JP2012108685A - Load distribution system - Google Patents

Abstract

【Task】
When a part of a server performing session management is systematically stopped, the continuation request for that server is distributed to another server.
[Solution]
The present invention determines a continuation request addressed to a stop target server after the stop time is exceeded, determines a new distribution server, the new distribution server acquires memory information on the stop target server, and then sends the continuation request to the new distribution server Send to. In addition, a client that does not issue a continuation request even after a certain period of time has elapsed is specified, and the memory information is forcibly moved.
[Selection] Figure 1

Description

  The present invention relates to a load distribution system including a plurality of clients, a plurality of servers that manage sessions from requests from the clients, and a load distribution apparatus that distributes loads of the plurality of servers.

  In a load distribution system including a plurality of clients, a plurality of servers, and a load distribution apparatus, the client sequentially reads pages on the server, inputs necessary data on each page, and transmits the data to the server. For this reason, an application running on the server performs session management in order to associate a request from a client with data on a memory. As a session management method, each server generates a session ID for each page, embeds it in response data, and returns it to the client. At the next request, the client places the session ID embedded by the server in the request data and transmits it to the server.

  By examining the session ID sent from the client, it is possible to determine whether the authentication has been completed or the session ID generated by itself, and proceed with the processing of a request with a valid session ID. Can do. However, in a session-managed system, if you want to stop a specific server for maintenance, etc., you can prevent the impact on the client by changing the distribution destination to a server other than the scheduled stop server. A continuation request from before cannot be distributed to other servers. If the distribution destination is simply changed, an error occurs because there is no client session information (memory information) requested from the changed server.

  Therefore, Patent Document 1 discloses a method of moving session information corresponding to a continuation request to another server and allocating the request to the moved server.

JP2007-219964

  However, the change of the distribution destination by the conventional session information moving method has the following problems. In the conventional technology, a continuation request after the stop time is exceeded is distributed to a server that has been connected once and processed by the server. After the process is completed, the session information is moved from the server to the server that will be distributed next time. When the movement of the session information is completed, the server returns Cookie information for identifying the distribution server to the embedded client in the response data. This cookie information describes the next distribution destination server.

  In this method, the server generates a cookie and embeds it in the response data, and a load distribution apparatus that is generally used cannot use the function of generating the cookie. In general, when a first request arrives from a client, the load balancer determines a server to distribute and transmits the request to the server. When the server performs processing and returns response data to the client, the load balancer generates a cookie and embeds it in the response data. The client sends a request with the cookie generated by the load balancer at the next request. The load balancer analyzes the cookie in the request to determine a server to be connected, deletes the cookie, and transmits the request to the server.

  In order to embed a cookie instructing the distribution destination in the response data and let the load balancer understand the load distribution device while performing such processing, there are two different cases unless cookie generation in the load balancer is stopped. There is a high possibility that the distribution destination is included in the response data, and accurate distribution cannot be performed.

  Also, in the conventional technology, session information can be moved if a continuation request comes after the stop time is exceeded, but session information cannot be moved if the request does not come for a while after the stop time is exceeded. Can not stop the server to rescue the client.

  Even if all the session information is forcibly moved, there is no means to return a cookie indicating the next distribution destination to the client, so the client continues to be distributed to the server to be stopped. In addition, the server newly selected as a distribution destination cannot be processed because a request is suddenly received from a new client that is not managed by itself.

  The present invention has been made to solve such problems, and an object of the present invention is to provide a load distribution system that can distribute a continuation request to another server without affecting a client.

  In order to achieve the above object, the present invention takes the following measures.

  The invention of claim 1 includes a plurality of clients, a plurality of servers, and a load distribution device that distributes requests from the plurality of clients and distributes the requests to the servers, and selects a stop target server, a stop time, and a waiting time until memory information is forcibly moved. A distribution management means for monitoring the scheduled stop time of the stop target server, setting a connection prohibition flag for the stop target server and determining a new distribution server when the scheduled stop time is reached, and a client Analyzing the connection request and categorizing the processing contents such as whether to connect to the server to be stopped or distributed to a new server, and when the connection information management unit determines that the server to be distributed needs to be changed Memory information from the server to be stopped for the server selected by the server distribution management means A server determination unit that transmits to the new server a pending continuation request when the movement of the memory information is completed, and a transmission / reception unit that transmits the data sent from the server determination unit to the LAN, A memory information acquisition unit that issues a memory information acquisition request to the server to be stopped when a memory information acquisition instruction is received from the server determination unit; and a memory information response that is received from the memory information acquisition unit Means are provided.

