'to the mirror server 40-2 of MAC3 which currently holds the token (step ST51).

In the connection distribution device 30, when the connection monitor processing unit 35 receives the packet transferred from the edge router 21 via the physical port 31, the set of the IP address and the port number of the client 10 'side of the packet is used as the connection. The connection and the destination MAC address (MAC3) of the packet (frame) are added as new entries to the connection-to-MAC address correspondence table 36 (step ST52).

Since the MAC address conversion processing unit 37 has no entry corresponding to the new connection in the connection-to-MAC address correspondence table 36, M of the packet is
The packet is sent to the switching processing unit 38 without converting the AC address.

The switching processing unit 38 follows the physical port 3 according to the destination MAC address (MAC3) of the packet.
Switch to 3. As a result, the packet is transferred to the mirror server 40-2 (step ST
53).

In the mirror server 40-2, the control unit 44
Receives the packet transmitted from the connection distribution device 30 via the communication unit 41 (step ST5
4) Execute application processing according to the control program 42a and send a response packet (step S
T55).

In the connection distribution device 30, when the switching processing unit 38 receives the response packet transmitted from the mirror server 40-2 via the physical port 33,
The response packet is switched to the physical port 31. Therefore, the response packet is the switching processing unit 38.
From the MAC address conversion processing unit 37.

The MAC address conversion processing unit 37 refers to the entry of the connection-to-MAC address correspondence table 36 to refer to the source MAC address (MAC3) of the response packet as the mirror server 4 which holds the token.
The MAC address of 0-2 (MAC3) is rewritten (step ST56). At this time, the source MAC of the packet
Mirror server 40 holding address and token-
Since the MAC address of 2 is MAC3 and has the same value, the MAC address is overwritten.

After the MAC address conversion processing unit 37 performs the MAC address conversion processing, the response packet passes through the connection monitoring processing unit 35 and is transferred to the edge router 21 via the physical port 31 (step ST5
7).

The edge router 21 sends the response packet transferred from the connection distributor 30 to the mirror server 40.
-2 as a response packet. Upon receiving the response packet, the edge router 21 performs routing and transfers the response packet to the client 10 '(step ST58).

The client 10 'is connected to the edge router 21
The response packet transmitted from is received via the Internet 100 (step ST59). After that, when the client 10 ′ transmits a packet in response to the response packet, the steps ST26 to ST in FIG.
30 processes are performed.

FIG. 11 is a sequence diagram for explaining the MAC address conversion operation in the content distribution system of the present invention.

As shown in FIG. 8, the client 1
0 and the mirror server 40-1 are already performing application communication, and as shown in FIG. 9, the token is passed to the mirror server 40-2, and the mirror server 4
It is assumed that 0-2 is actually in a state of holding a token.

When a packet is transmitted from the client 10 (step ST60), the packet is routed by the edge router 21. At this time, the mirror server 40-2 holds the token, and the ARP cache 22 stores the MAC address (M
Since AC3) is stored, the edge router 21
Mirror server 40-
It will be transferred to 2 (step ST61).

In the connection distribution device 30, when the packet transmitted from the edge router 21 is received via the physical port 31, the connection monitoring processor 35 stores the same connection as the packet in the connection-to-MAC address correspondence table 36. No entry is added to the connection-to-MAC address correspondence table 36 to find the connection.

When the MAC address conversion processing unit 37 confirms that the same connection as the packet connection exists in the connection-to-MAC address correspondence table 36, the destination MAC address (MAC3) of the packet corresponds to the connection-to-MAC address. The MAC address (MAC2) of the mirror server 40-1 corresponding to the connection in Table 36 is rewritten (step ST28).

The packet includes the MAC address conversion processing unit 3
After the MAC address conversion processing is performed at 7, the data is sent to the switching processing unit 38. Switching processing unit 38
Switches the packet toward the physical port 32 according to the destination MAC address (MAC2) of the packet. As a result, the packet is transferred to the mirror server 40-1 (step ST63).

When the communication section 41 of the mirror server 40-1 receives the packet transferred from the connection distribution apparatus 30 (step ST64), the control section 44 follows the application processing from the communication section 41 to the client 1.
A response packet is transmitted to 0 (step ST6)
5).

In the connection distribution device 30, when the switching processing unit 38 receives the response packet transmitted from the mirror server 40-1 via the physical port 32,
The response packet is switched to the physical port 31. Therefore, the response packet is the switching processing unit 38.
From the MAC address conversion processing unit 37.

