CN100435530C - A Realization Method of Two-way Load Balancing Mechanism in Multi-machine Server System - Google Patents

Abstract

The present invention relates to a method for realizing a two-way load equalizing mechanism in a multi-machine server system. A load equalizing system composed of one or more load equalizer nodes is connected with an outer network, request and return data packets pass through the load equalizing system, server nodes in the system are externally shielded, and a server system has high safety. Simultaneously, the load equalizer nodes are used for equalizing the load of the request from a client terminal, and the objective MAC addresses of the data packets are modified to distribute the data packets to improve performance when the load of the request is equalized; when returned data packets are uniformly distributed to the load equalizer nodes when pass through the load equalizing system, and the whole server system has the function of bidirectional load equalization; when the load equalizer nodes have faults, the request data packets and the return data packets can be transferred to other load equalizer nodes to realize high usability.

Description

A Realization Method of Two-way Load Balancing Mechanism in Multi-machine Server System

technical field

The invention relates to the field of computer technology, and specifically provides a bidirectional load balancing mechanism used in a multi-machine server system.

technical background

Due to the rapid development of the Internet, the rapid growth of various applications has greatly increased the number of visits to network servers, which has led to the emergence of multi-machine server systems (such as cluster systems, also known as cluster systems) to meet the growing variety of applications. kind of demand. Load balancing technology is the key technology in multi-machine server system. Its main function is to balance the load so that the entire multi-machine server system can achieve the best performance. At present, the load balancing system based on multiple load balancers is one of the most used solutions, but the current load balancing generally refers to the load on the client request data (also called uplink data) packets. balanced. For the response data (also called downlink data) package returned from the server node, some systems directly return it to the client through the server node. At this time, there is no need to consider load balancing. In this case, the back-end server system ( Generally composed of server nodes connected through the network) exposed to the external network, the security of the system is not good; some systems use the Network Address Translation (NAT, Network Address Translation) mechanism (strategy), although the request data packet and the return data packet All go through the load balancing system to ensure security, but there are also two shortcomings. One is that the network address translation makes the load balancing system more expensive, which affects the performance of the load balancing system; the other is that the returned data packets are not load balanced.

Contents of the invention

The purpose of the present invention is to overcome the above-mentioned deficiencies in the prior art, and provide a method for realizing a bidirectional load balancing mechanism in a multi-machine server system, which has a high availability function and can improve the performance and security of the server system.

Technical scheme of the present invention is realized like this: carry out as follows:

1) In a multi-machine server system composed of many computers, use one or more load balancer nodes to form a load balancing system, each balancer node has two Ethernet ports, one is connected to the external network, Responsible for receiving the client's request packet; the other is connected to the internal network and is responsible for communicating with the back-end server system;

2) When a client request data packet arrives, the load balancer node in the load balancing system selects the sending destination of the request data packet according to the load and survival status of the backend server node, and assigns the destination MAC address of the data packet to Modify the MAC address of the selected back-end server node, and then distribute the data packet to the back-end server node;

3) At least one network card of the load balancer node and the back-end server node is in the same network segment, so as to ensure that the data packet can reach the destination back-end server node by modifying the destination MAC address of the data packet;

4) Write the internal IP addresses of all load balancer nodes and the IP address information of the backend server nodes in the configuration file, and save them in the console of the load balancing system;

5) On the console of the load balancing system, the administrator numbers the load balancer nodes and the backend server nodes with consecutive integer numbers starting from 0, and uses these numbers to take the modulus of the total number of load balancer nodes. The number of the end server node is j, and j takes the modulus of the load balancer node, and the modulus result is i, then the internal IP address of the i-th load balancer node is used as the default gateway address of the back-end server node, In this way, the internal IP address of the load balancer node is balanced as the default gateway address of each back-end server node;

6) After the backend server node processes the request packet from the client, it forwards the returned data to its corresponding default gateway, which is the corresponding load balancer node;

7) When the load balancer node is implemented based on the host of the Linux operating system, the load balancer node will modify the Linux operating system kernel to allow the Linux operating system kernel to receive packets whose source IP is the same as its own IP and comes from the outside, and Open its own forwarding function in the Linux kernel, and directly forward the return data packet sent by the back-end server node to the external network;

8) When a load balancer node is added or deleted, the resident program on the console of the load balancer system will automatically modify the saved internal IP address information table of the load balancer node, and then re-take the model to check the backend The server nodes and load balancer nodes are re-divided, and the internal IP of the normal working load balancer node is re-distributed to each back-end server node as their default gateway address to ensure the high availability and load balancing of the balancer system ;

9) When a back-end server node is added or deleted, the administrator needs to re-number the back-end server nodes with consecutive integer numbers starting from 0 on the console of the load balancing system, and use the new number to count the total number of load balancer nodes. Take the modulus and reconfigure the default gateway of the backend server node.