  The invention of claim 2 specifies a client that needs to move memory information based on connection information recorded in the load balancer when a continuation request is not transmitted even after the waiting time input by the input means has elapsed. The memory information on the stop target server reserved for the client is forcibly moved to the server determined as the new distribution destination.

According to the present invention, after the stop time is exceeded, the continuation request to the stop target server can be distributed to the newly selected server and the processing can be continued without interruption.

Configuration diagram showing a configuration example of a load balancing system Block diagram of load balancer Functional diagram of load balancer Flow chart of transceiver Flowchart 1 of connection information management unit Flowchart 2 of the connection information management unit Flowchart 1 of server determination unit Flowchart 2 of server determination unit Flowchart 3 of server determination unit Flowchart 4 of server determination unit Flow chart of distribution management unit State management table Server management table Inter-server communication status table Outline operation sequence diagram

  Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings.

  As shown in FIG. 1, the load distribution system according to the present embodiment includes a plurality of clients 100 (# 1, # 2,... #N) and a plurality of servers 102 (# 1, # 2,... #N). Load distribution device that distributes connection requests from clients 100 (# 1, # 2,... #N) to each server so as to distribute the load of servers 102 (# 1, # 2,... #N) This is applied to a load distribution system 10 having 101.

  Since there are many types of applications in which a client sequentially calls pages on a server and inputs data on each page, requests from the same client must be connected to the same server from the start to the end of business. If you connect to another server in the middle of business with this type of application, there will be no data entered by the client and an error will occur. Connecting a request from the same client to the same server in this way is called session maintenance.

  In a business system that maintains a session in this way, when a specific server is scheduled to be stopped for maintenance, the load distribution system 10 monitors the stop time and plans to stop when the scheduled stop time is reached. Set a connection prohibition flag for the server, and prohibit subsequent connections to the server for both new and continued requests. When a connection prohibition flag is set for a server that is a selection target by the load balancing algorithm, the server is excluded from the selection target and another server is determined as a new distribution destination.

  Another server determined as a new distribution destination is hereinafter referred to as a new server.

  In addition, when a continuation request arrives at the server for which the connection prohibition flag is set, the new server is instructed to acquire memory information from the server connected so far (hereinafter referred to as the old server). In response to the memory information acquisition instruction, the new server issues a memory information acquisition request to the old server to acquire memory information. After the memory information acquisition is completed, the load balancer 101 transmits a continuation request from the client that has been suspended to the new server.

  The clients 100 (# 1, # 2,... #N), the load balancer 101 and the servers 102 (# 1, # 2,... #N) are connected to the LAN 103 and are connected from a plurality of subnets via routers. Sometimes configured. In order to realize this, the load distribution apparatus 101 has a configuration shown in FIGS.

  First, an outline of the operation of this embodiment will be described with reference to FIG.

  The description here is an example where the old server 1203 is a scheduled stop server and the stop time is reached between the sequence 55 and the sequence 56.

  In a business system that maintains a session, 1201 is a client that transmits a request, 1202 is a load balancer that distributes processing, 1203 is an old server, and 1204 is a new server. When a connection request is received from the client 1201 to the virtual IP address of the load balancer 1202, a connection is established (sequence 50). After the connection is established, a new request is transmitted from the client 1201 to the virtual IP address (sequence 51). When receiving a new request, the load distribution apparatus 1202 issues a connection request to the old server 1203 to establish a connection (sequence 52), and transmits a request from the client to the old server 1203 (sequence 53). When receiving the request, the old server returns a response to the load balancer 1202 accordingly (sequence 54). The load balancer 1202 returns a response from the old server 1203 to the client 1201 (sequence 55).

  At this time, since the old server 1203 reaches the stop time, the distribution is prohibited. The new server 1204 is determined as a new distribution server.