The MAC address conversion processing unit 37 refers to the MAC address of the mirror server 40-2 holding the token stored in the connection-to-MAC address correspondence table 36, and determines the source MAC address (MAC2) of the response packet. ) Mirror server 4 holding a token
It is rewritten to the 0-2 MAC address (MAC3) (step ST66).

The response packet is subjected to MAC address conversion processing by the MAC address conversion processing unit 37, then passes through the connection monitoring processing unit 35 and is transferred to the edge router 21 via the physical port 31 (step ST6
7).

The edge router 21 sends the response packet transferred from the connection distributor 30 to the mirror server 40.
-2 as a response packet. in this way,
The edge router 21 actually receives the response packet transmitted from the mirror server 40-1 as the response packet from the mirror server 40-2. Therefore, when the edge router 21 routes the packet, the consistency with the information stored in the ARP cache 22 is maintained.

The edge router 21 transfers the received response packet to the client 10 (step ST
68). The client 10 receives the packet transferred from the edge router 21 via the Internet 100 (step ST69).

FIG. 12 is a sequence diagram for explaining the operation of deleting an entry by ARP request / ARP response and connection termination in the content distribution system of the present invention.

Generally, the ARP cache 22 (ARP
Information on the IP address and the MAC address stored in the cache 43 is the same as the age time (for example, 10
Minutes) automatically erases (clears) (step S
T70).

Therefore, in the edge router 21, when the packet is transmitted from the client 10 after the IP address and the MAC address of the ARP cache 22 are erased by the age (step ST71), the routing is performed and the packet is transferred. In order to do so, it is necessary to make an ARP request. The edge router 21 receives the packet transmitted from the client 10 (step S
T72), ARP request is transmitted (step ST7)
3).

In the connection distribution device 30, the MAC address conversion processing unit 37 stores the destination MAC address, which is the broadcast address in the ARP request transmitted from the edge router 21, in the connection-to-MAC address correspondence table 36. MAC address of the mirror server (for example, 40-2) that holds the
It is rewritten to 3) (step ST74).

The switching processing unit 38 switches to the physical port 33 as an ARP request having the MAC address (MAC3) of the unicast address rewritten by the MAC address conversion processing unit 37.
As a result, the ARP request is transferred only to the mirror server 40-2 holding the token (step ST75).

In the mirror server 40-2, the control unit 44
When the ARP request is received via the communication unit 41, the AR
The P response is transmitted via the communication unit 41 (step ST7).
6).

In the connection distribution device 30, the switching processing unit 38 switches the ARP response transmitted from the mirror server 40-2 toward the physical port 31, and the MAC address conversion processing unit 37 determines the source MAC of the ARP response. The address is rewritten to the MAC address (MAC3) of the mirror server 40-2 holding the token, that is, overwritten (step ST77), and the ARP response is transferred from the physical port 31 to the edge router 21 (step ST78).

When the edge router 21 receives the ARP response transferred from the connection distributor 30, the edge router 21 receives the ARP response.
Information on the IP address (IP2) and the MAC address (MAC3) of the ARP cache 22 is stored according to the content of the response (step ST79). After that, the edge router 21 transfers the packet from the client 10 to the mirror server 40-2 (step ST80).

In the connection distribution device 30, when the connection monitoring processor 35 receives a packet, the connection monitoring processor 35 indicates in the connection-to-MAC address correspondence table 36 the client 10
The entry corresponding to the connection from (1) is added (step ST81), and the switching processing unit 38 switches the packet toward the physical port 33. As a result, the packet is transferred to the mirror server 40-2 (step ST82). Then, the mirror server 40-2 is
The packet from the connection distribution device 30 is received (step ST83).

After that, the processes of steps ST22 to ST30 of FIG. 8 are performed.

When the connection monitor processor 35 of the connection distributor 30 detects, for example, the end (disconnection) of the connection between the client 10 and the mirror server 40-2 (step ST90), the connection-to-MAC address correspondence table. The entry corresponding to the connection stored in 36 is deleted (step ST
91). This avoids duplicate entries of the same connection stored in the connection-to-MAC address correspondence table 36.

As described above, according to the first embodiment, the connection distribution device 30 installed between the edge router 21 and the content server 40 causes the plurality of mirror servers 40-1 constituting the content server 40.
To 40-4 are configured to appropriately distribute the connection from the client 10 to each of the mirror servers 40-1 to 40-4, the loads on the plurality of mirror servers 40-1 to 40-4 can be evenly distributed. It is possible to reduce the content transfer processing time on the side of 10. As a result, the response to the content server 40 on the Internet 100 is further improved.