By adopting the above method, the present invention has the following technical effects:

1. Two-way load balancing

This system can not only perform load balancing on the request data packets, but also perform load balancing on the returned data packets when the returned data packets are returned through the load balancing system, so as to better improve the load balancing effect of the server system.

2. High performance

When performing forward load balancing, the data packets are distributed to the back-end server nodes by modifying the destination MAC address of the data packets; when performing reverse load balancing, it is by setting the gateway address of the server nodes, the essence of which is also This is achieved by modifying the destination MAC address of the data packet. Compared with the NAT mechanism, there is no need to perform network address translation on the data packets, and the overhead of the load balancing system is relatively small, thereby improving the performance of the server system.

3. High availability

When a load balancer node fails, the task can be migrated by dynamically modifying the gateway address of the backend server node to the internal IP address of the normal working load balancer node, thereby achieving high availability.

4. Security

Both request and return data packets pass through the load balancing system, and the internal server nodes of the entire multi-machine server system are shielded from the external network. Compared with the DR (Direct Routing) mechanism (strategy), it can more effectively ensure the security of the entire multi-machine server system sex.

5. Scalability

The entire server system can dynamically add or delete the number of load balancer nodes as needed to achieve the best cost performance.

Description of drawings

FIG. 1 is a schematic diagram of the working principle of the present invention when performing forward load balancing.

FIG. 2 is a schematic diagram of the working principle of the present invention when performing reverse load balancing.

FIG. 3 is a conversion diagram of addresses and port numbers in data packets when the present invention performs load balancing.

FIG. 4 is a topology diagram of gateway division for all server nodes.

FIG. 5 is a topology diagram for dividing service pools into gateways.

Accompanying drawing is the specific implementation example of the present invention.

The content of the present invention will be described in further detail below in conjunction with the accompanying drawings.

Detailed ways

Referring to Figure 1, the load balancing system is composed of multiple balancer nodes, and each balancer node has two Ethernet ports, one is connected to the external network and is responsible for receiving client request packets; one is connected to the internal network, Responsible for communicating with the backend server system. The dotted oval box above the dotted line in the figure is where forward load balancing is performed. It can be seen from Fig. 1 that when performing forward load balancing, the balancer node i (i∈0~n) realizes the distribution of data packets by modifying the destination MAC address of the data packets. When the request packet from the client passes through the load balancing system, the load balancer node i (i∈0~n) decides which server node j (j∈ 0~m) to process, and then modify its destination MAC address to the MAC address of the selected server node j (j∈0~m) and forward it.

Referring to Figure 2, the dotted ellipse box below the dotted line in the figure is the place where reverse load balancing is performed. When performing reverse load balancing, it is realized by setting the gateway address of the server node j (j∈0~m) as the internal IP address of the balancer node i (i∈0~n). When the return data packet is output from the server node j (j∈0~m), it is forwarded to the equalizer node i (i∈0~n) corresponding to the gateway address of the server node j (j∈0~m), and then It is forwarded to the client by the equalizer node i (i∈0~n).

Referring to Figure 3, Cip refers to the client IP, and Vip refers to the IP provided by the load balancing system to the external network, which is generally called a single IP or virtual IP; Cport refers to the (network) port number of the client, Vport refers to the destination port number; Vmac refers to the virtual MAC address provided by the load balancing system, Rmac refers to the MAC address of the selected backend server node, and Gmac refers to the MAC address of the gateway of the backend server node The address is the MAC address of the internal network card of the corresponding balancer node. It can be seen from Figure 3 that the balancer node in the load balancing system receives the request data packet from the client, and when performing forward load balancing, the destination MAC address of the data packet is changed to the MAC address of the selected server node ; The return data packet of the server node, when reverse load balancing is performed, the destination MAC address of the data packet is modified to the gateway address of the server, which is the MAC address of the corresponding balancer node.