  The client sends a continuation request (sequence 56) to the virtual IP address, but since distribution to the old server 1203 is prohibited, the load balancer 1202 temporarily holds the request and sends it to the new server 1204 An instruction is issued to acquire memory information on the old server reserved for the requesting client (sequence 57, sequence 58). In accordance with the instruction, the new server 1204 issues a connection request for acquiring memory information of the old server to the virtual IP address (sequence 59). When the connection is established, a memory information acquisition request is similarly sent to the virtual IP address (sequence 60).

  When receiving the memory acquisition request from the new server, the load balancer 1202 issues a connection request to the old server and establishes a connection (sequence 61). After the connection is established, a memory acquisition request is transmitted to the old server 1203 (sequence 62). When the old server 1203 receives the request, it sees the session information of the client 1201 in the request data and responds with the corresponding memory information (sequence 63). When the new server receives this response (sequence 64), the memory information can be acquired.

  When the acquisition of the memory information is completed, the load distribution apparatus 1202 issues a connection request to the new server and establishes a connection (sequence 65). Thereafter, a request is transmitted from the suspended client (sequence 66). The new server 1204 that has received the request performs processing and returns a response (sequence 67). The load balancer 1202 returns a response from the new server 1204 to the client (sequence 68).

  Details of the present embodiment will be described below.

FIG. 2 is a schematic configuration diagram of the load balancer 101, and the clients 100 (# 1, # 2,.
.. #N) and the server 102 (# 1, # 2,... #N) are connected by the LAN 103, the switch unit 200, the controller 201 that controls the switch unit 200, and the distribution server determination for the connection request from the client Instructions for acquiring memory information from the old server to the new server, monitoring the scheduled stop time of the scheduled stop server, analyzing the packet and determining the processing content, the memory 203 used by the CPU unit 202, and various information It comprises a disk unit 204 for storing.

  FIG. 3 is a functional block diagram provided in the load balancer 101 that performs transmission / reception of packets, analysis of packets, determination of a distribution destination server, monitoring of stop time, and input of setting information. The operation of the transmission / reception unit 301 will be described using the flowchart of FIG.

  A packet addressed to the virtual IP address of the load balancer 101 input to the load balancer 101, a response from the server, and a packet passing through the load balancer 101 are received (step S1) and transmitted to the connection information management unit 302 (step S1). S2). Also, the packet and data processed by the server determination unit 303 are received (step S3), and it is determined with reference to the connection source IP address 801 and connection source port number 802 in the state management table of FIG. (Step S4). Since the connection source IP address 801 and the connection source port number 802 are registered immediately after the connection is established, the connection source IP address 801 and the connection source port number 802 are determined to be already established.

  If the connection has been established, the packet received from the server determination unit 303 is transmitted to the LAN 103 (step S5). If the connection has not been established, connection establishment processing is performed with the client or server (step S6), and the data received from the server determination unit 303 is transmitted to the LAN 103 (step S5).

  An example of connection establishment is a TCP three-way handshake.

  The operation of the connection information management unit 302 will be described using the flowcharts of FIGS. 5A and 5B.

  The connection information management unit 302 receives the packet sent from the transmission / reception unit 301 (step S11) and analyzes the packet (step S12). First, it is determined whether the packet is addressed to the virtual IP address of the load balancer 101 (step S13).

  The virtual IP address may be abbreviated as VIP, and a plurality of IPs of the server 102 (# 1, # 2,... #N) with respect to the client 100 (# 1, # 2,... #N). It has a role of showing an address as one IP address.

  If it is not addressed to the virtual IP address, it is a packet that passes through the load balancer 101 or a response from the server, so 1 is set in the forward flag that passes the packet (step S20), and the server determination unit 303 is notified ( Step S19).

  The operation of the server determination unit 303 will be described using the flowcharts of FIGS. 6A, 6B, and 6C.

  When the server determination unit 303 receives data in which the processing content flag is set from the connection information management unit 302 (step S41), the server determination unit 303 analyzes the set flag (step S42).

  When the forward flag is 1 (step S43), it is determined whether the packet passes through the load balancer 101 or the response from the server (step S44).

  If the result of determination in step 44 is not a response from the server, it is a packet that simply passes through the load balancer 101, so that the transmission / reception unit 301 is notified without performing any processing (step S46). Whether or not the response is from the server, the IP address and port number registered with reference to the connection destination IP address 804 and connection destination port number 805 in the state management table of FIG. 8 are used as the source IP address and source port number. Can be determined by whether or not

  When receiving the request addressed to the virtual IP address from the client, the load balancer 101 converts the destination IP address into the IP address of the selected server and transmits it to the server. At the time of response, the server IP address is designated as the source IP address, and the client IP address is designated as the destination IP address. The same applies to the port number. For this reason, if the IP address of the connected server is used as the source IP, it can be determined that the response is from the server.