Further, the system administrator does not need to make any special setting for the edge router 21 in order to install the load balancing system of the present invention. This is because the edge router 21 operates in the same way as one content server 40 to which a global address is given is connected. Therefore, the burden on the system administrator required to install the load balancing system of the present invention is light.

The system administrator and the mirror server 40
For -1 to 40-4, it suffices to install software (control program 42a) that executes a protocol (token processing unit) for transmitting and receiving tokens, and make any special settings for existing applications. No need. Therefore, the work load for installing the load balancing system of the present invention is light to the system administrator.

Further, the plurality of mirror servers 40-1 to 40-4 constituting the content server 40 not only have exactly the same contents, but are also given the same IP address (global address). Therefore, each mirror server 40-1 to 40-4 may be a dead copy of the original server, and since the work of performing a dead copy is simple, the burden on the system administrator is further reduced.

Further, each mirror server 40-1 to 40-
When it is desired to increase the degree of load distribution for 4, it is sufficient to add a physical port on the LAN side of the connection distribution device 30 and connect a dead copy mirror server.
The system administrator can quickly respond.

Embodiment 2. In the connection distribution device 30, it is possible to add a physical port on the WAN side and connect a plurality of edge routers. In this case, the "irrelevant ARP request" transmitted from the mirror server (for example, 40-1) is transferred to all edge routers, and all edge routers update the information in the ARP cache. Therefore, as described in the first embodiment, since the edge router operates as if only one content server 40 is connected, the same effect as in the first embodiment can be obtained. Further, since the connection distribution device 30 can connect a plurality of edge routers, the connection distribution device 30 can be applied to networks of various forms (topologies).

Further, the connection distribution device 30 is a WA
Not only the edge router 21 but also a plurality of normal IP connection devices (for example, firewall, authentication server, etc.) are connected to the physical port on the N side, that is, the edge router 21.
Even if a normal IP connection device is connected between the connection distribution device 30 and the connection distribution device 30, the same effect as above can be obtained.

Further, the round-robin method and the priority method have been mentioned as the method of determining the mirror server of the connection destination of the connection, but other than these methods, for example, a random method of randomly selecting the connection destination of the connection is used. It may be law or the like.

[0135]

As described above, according to the present invention, there is provided a load balancing method in a system in which a client receives content transfer from any one of a plurality of mirror servers via a communication network and a network connection device. To the network connection device from the one mirror server to which the right is assigned, and assigns the destination MAC address to the one of the plurality of mirror servers. The request to make an address is sent. This allows you to evenly distribute the load on multiple mirror servers.
It is possible to further reduce the content transfer processing time on the client side.

Further, according to the present invention, the information of the ARP cache of the network connection device, which is transmitted from one mirror server when the right is assigned, is automatically updated to the IP address and MAC address of the one mirror server. When a packet of a new connection from a client, which is transferred from the network connection device, is transmitted to the network connection device, the “unfounded ARP request” is received, and the packet is transferred to one mirror server that owns the right. It was transferred, and a connection was established between one client and one mirror server to execute application communication. As a result, in order to realize the load distribution method of the present invention, it is not necessary to make any special setting for the network connection device, and the burden on the system administrator can be reduced.

Further, according to the present invention, when a right is returned from one mirror server when a connection is established and the returned right is received, the right is assigned to another mirror server, and at the same time, from the other mirror server. When a "groundless ARP request" that is sent is transferred to the network connection device and a packet for a new connection from another client that is transferred from the network connection device is received, the packet is actually held by the right. Transferred to another mirror server and established a connection between another client and another mirror server to execute application communication. As a result, the connection from the client can be surely allocated to the plurality of mirror servers, and the load on the plurality of mirror servers can be surely distributed.

Further, according to the present invention, there is provided a content distribution system in which a client receives a content transfer from any one of a plurality of mirror servers via a communication network and a network connection device, the network connection device and the plurality of mirror servers. A connection distribution device is provided between the mirror server and the mirror server, and when the right to allocate the connection is assigned from the connection distribution device, the mirror server sends an "irrelevant ARP request", and the connection distribution device detects The right is assigned to one of the mirror servers, and a new connection from the client is assigned to the one mirror server that holds the right. With this configuration, it is possible to evenly distribute the load on the plurality of mirror servers and to further shorten the content transfer processing time on the client side.