Referring to Figure 4, the server nodes are put together to uniformly model the number of equalizer nodes, so as to divide the server node gateways in a balanced manner. In the figure, it is assumed that there are n equalizer nodes, numbered 0-n-1; there are (k+1)n server nodes, and the results after division by modulus are shown in the figure: the numbers are 0, n...kn The server nodes are assigned to the equalizer node numbered 0; the server nodes numbered 1, n+1...kn+1 are assigned to the equalizer node numbered 1; the numbered n-1, 2n-1... ...kn+n-1 server nodes are assigned to equalizer nodes numbered n-1. The purpose of the division is to make each equalizer undertake relatively equal tasks.

Referring to Fig. 5, the backend server system is composed of many service pools providing different services. At this time, the service pools need to be divided separately, that is, each service pool is divided according to the method shown in Figure 4. In order to avoid the load imbalance of the balancer when there are few server nodes in the service pool, the service pool and service Two-level division strategy for server nodes in the pool.

The two-way load balancing mechanism provided in the present invention includes forward load balancing and reverse load balancing. Requests are evenly distributed to each server node according to a preset strategy; reverse load balancing refers to the balanced distribution of data packets returned from server nodes to each load balancer node, and then through the load balancer The node is returned to the client. This method is used when request and return data packets are required to pass through the load balancing system.

In the load balancing system, when a load balancer node fails, it will have an impact on the entire multi-machine system. The gateway address of the backend server node provided by the invention is redirected to any other normal working load balancing system. The server node method can solve this problem very well. While ensuring high availability, it also distributes the load of the failed node to other load balancing nodes for processing, avoiding the traditional task of completely taking over the failed node by the backup node. The situation that the load is all added to the backup node.

The data packet transmission method provided in the present invention is: when performing forward load balancing, the load balancer node modifies the destination MAC address of the data packet; when performing reverse load balancing, the server node is used to set the gateway . The system overhead of modifying the destination MAC address of the data packet is very small; the load balancer node, as the gateway of the back-end server node, just makes a simple judgment and forwards the data packet when the data packet passes, and the system overhead is also very small. In this way, the entire multi-machine server system can have high access performance.

The two-way load balancing mechanism of the present invention supports scalability. Scalability refers to reducing costs by adding resources to meet growing demands for performance and functionality, or by shrinking resources. The total service capacity of the system should increase proportionally with the increase of resources. Ideally, the rate of growth is linear. The cost increase should be less than N (N refers to the number of repeated resources) or a linear coefficient of NlogN. In the load balancing mechanism of the present invention, the number of load balancer nodes can be dynamically increased and deleted as required, and the performance of the system has an approximately linear relationship with the number of load balancer nodes.

The present invention is realized in the following way: the request data packet from the client, when performing forward load balancing, the load balancer node directly forwards it to the back-end server node by modifying the destination MAC address of the data, without going through Network address translation; in order to enable the data packets returned to the client to be forwarded through the load balancer node, the information list of the load balancer node and the back-end server node is saved on the console of the multi-machine system, which is set according to the administrator The load balancing algorithm, the division program resident in the console divides the server nodes into gateways, so that the server nodes correspond to the load balancer nodes as balanced as possible, and then set the default gateway of the back-end server nodes according to the calculated correspondence. In this way, the data packet returned from the backend server node is forwarded to its default gateway, which is the corresponding load balancer node. At the same time, the console management monitoring program simultaneously monitors the status of each load balancer node and back-end server node, so as to dynamically modify the corresponding relationship between the server node and the load balancer node to ensure its high availability.

Since the forward load balancing is achieved by directly modifying the destination MAC address of the data packet, and the load balancer node is used as the default gateway of the backend server node during the reverse load balancing, so the load balancer is required At least one port of the node and the server node is in the same network segment. Only in this way can we ensure that the data packets accessing the server system can reach the server node through the load balancer node, and ensure that the returned data packets can be forwarded through the load balancer node.

When performing forward load balancing, in order to distribute the access data packets to the back-end server nodes in a balanced manner, various load balancing algorithms can be used, such as round-robin method, minimum number of connections method, etc.

When performing reverse load balancing, in order to distribute the returned data packets to each load balancer node in a balanced manner, it is necessary to divide the corresponding relationship between the backend server node and the load balancer node. The specific division method is as follows:

Suppose there are n load balancer nodes and m backend server nodes (m>n), the load balancer nodes are numbered 0, 1...n-1, and the backend server nodes are numbered 0, 1... m-1, the distribution algorithm can be realized by the following method: Take the number of the back-end server node modulo n, if the modulus value is i (0=<i<=n-1), then load the i-th The internal IP address of the balancer node is set as the default gateway address of the server node.