  If the result of determination in step 44 is a response, response return processing is performed (step S45). There are two types of responses from the server: a response for moving memory information from the old server (example: sequence 63) and a response from the server to the request from the client (example: sequence 54). For this reason, which response is determined.

  The source IP address and connection destination IP address 804 of the response are compared, and the source port number and connection destination port number 805 are compared, and the destination IP address is the load balancing target server 102 (# 1, # 2,... If #n), it can be determined that the response is a memory information transfer from the old server.

  If the destination IP address is not the load distribution target server 102 (# 1, # 2,... #N), it can be determined that the response is a response to the client request.

  As a response return process for the request from the client, the source IP address is changed from the server IP address to the virtual IP address of the load balancer 101. The same replacement process is performed for the MAC address, and the transmission / reception unit 301 is notified (step S46). Since the client issues a request to the virtual IP address of the load balancer 101, the client recognizes that a response has come from another device and cannot receive a response unless the source IP address is converted into a virtual IP address at the time of response. The same applies to the MAC address.

  The response return process from the old server for moving the memory information is the same.

  If the result of determination in step 13 is that the packet is addressed to the virtual IP address, the connection information management unit 302 is a connection request from a client or a connection request from a new server for acquiring memory information. It is determined whether the address is the IP address of the load balancing target server 102 (# 1, # 2,... #N) (step S14). The IP address of the load distribution target server 102 (# 1, # 2,... #N) is acquired by referring to the connection destination server IP address 902 of the server management table of FIG.

  If the source IP address is not the address of the load balancing target server 102 (# 1, # 2,... #N), it is a connection request from the client 101 (# 1, # 2,... #N). Whether the request is a new request or a continuation request is determined (step S15). Since the connection destination server name 803 to the connection destination port number 805 in the state management table are registered when a new quest is assigned to the server, if it is not registered, it can be determined that the request is a new request.

  If the result of determination in step S15 is that it is not a new request, it will be a continuation request, so it is determined whether or not the stop time has exceeded based on the notification from the distribution management unit 304 (step S16). Since notification is received from the distribution management unit 304 when the stop time is reached, it is determined whether or not the stop time is exceeded based on the presence or absence of the notification.

  If the stop time has not been reached, the load distribution flag is set to 1 to execute the continuation request process (step S18), and the server determination unit 303 is notified (step S19).

  The server determination unit 303 analyzes the received data flag, recognizes that the load distribution flag is 1 (step S67), and refers to the connection source IP address 801 and connection source port number 802 in the state management table. The client request is specified (step S68), and the server to be connected is specified with reference to 804 (step S69).

  After the server is determined, source IP address conversion and MAC address conversion are performed as transfer processing to the server (step S70), the load distribution processing flag is set to 0 (step S71), and the transmission / reception unit 301 is notified (step S46). .

  If the result of determination in step S16 indicates that the stop time has been exceeded, the connection information management unit 302 needs to change the distribution destination. Therefore, the connection information management unit 302 temporarily holds the continuation request from the client on the memory or the disk. The server change flag is set to 1 (step S17), and the server determination unit 303 is notified (step S19). The temporarily held continuation request is processed after the end of the memory movement between the new server and the old server.

  When the server change flag is determined to be 1 by the server determination unit 303 (step S81), the server 904 refers to the distribution target server 904 (step S82) and determines the distribution destination server (step S83). The server to be selected is a server to which 1 is set in the distribution target server 904. In the case of round robin, the servers to be selected in order are changed. Note that other methods may be designated as the load distribution method.

  The operation of the distribution management unit 304 will be described using the flowchart of FIG.

  The distribution management unit 304 monitors the stop time input through the input unit 305 (step S201), and when the stop time is reached, notifies the connection information management unit 302 and the server determination unit 303 that the stop time has been exceeded (step S202). ), 9 is set in the connection prohibition flag 905 of the server management table of FIG. 9 for the server to be stopped (step S203). If the server for which the connection prohibition flag is set is a selection target (1 is set in the distribution target server 904) (step S204), 0 is set in the distribution server 904. When the distribution target server 904 is changed from 1 to 0 and there is no selection target server, 1 is set so that the next server is selected as a new selection target (step S205).