Further, according to the present invention, the connection allocation device includes a right allocation processing section for allocating a right to one mirror server among the plurality of mirror servers, and one mirror server among the plurality of mirror servers. Between the client and one mirror server, and a table that stores the correspondence relationship between the IP address and port number of the client that has already established a connection between them and the MAC address of the one mirror server to which the connection is established. Connection monitoring processing unit that constantly monitors connection and disconnection, adds correspondence to the table when the connection is connected, and deletes the correspondence from the table when the connection is disconnected, and transfer from the network connection device Packet sent from the client to one mirror server MAC to convert the destination MAC address, actually from other MAC address of the mirror server owns the rights to the MAC address of one server mirror
It is composed of an address conversion processing unit and a switching processing unit that performs switching based on the destination MAC address of the packet. With this, it is possible to surely configure the connection distribution device that appropriately distributes the connection from the client to the plurality of mirror servers, and it is possible to more easily achieve the above effect.

Further, according to the present invention, the switching processing unit receives the "irrational AR received from one mirror server.
Since the "P request" is not transferred to another mirror server but is transferred to the network connection device, it is possible to surely make the network connection device recognize the one mirror server that has the right. It is possible to avoid unnecessary frame transfer from one mirror server to another mirror server.

Further, according to the present invention, since the one mirror server that holds the right is configured to return the right when the response packet to the client is sent, the right is surely given to the other mirror server. It can be handed over, and as a result, a new connection from the client can be reliably distributed to other mirror servers.

Further, according to the present invention, the client is arranged between the network connection device and the plurality of mirror servers in the system in which the content is transferred from any of the plurality of mirror servers via the communication network and the network connection device. A load balancer,
The load balancer is configured to allocate a right to one mirror server among a plurality of mirror servers and allocate a new connection from a client to the one mirror server that holds the right. As a result, the load applied to the plurality of mirror servers can be evenly distributed, and the content transfer processing time on the client side can be further shortened.

Further, according to the present invention, the load balancer includes a right allocation processing unit for allocating a right to one mirror server among the plurality of mirror servers and one mirror server among the plurality of mirror servers. Between the client and one mirror server, and a table that stores the correspondence relationship between the IP address and port number of the client that has already established a connection between them and the MAC address of the one mirror server to which the connection is established. Connection monitoring processing unit that constantly monitors connection and disconnection, adds correspondence to the table when the connection is connected, and deletes the correspondence from the table when the connection is disconnected, and transfer from the network connection device Of packets sent from the client to one mirror server A MAC address conversion processing unit that converts the MAC address from the MAC address of another mirror server that currently holds the right to the MAC address of one mirror server, and a switching processing unit that switches based on the destination MAC address of the packet. Composed from. As a result, it is possible to surely configure the load distribution device that distributes the load applied to the plurality of mirror servers, and it is possible to more easily achieve the above effect.

[Brief description of drawings]

FIG. 1 is a diagram showing a network configuration to which a content distribution system of the present invention is connected.

FIG. 2 is a block diagram showing a configuration of a connection distribution device.

FIG. 3 is a diagram showing an example of a connection-to-MAC address correspondence table.

FIG. 4 is a block diagram showing a configuration of a mirror server.

FIG. 5 is a diagram showing a protocol stack used by a mirror server for communication.

FIG. 6 is a diagram showing a frame structure of “Unfounded ARP Request”.

FIG. 7 is a sequence diagram for explaining a token allocation operation in the content distribution system of the present invention.

FIG. 8 is a sequence diagram for explaining a connection allocation operation in the content distribution system of the present invention.

FIG. 9 is a sequence diagram for explaining the token allocation operation after the token is returned in the content distribution system of the present invention.

FIG. 10 is a sequence diagram for explaining another new connection allocation operation in the content distribution system of the present invention.

FIG. 11 is a sequence diagram for explaining a MAC address conversion operation in the content distribution system of the present invention.

FIG. 12 is a sequence diagram for explaining an operation of deleting an entry by ARP request / ARP response and connection termination in the content distribution system of the present invention.

