JP3825333B2 - Load distribution method using tag conversion, tag conversion apparatus, and load distribution control apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention belongs to the technical field of server load balancing in a computer network.
[0002]
[Prior art]
In a Web site or the like, load distribution and redundancy of servers are important in order to ensure high reliability and availability of servers. In general, the load distribution method includes a method using a domain name system (DNS), a method using a load distribution device, a method using a cluster server, and the like.
[0003]
FIG. 1 is an explanatory diagram of a load distribution method using DNS.
[0004]
The client information terminals 11 and 12 issue a domain name address resolution query to the DNS server 31 when accessing the server. The DNS server 31 performs load distribution for these queries by designating different servers 41 and 42 by round robin or the like.
[0005]
FIG. 2 is an explanatory diagram of a load distribution method using a load distribution apparatus.
[0006]
A virtual server IP address different from the IP addresses of the servers 43 to 45 is set in the load balancer 81. The client information terminals 13 and 14 access the IP address. The load distribution device 81 performs load distribution by redirecting these accesses to different servers from the load states of the servers 43 to 45, for example. Since the load balancer 81 monitors the states of the servers 43 to 45, it is possible to cancel redirection to the failed server when the server fails. Further, in order to increase the reliability of the load balancer 81, it is possible to provide a load balancer 82 for redundancy.
[0007]
FIG. 3 is an explanatory diagram of a load distribution method using a cluster server.
[0008]
The servers 48 to 50 are virtually set with the same IP address and MAC address by the installed adapter software. The client information terminals 17 and 18 access the IP address. In response to access from the information terminal, the servers 48 to 50 perform load distribution by any one server responding according to the load distribution algorithm. When a failure occurs in the servers 48 to 50, the failure server is detected by exchanging heartbeat messages with each other, and the load distribution algorithm is corrected, whereby distribution to other servers becomes possible. In addition, since the servers 48 to 50 are completely parallel in terms of network, it is easy to increase or decrease the number of servers and replace the spare servers 72.
[0009]
[Problems to be solved by the invention]
In the method using DNS, since the DNS server does not know the load state of the server, it may be allocated to a high-load server. In addition, when address resolution is routed through another DNS server in the network, if the address resolution information is cached in that server, the DNS server designates the server based on the cache information, so the load is not distributed. There is a problem that there are cases.
[0010]
In FIG. 1, for example, when the server 41 has a high load and the load of the server 42 has a low load, when the information terminal 11 issues an address resolution query to the server, the DNS server 31 designates the server by round robin or the like. For this reason, the server 41 may be distributed to a high load regardless of the load state of the server. In addition, when the address resolution query reaches the DNS server 31 via the DNS server 32, the DNS server 32 caches the address resolution information. Therefore, other address resolution queries are also cached. There is a problem that the server is specified based on the information and the load cannot be distributed.
[0011]
In the method using the load balancer, when there are a plurality of networks, the load balancer is necessary for each network. For example, when a network is divided between a user's tenant building and a data center by a virtual LAN (Virtual LAN: hereinafter referred to as VLAN) for each user group, a load balancer is installed for each user group in the data center. There is a need to. Since each server has its own IP address, it is necessary to set the IP address of the spare server as the IP address of the failed server in the event of a server failure. Furthermore, in order to install a server shared between different networks (groups), there is a problem that it is necessary to install two physical interfaces on the server or terminate the VLAN on the server.
[0012]
In FIG. 2, for example, when there is a network 3 different from the network 2, the network 3 also needs load distribution devices 83 and 84, SW 92, and servers 46 and 47. Further, since the servers 43 to 47 are assigned their respective IP addresses, when the server 47 fails, it is necessary to set the IP address of the spare server 71 to the IP address of the server 47 and replace it. Further, in order to install the server 45 shared by the group A 141 to which the information terminals 13 and 14 belong and the group B 142 to which the information terminals 15 and 16 belong, the server 45 has two physical IFs to be connected to each load balancer. There is a problem that it is necessary to install one network or terminate the VLAN of the network with a server.