  The server determination unit 303 creates an inter-server communication activation packet that instructs the new server to acquire the memory information of the client that has issued a continuation request on the memory of the old server after the new server is determined (step S84). ). This instruction packet includes session information issued by the server extracted from the pending client request and a memory information acquisition instruction flag set by the server determination unit 303. After creating the inter-server communication start packet, the server change flag is set to 0 (step S85), and the transmission / reception unit 301 is notified (step 46).

  If the result of determination in step S15 is a new request, the connection information management unit 302 sets a new load distribution flag to 1 (step S21), and notifies the server determination unit 303 of it.

  For example, the HTTP method such as POST or GET is determined as a request, and if the connection destination server name 803 to the connection destination port number 805 in the state management table of FIG.

  The connection destination server name 803 is the name of the server that performed the distribution, and is defined in the setting information of the load balancer 101. The connection destination server IP address 804 is the IP address of the server that performed the distribution, and this information is also defined by the setting information of the load balancer 101. A connection destination port number 805 is a service port number to which connection has been made, and 80 indicates HTTP. Since the connection destination server name 803 to the connection destination port number 805 are registered when a new quest is assigned to the server, if it is not registered, it can be determined that the request is a new request.

  When the server determination unit 303 determines that the new load distribution flag is 1 (step S61), the virtual IP address specified as the destination is selected with reference to the virtual IP address 901 in the server management table of FIG. Since the distribution target server is specified for each virtual IP address, the distribution target server 904 is referred to and updated in order to select the server 902 belonging to the group of the selected virtual IP address 901 (step S62). The server to be selected is described in the distribution target server 904, and the server for which the distribution target server 904 is 1 is selected (step S63). After the server is selected, the selected server is set to 0, and 1 is set to the server to be selected next time based on the load distribution method of the load distribution method 903 of the state management table.

  Since the information on the virtual IP address 901, the server name 902, and the load distribution method 903 is defined in the setting information of the load distribution apparatus 101, it is extracted from the setting information and a state management table is created.

  After the server is determined, the server information to be connected to the connection destination server name 803 to the connection destination port number 805 in the state management table is registered (step 64), and connection processing with the server is performed (step S65). Since the client transmits a request to the virtual IP address of the load balancer 101, it is necessary to perform address conversion of the server that has selected the destination IP address and MAC address, and therefore performs address conversion processing. After the process of step S65 is completed, the new load distribution flag is set to 0 (step S66), and the transmission / reception unit 301 is notified (step S46).

  If the result of determination in step 14 is that the source IP address is the load distribution target server 102 (# 1, # 2,... #N), the connection information management unit 302 stores the memory information of the new server and the old server. Since it is a takeover process, the process status is determined by the packet acquired via the transmission / reception unit 301 (step S24), and the new server 1002 and the old server 1003 in the inter-server communication status table of FIG. 10 are updated (step S25). . Reference numeral 1001 in the inter-server communication state table indicates items to be monitored. In this embodiment, FIN and ACK for FIN are monitored. For example, if the FIN flag from the new server is set, 1 is set in the FIN column of the new server 1002. In this way, when 1 is set in all ACKs for the FIN and FIN of the new server and the old server, it is determined that the connection is closed.

  When it is determined that the processing state is the connection closed (step S26), it indicates the end of the movement of the memory information, so that the server determination unit transmits the continuation request from the client held by the load balancer 101 to the new server. 303 is notified (step S19).

  When the server determination unit 303 determines that the inter-server communication flag is not 1 (step 86), information on the new server is registered in the state management table (step 89), and the continuation request from the suspended client is sent to the new server. The connection process is executed (step 90) and the transmission / reception unit 301 is notified (step 46).

  If the processing status is not connection closed, the connection information management unit 302 indicates that the movement of memory information between the new server and the old server has not been completed. The inter-communication flag is set to 1 (step S27), and the server determination unit is notified (step S19).

  If the server determination unit 303 determines that the server-to-server communication flag is 1 (step S86), the server-to-server communication process is performed (step S87) because the communication is between the new server and the old server for acquiring memory information.