When a load balancer node fails, in order to ensure high availability, it is necessary to reset the gateway of the back-end server node. After resetting, it is still necessary to ensure that load balancing works. The redistricting method is as follows:

When the load balancer node numbered i fails, the load balancer node needs to be removed from the system. Now the total number of load balancer nodes is n-1, and the remaining load balancer nodes are restarted from 0. number, and then re-modulo n-1 the number of the backend server node to reconfigure the default gateway address of the backend server node. It is also possible to improve efficiency by re-corresponding the server node corresponding to the load balancer node numbered i to other load balancer nodes, and the other server nodes will not be changed again to improve efficiency, that is, when the load balancer node numbered i When the load balancer node fails, the backend server node corresponding to the load balancer node is reassigned to other load balancer nodes.

In order to achieve scalability, when it is necessary to add load balancer nodes, it is also necessary to re-partition, the algorithm is as follows:

When there are n load balancer nodes and a new one needs to be added, number it n, the total number of load balancers is now n+1, and then re-modulate the number of the backend server node to n+1, Use this to reconfigure the default gateway address of the backend server node. It is also possible to improve efficiency by removing some of the server nodes corresponding to the original n load balancer nodes and re-corresponding to the newly added load balancer node, while other server nodes will not be changed again. That is, when a new load balancer node is added, it is numbered n, and then the back-end server node is taken out (m/n+1) (rounded down) and assigned to the newly added load balancer node , that is, in the original n load balancer nodes, take (m/n*(n+1)) (rounded down) from each corresponding server node, and make it correspond to the newly added On the load balancer node, re-modify its default gateway address.

If the load balancer node is implemented based on the host of the Linux operating system, when processing the returned data packet, it is necessary to open its own data forwarding function in the kernel of the Linux operating system to ensure that the data packet whose destination IP is not its own IP is forwarded.

Claims (1)

1, the implementation method of two-way load equalizing mechanism in a kind of multiple machine servicer system is characterized in that, carries out as follows:

1) in the multiple machine servicer system that is constituted by a lot of platform computers, form SiteServer LBS with one or more load equalizer node, every equalizer node all has two ethernet ports, and one links to each other with external network, is responsible for receiving the request data package of client; Another links to each other with internal network, is responsible for and back-end server system communication;

2) when the client-requested packet arrives, the transmission destination that load equalizer node in the SiteServer LBS is selected request data package according to the load and the survival condition of back-end server node, the target MAC (Media Access Control) address of packet is revised as the MAC Address of selected back-end server node, gives this back-end server node with this packet delivery then;

3) to have a network interface card at least be in the same network segment for load equalizer node and back-end server node, to guarantee just can to make packet arrival purpose back-end server node by the target MAC (Media Access Control) address of revising packet;

4) the internal IP address of all load equalizer nodes and back-end server IP addresses of nodes information are write in the configuration file, be kept in the control desk of SiteServer LBS;

5) on the control desk of SiteServer LBS, the keeper carries out numbering since 0 continuous integral number to load equalizer node and back-end server node respectively, with these numberings load equalizer node sum is carried out delivery, if certain back-end server node be numbered j, j carries out delivery to the load equalizer node, the delivery result is i, then with the internal IP address of i platform load equalizer node default gateway address, by this method with the default gateway address as each back-end server node of the internal IP address equilibrium of load equalizer node as this back-end server node;

6) after the request data package that the intact client of back-end server node processing brings in, return data is transmitted to own corresponding default gateway, just Dui Ying load equalizer node;

7) when the load equalizer node is based on the main frame realization of (SuSE) Linux OS, the load equalizer node will be revised the (SuSE) Linux OS kernel, allow (SuSE) Linux OS kernel reception sources IP identical with self IP and be from the outside packet, and in linux kernel, open the forwarding capability of self, the return data bag that the back-end server node is sent directly is forwarded to outer net;

8) when the load equalizer node increases or deletes, the program of garrisoning on the control desk of SiteServer LBS can be revised the internal IP address information table of the load equalizer node of preserving automatically, carry out delivery again then, back-end server node and load equalizer node are repartitioned, the internal IP of the load equalizer node of operate as normal is given the default gateway address of each back-end server node as them again, to guarantee the high availability and the load balancing of equalizer system;

9) when the back-end server node increases or deletes, the keeper will carry out since 0 continuous integer numbering the back-end server node on the control desk of SiteServer LBS again, with new numbering load equalizer node sum is carried out delivery, reconfigure the default gateway of back-end server node.

Images