[0013]
In the method using a cluster server, if there are a plurality of networks, it is necessary to install a cluster server for each network. In addition, since each server is completely parallel in terms of network, it is difficult to distribute a request from a specific client to a specific server, and it is also difficult to install a server shared between different networks. Furthermore, since the servers are connected by HUB, the influence of one server failure may spread throughout. As the number of servers increases, the time required for exchanging heartbeat messages with each other and the time required for the correction of the load balancing algorithm to converge increase, which makes it difficult to increase the scale of the servers.
[0014]
In FIG. 3, for example, when there is a network 5 different from the network 4, the network 5 also needs a cluster server of HUB 94 and servers 51 to 53. Further, since the servers 48 to 50 are parallel in the network, it is difficult to set the server 48 so that requests from the information terminal 20 are fixedly distributed to the server 53, and a server that is shared between the network 4 and the network 5 can also be installed. difficult.
[0015]
Furthermore, if the server 48 continues to output a signal to the HUB 93 due to a failure of the server 48, there is a possibility that the influence of one server failure may spread to the whole, that is, communication with the servers 49 and 50 becomes impossible. is there. In addition, as the number of servers increases, the time required for exchanging heartbeat messages with each other and the time required for the correction of the load balancing algorithm to converge increase, making it difficult to increase the scale of the servers.
[0016]
The present invention has been made in view of the drawbacks of the conventional load distribution method as described above, and enables server requests to be distributed according to the load state of the server or external settings, and includes a spare server. The purpose is to provide a load balancing method that makes server settings uniform, makes load balancing devices redundant, shares load balancing devices across multiple networks, installs servers that can be shared between networks, and prevents server failures from affecting the entire server To do.
[0017]
[Means for Solving the Problems]
In order to achieve the above object, in a load distribution method in a network system including a plurality of information terminals and a plurality of servers, a plurality of servers are connected to VLANs having different VIDs, and the load of each server is determined using a load distribution control device. Using a tag conversion device that detects the status and failure and converts the VID of the input packet into the VID of the connection destination VLAN, the VID of the packet from the information terminal is changed to the load status and failure of each server, and the load distribution By performing conversion according to the policy , the server connected to the information terminal is switched to perform load distribution. Specifically, a server to be distributed is determined according to a source IP address, a destination IP address, a load state of each server, an external distribution policy setting, etc., and the VID of the input packet is replaced with the VID of the server, Dispersion is performed (claim 1).
[0018]
Further, since the server is connected to different VLAN, it is possible to have the same IP address to a plurality of servers including the spare server (claim 2), by redundant tags converter The load on the tag conversion device can also be distributed for each access from the information terminal (claim 3). The load distribution control device detects the load state of the server by switching the VID of the packet for detecting the load state of the server to a VID corresponding to the VLAN to which each server is connected and detecting it from the response time or the like. (Claim 4).
[0019]
As a device for achieving the above object, a tag conversion device includes a tag conversion table in which a VID corresponding to a VLAN to which an input packet is connected is described for each VID, IP address, and port number of the input packet, and an input The VID of the packet can be configured to be converted according to the tag conversion table and output, and the tag conversion table can be set according to an instruction from the outside (Claim 5). A tag conversion table describing a VID corresponding to the VLAN to which each server is connected and a packet for detecting the load state of the server are switched and output according to the tag conversion table according to the tag conversion table. It is possible to comprise a means for detecting the state and a means for notifying the detected result to the outside. 6).
[0020]
Further, the load distribution control device can be made unnecessary by detecting the load state and failure of each server using the tag conversion device (claim 7). At this time, the tag conversion device is provided with means for switching the VID to the packet for detecting the load state of the server according to the tag conversion table and outputting it, and detecting the load state of each server based on the response. 8).
[0021]
According to the present invention, by setting the tag conversion table of the tag conversion device based on the load status information of each server from the load balancing control device and the distribution policy to the server, not only the load status of the server, Requests from a specific client can be distributed to a specific server. Further, since each server is connected via a VLAN and logically divided, it is possible to set the same IP address or the like on a plurality of servers including a spare server. Furthermore, since the network to which the user belongs is also logically separated by VLAN, the load distribution of a plurality of networks can be performed by one tag conversion device.
[0022]
By making the tag conversion device redundant, duplicating the server access path from the client, and specifying which tag conversion device is routed by the load balancing control device, the load on the tag conversion device as well as the server is accessed by the client. Can be distributed every time. Each server is logically divided by VLAN, and the tag conversion device is equivalent to a layer 2 switch (L2SW) in terms of network. Therefore, a single server failure affects other servers connected to the HUB, such as a HUB connection. Never do.