[Explanation of sign]

10 client 21 edge router (network connection device) 22 ARP cache 30 connection distribution device (load distribution device) 34 token allocation processing unit 35 connection monitoring processing unit 36 connection-to-MAC address correspondence table (table) 37 MAC address conversion processing unit 38 switching Processing unit 40 Content server 40-1 to 40-4 Mirror server 100 Internet (communication network)

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) G06F 15/177 674 G06F 15/177 674A H04L 12/56 100 H04L 12/56 100Z F term (reference) 5B045 BB28 BB42 GG02 5B085 BE07 BG03 BG07 5B089 GA11 GA31 GB03 JA33 KA06 KB06 MA03 5K030 HA08 HD09 LB05 LE03 5K033 BA13 CA15 CB06 CC01 DA06 DB18 EC03

Claims (9)

[Claims] 1. A load balancing method in a system in which a client receives content transfer from any of a plurality of mirror servers via a communication network and a network connection device, wherein the right from which the connection from the client is allocated is The request is assigned to one of a plurality of mirror servers, and the one mirror server to which the right is assigned sends a request for making the destination MAC address the MAC address of the one mirror server to the network connection device. A load balancing method characterized by: 2. Automatically updating the information of the ARP cache of the network connection device, which is transmitted from one mirror server when the right is assigned, to the IP address and MAC address of the one mirror server. AR
P request ”is transferred to the network connection device, and a packet of a new connection from the client, which is transferred from the network connection device, is received,
The load distribution according to claim 1, wherein the packet is transferred to the one mirror server that holds the right, and a connection is established between the one client and the one mirror server to execute application communication. Method. 3. When a right is returned from the one mirror server when a connection is established, and when the returned right is received, the right is assigned to another mirror server and transmitted from the other mirror server. When an "unfounded ARP request" is transferred to the network connection device and a new connection packet from another client transferred from the network connection device is received, the packet is actually held by the right. 3. The load distribution method according to claim 2, wherein the load is transferred to the other mirror server, and a connection is established between the other client and the other mirror server to execute application communication. 4. A content distribution system in which a client receives content transferred from any of a plurality of mirror servers via a communication network and a network connection device, wherein the content distribution system is provided between the network connection device and the plurality of mirror servers. A connection distributor is provided, and the mirror server is assigned a right to allocate a connection from the connection distributor,
Sending an "unfounded ARP request", the connection distribution device assigns the right to one mirror server of the plurality of mirror servers, and a client to one mirror server that holds the right. Content distribution system characterized by allocating new connections from. 5. The connection distribution device is configured to establish a connection between a right allocation processing unit that allocates a right to one mirror server among a plurality of mirror servers and one mirror server among the plurality of mirror servers. IP of the client
A table that stores the correspondence between the address and port number and the MAC address of the one mirror server to which the connection is established, and constantly monitors the connection and disconnection of the connection between the client and the one mirror server. Then, a connection monitoring processing unit that adds the correspondence relationship to the table when a connection is connected and deletes the correspondence relationship from the table when the connection is disconnected, and is transferred from the network connection device, A MAC address for converting the destination MAC address of the packet transmitted from the client to the one mirror server from the MAC address of another mirror server that currently holds the right to the MAC address of the one mirror server. A conversion processing unit and a destination MAC address of the packet Content delivery system according to claim 4, further comprising a switching unit for switching on the basis of the scan. 6. The content according to claim 4 or 5, wherein the switching processing unit transfers the “unfounded ARP request” received from one mirror server to the network connection device, not to the other mirror server. Delivery system. 7. The content distribution system according to claim 4, wherein the one mirror server that owns the right returns the right when sending a response packet to the client. 8. A load balancer disposed between the network connection device and the plurality of mirror servers in a system in which a client receives a content transfer from any one of the plurality of mirror servers via a communication network and a network connection device. A device, wherein the load balancer allocates the right to one mirror server of the plurality of mirror servers and allocates a new connection from a client to the one mirror server that holds the right. A load balancer characterized by the above. 9. A right allocation processing unit that allocates a right to one mirror server among a plurality of mirror servers, and a client that has already established a connection between one right mirror server among the plurality of mirror servers. IP
A table that stores the correspondence between the address and port number and the MAC address of the one mirror server to which the connection is established, and constantly monitors the connection and disconnection of the connection between the client and the one mirror server. Then, a connection monitoring processing unit that adds the correspondence relationship to the table when a connection is connected and deletes the correspondence relationship from the table when the connection is disconnected, and is transferred from the network connection device, A MAC address for converting the destination MAC address of the packet transmitted from the client to the one mirror server from the MAC address of another mirror server that currently holds the right to the MAC address of the one mirror server. A conversion processing unit and a destination MAC address of the packet Load balancer according to claim 8, further comprising a switching unit for switching on the basis of the scan.