  The new server that has received the packet created in step 84 starts communication to acquire memory information of the old server. First, the new server issues a connection request to the virtual IP address of the load balancer 101 and establishes a connection (sequence 59). Thereafter, the new server transmits a memory information acquisition request to the virtual IP address (sequence 60). This request includes the session information included in the client request embedded in the packet created in step 84.

  The load distribution apparatus 101 that has received the request refers to the session information 806 in the state management table and identifies the old server. When the old server can be identified, the destination IP address is converted to the IP address of the old server, a connection request is issued to the old server, a connection is established (sequence 61), and a request from the new server is transmitted (sequence 62). ).

  The old server that has received the request from the new server returns the corresponding memory information as a response based on the session information included in the request (sequence 63). The load balancer 101 that has received the response from the old server converts the destination IP address converted at the time of transmission into a virtual IP address and transmits the response to the new server (sequence 64).

  When the processing in step S87 is completed, the server-to-server communication flag is set to 0 (step S88), and the transmission / reception 301 is notified (step S46).

  As a result of the determination in step S86, if the inter-server communication flag is not 1, information on the new server is registered in the state management table (step 89), and a continuation request from the suspended client is connected to the new server. (Step 90), and notifies the transmitter / receiver 301 (step 46).

  The operation of the server determination unit 303 according to the second embodiment of this invention will be described using the flowchart of FIG. 6D.

  For clients that have accessed immediately after exceeding the scheduled stop time, the connection server can be changed by the above method. However, even if the scheduled stop time elapses, some clients do not access for a while. The waiting time from the stop time is monitored with reference to the waiting time 907 in the management table (step S101). For a client that does not access even after the waiting time specified in waiting time 907 has elapsed, refer to the connection destination server name 803 to the connection destination port number 805 in the state management table (step S102), and select the client. Identify. Next, a new server is determined with reference to the distribution target server 904 in the server management table (step S103), a server-to-server communication activation packet is forcibly created (step S104), and notified to the transmission / reception unit 301 (step S105). ). This process forcibly starts the movement of memory information.

  Even if there is a client that has accessed after exceeding the waiting time, since the memory information has been moved, the processing can be continued.

100 Client 101 Load Balancing Device 102 Server 200 Switch Unit 201 Controller 202 CPU
203 Memory 204 Disk 301 Transmission / reception unit 302 Connection information management unit 303 Server determination unit 304 Distribution management unit 305 Input unit 801 Connection source IP address 802 Connection source port number 803 Connection destination server name 804 Connection destination server IP address 805 Connection destination server Port number 806 Session information system headquarters included in continuation request 901 Virtual IP address 902 set in load balancer IP address 903 of destination server Load distribution method 904 Distribution target server 905 Connection prohibition flag 906 Scheduled stop time 907 Memory information Waiting time until forced movement

Claims (2)

A load distribution system including a load distribution device that distributes connection requests from a plurality of servers and a plurality of clients and distributes the requests to the plurality of servers,
An input means capable of inputting a stop target server, a stop time, and a waiting time until forced movement of memory information in order to systematically stop a specific server among the plurality of servers;
A distribution management means for monitoring a scheduled stop time of the stop target server and setting a connection prohibition flag on the stop target server and determining a new distribution server when the scheduled stop time is reached;
A connection information management means for analyzing a connection request from the client and classifying processing contents such as whether to connect to the server to be stopped or to distribute to a new server;
When the connection information management unit determines that the distribution server needs to be changed, the server selected by the distribution management unit is instructed to acquire memory information from the server to be stopped, and the memory information is moved. Server determination means for sending a continuation request that has been suspended to the new server when
A transmission / reception means for transmitting data sent from the server determination unit to the LAN;
Memory information acquisition means for issuing a memory information acquisition request to the stop target server when receiving an instruction to acquire memory information from the server determination unit;
A load distribution system comprising memory information return means for returning memory information upon receiving a memory information movement request from the memory information acquisition means.   2. The load distribution system according to claim 1, wherein the memory information needs to be moved based on connection information recorded in the load distribution apparatus when a continuation request is not transmitted even after the waiting time input by the input means has elapsed. Identify the client and forcibly move the memory information on the server to be stopped reserved for the client to the server determined as the new distribution destination, and change the connection information in the load balancer And a load balancing system.

Images