[0023]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 4 shows an example of an embodiment of the load distribution method of the present invention.
[0024]
In this embodiment, the clients (information terminals) 21 and 22 pass through the VLAN 101 (VID = 1), the information terminals 23 and 24 pass through the VLAN 102 (VID = 2), and the servers 54 to 58 are connected to the VLANs 103 to 107 ( VID = 3 to 7) is connected to the tag conversion device 121, and the load distribution control device 131 is also connected to the tag conversion device 121.
[0025]
Information terminals 21 and 22 belong to group A145 and servers 54 and 55 are used in group A145. Similarly, information terminals 23 and 24 belong to group B146 and servers 57 and 58 are used in group B146. ing. It is assumed that the group A 145 and the group B 146 are different networks, and the server 56 can be accessed from both the information terminals of the group A 145 and the group B 146.
[0026]
For example, the tag conversion device 121 determines the VID of the server access packet from the information terminal 21 based on the source IP address, VID (= 1), the load state of the server from the load balancing control device 131, and the like.
VID = 1 → VID = 3
Can be distributed to the servers 54. Similarly, by distributing the packet of the information terminal 22 to the server 55, load distribution to the server is possible.
[0027]
When the server 54 fails,
VID = 1 → VID = 4
Failover is possible by switching the servers connected as.
[0028]
Since the servers 54 to 58 are logically separated from each other by VLAN, it is possible to set the same IP address to these, and the information terminals 21 to 24 use only the IP address regardless of the server. Can be accessed. By setting the same IP address as that of the servers 54 to 58 in the spare server 73, even if any of the servers 54 to 58 fails, the spare server 73 can be replaced as it is, and the setting is changed depending on the server to be replaced. There is no need.
[0029]
The load distribution control device 131 can check the load state of each server by sending ping (Packet Internet Grouper) or the like to each server while switching the VID. This load state confirmation function can also be incorporated in the tag converter 121.
[0030]
Although the networks are different between group A 145 and group B 146, each information terminal and server are logically separated by VLAN, so they can be connected by one tag conversion device 121, and a tag conversion device (load distribution device) for each network. Do not need. Also, the server 56 shared between different networks can be accessed from either the group A 145 or the group B 146 by just converting the VID, as with other servers.
[0031]
FIG. 5 is a detailed configuration diagram of the tag conversion device 121 and the load distribution control device 131.
[0032]
A packet input to the packet input unit 211 is converted by the tag conversion unit 212 into a VID of a connection destination VLAN according to the tag conversion table 215 and transmitted from the packet output unit 213. In the tag conversion table 215, the VID of the connection destination VLAN of the input packet is described for each VID and IP address of the input packet. The table setting unit 214 is based on the VID of the input packet of the packet input unit 211, the source IP address, the port number, and the like, information from the load distribution control device 131 input via the communication unit 216, and the like. Set.
[0033]
The load distribution control device 131 outputs a packet such as ping from the load state detection unit 311 to each server via the packet input / output unit 313 while switching the VID by the tag conversion unit 312 according to the tag conversion table 315, and the response time Thus, it is detected whether the load of the server is light / heavy / impossible to connect, for example. The tag conversion table 315 describes the VID corresponding to the VLAN to which each server is connected. The table setting unit 314 rewrites the tag conversion table 315 according to an instruction from the load state detection unit 311 and switches the connection destination (test packet output destination) of the load state detection unit 311.
[0034]
From the distribution setting input unit 317, a load distribution policy such as specification of connection of a specific server or specification of priority of server distribution is input to a specific client. For example, priorities such as designation of connection of the information terminal 21 to the server 54 and high / medium / low / connection impossibility depending on the processing capability of the server are input. The detected server load state and the input load distribution policy are notified to the tag conversion apparatus 121 via the communication unit 316.
[0035]
6 to 8 show the server distribution operation according to the distribution policy together with server status and table setting examples.
[0036]
The information terminal 21 is connected to the server 54 by connection designation of the load distribution control device 131 (FIG. 6). When the information terminal 22 sends a packet, the tag conversion device 121 does not have a corresponding conversion table in the tag conversion table 215, so the source IP address, VID, port number of the information terminal 21, and the server of the load distribution control device 131. Set the conversion table based on the load status and priority. Here, since the load states of the servers are the same, they are distributed to the servers 55 having higher priority (FIG. 7). The conversion table 215 is held for a while after the packet from the information terminal 22 is not transmitted, but is deleted when the aging time has passed (FIG. 8).
[0037]
FIG. 9 to FIG. 11 show the operation at the time of distribution change due to a server failure together with server status and table setting examples.
[0038]
The information terminal 22 is connected to the server 55 (FIG. 9), but when the server 55 fails (FIG. 10), the tag conversion device 121 distributes the information terminal 22 to another server. Specifically, the load distribution control device 131 detects a failure and notifies the tag conversion device 121, and the tag conversion device 121 distributes the server 50 based on the server priority (FIG. 11).
[0039]
FIG. 12 shows another example of the embodiment of the load distribution method of the present invention, in which the tag conversion device is made redundant, that is, the load of the tag conversion device is distributed using the two tag conversion devices 122 and 123. An example is shown.
[0040]
FIGS. 13 to 15 show the operation at the time of distribution change due to a tag converter failure together with server status and table setting examples.
[0041]
The server access route from the information terminal 26 is set via the tag conversion device 122 according to the load distribution policy from the load distribution control device 132. Specifically, the tag conversion device 122 transmits the packet of the information terminal 26 to the server 60 by converting the tag, and the tag conversion device 123 discards the packet by not setting the connection destination (see FIG. 13). If the load distribution control device 132 detects that the connection of the server 60 via the tag conversion device 122 is not connectable and the connection via the tag conversion device 123 is possible (FIG. 14), the load distribution policy is changed and tag conversion is performed. Switch to device 123. As a result, the tag conversion device can be made redundant (FIG. 15).
[0042]
FIG. 16 shows still another example of the embodiment of the load distribution method of the present invention, here, an example using the tag conversion device 124 having the function of the load distribution control device.
[0043]
FIG. 17 is a detailed block diagram of the tag conversion device 124 in FIG.
[0044]
The tag conversion device 124 outputs a packet such as ping from the load state detection unit 246 to each server while switching the VID by the tag conversion unit 242 according to the tag conversion table 245, and detects the load state from the response time. Also, in the tag conversion table 245, the VID of the connection destination VLAN of the input packet is described for each VID and IP address of the input packet. The table setting unit 244 sets the tag conversion table 245 based on the server switching instruction of the load state detection unit 246, the detection result of the server load state, and the distribution designation input from the outside via the distribution setting input unit 247. Thus, detection of the load state of each server and load distribution are realized.
[0045]
As described above, the tag conversion device 124 can also have the function of the load distribution control device by including the load state detection unit 246.
[0046]
【The invention's effect】
By dynamically switching the VLAN path to the server, it is possible to provide a highly reliable and highly available server system. By setting a table for switching VLAN paths, it is possible to set an external distribution policy while distributing the load.
[0047]
Since each server is logically divided by VLAN, it is possible to set the same IP address, and setting the spare server to the same setting facilitates switching between active / spare.
[0048]
Since each network and server are logically separated by VLAN, a load distribution device (tag conversion device) can be shared by a plurality of networks, and a server that can be shared between networks is easy. At this time, the tag conversion apparatus is equivalent to L2SW, and there is no problem that one failure is spread to the whole as in the case where the server is HUB-connected.
[Brief description of the drawings]
FIG. 1 is an explanatory diagram illustrating an example of a conventional load distribution method. FIG. 2 is an explanatory diagram illustrating another example of a conventional load distribution method. FIG. 3 is an explanatory diagram illustrating still another example of the conventional load distribution method. FIG. 4 is an explanatory diagram showing an example of an embodiment of a load distribution method of the present invention. FIG. 5 is a configuration diagram showing details of a tag conversion device and a load distribution control device in FIG. 4. FIG. FIG. 7 is an explanatory diagram showing an operation at the time of server distribution according to the distribution policy in the form of FIG. 7. FIG. 7 is an explanatory diagram showing an operation at the time of server distribution by the distribution policy in the embodiment of FIG. FIG. 9 is an explanatory diagram showing an operation at the time of server distribution according to a policy. FIG. 9 is an explanatory diagram showing an operation at the time of a distribution change due to a server failure in the embodiment of FIG. 4. FIG. FIG. 11 is an explanatory diagram showing an operation at the time of change. FIG. 11 is an explanatory diagram showing an operation at the time of distribution change due to a server failure in the embodiment of FIG. 4. FIG. 12 is another example of an embodiment of the load distribution method of the present invention. FIG. 13 is an explanatory diagram showing an operation at the time of distribution change due to a tag converter failure in the embodiment of FIG. 12. FIG. 14 is an operation at the time of distribution change due to a tag converter failure in the embodiment of FIG. FIG. 15 is an explanatory diagram showing an operation at the time of distribution change due to a tag conversion device failure in the embodiment of FIG. 12. FIG. 16 is a diagram showing still another example of the embodiment of the load distribution method of the present invention. FIG. 17 is a block diagram showing details of the tag conversion device in FIG. 16.
1-5: Network, 11-30: Information terminal, 31, 32: DNS server, 41-66: Server, 71-74: Spare server, 81-84: Load balancer, 91, 92: Switch, 93, 94 : HUB, 101 to 118: VLAN, 121 to 124: tag conversion device, 131, 132: load distribution control device, 141 to 148: information communication network (group), 211, 241: packet input unit, 212, 242, 312 : Tag conversion unit, 213 and 243 are packet output units, 214, 244 and 314: Table setting unit, 215, 245 and 315: Tag conversion table, 216 and 316: Communication unit, 246 and 311: Load state detection unit, 313 : Packet input / output unit, 247, 317: Distribution setting input unit.

Claims (8)

In a load distribution method in a network system including a plurality of information terminals and a plurality of servers,
Connect multiple servers to VLANs with different VIDs,
Detect the load status and failure of each server using the load balancing control device,
Using a tag conversion device that converts the VID of the input packet into the VID of the connection destination VLAN, the VID of the packet from the information terminal is converted according to the load status and failure of each server, and the load distribution policy The load distribution method characterized by switching a server connected to the information terminal. The load balancing method according to claim 1,
Load balancing method characterized by giving the same IP address to a plurality of servers including the spare server. The load balancing method according to claim 1,
A load balancing method characterized in that the tag converter is made redundant. The load balancing method according to claim 1,
A load distribution control apparatus detects a load state of each server by outputting a VID of a packet for detecting a load state of the server while switching to a VID corresponding to a VLAN to which each server is connected. . A tag conversion device used in the load balancing method according to claim 1,
A tag conversion table in which the VID corresponding to the VLAN to which the input packet is connected is described for each VID, IP address, and port number of the input packet;
Means for converting and outputting the VID of the input packet according to the tag conversion table;
A tag conversion apparatus comprising: means for setting a tag conversion table according to an instruction from the outside. A load distribution control device used in the load distribution method according to claim 1,
A tag conversion table describing a VID corresponding to a VLAN to which each server is connected;
A means for switching the VID to a packet for detecting the load state of the server according to the tag conversion table and outputting it, and detecting the load state of each server according to the response;
A load balancing control device comprising: means for notifying the detected result to the outside. In a load distribution method in a network system including a plurality of information terminals and a plurality of servers,
Connect multiple servers to VLANs with different VIDs,
Using a tag conversion device that converts the VID of the input packet into the VID of the connection destination VLAN, the packet for detecting the load state of the server is output while switching to the VID corresponding to the VLAN to which each server is connected. Switching the server connected to the information terminal by detecting the load state and failure of each server and converting the VID of the packet from the information terminal according to the load state, failure and load distribution policy of each server. A load balancing method characterized by the above. A tag conversion device used in the load balancing method according to claim 7,
A tag conversion table describing the VID corresponding to the VLAN to which the input packet is connected for each VID, IP address, and port number of the input packet and describing the VID corresponding to the VLAN connected to each server;
Means for converting and outputting the VID of the input packet according to the tag conversion table;
A means for switching the VID to a packet for detecting the load state of the server according to the tag conversion table and outputting it, and detecting the load state of each server according to the response;
Means for setting a tag conversion table according to a load state of each server and an instruction from the outside.

Images