CN107426332A - The load-balancing method and system of a kind of web server cluster - Google Patents

Abstract

The invention discloses a load balancing method and system of a WEB server cluster. Firstly, for each type of static page access, the server to be allocated is respectively designated, and at the same time, a load balancing target function is constructed. When a page method request is received, it is classified, When it is a static page access request, it is allocated to the corresponding server according to its type; when it is a dynamic page access request, it is obtained according to the optimal allocation weight of each server's dynamic page access in the current cycle obtained in the previous cycle Allocate the dynamic page access at each moment of the current cycle; when the time reaches the last moment of the current cycle, obtain the optimal allocation weight of each server's dynamic page access in the next cycle by balancing the objective function; the present invention is aimed at static pages and dynamic pages Different processing methods are proposed respectively, so that server resources can be fully utilized, and load balancing effects can still be achieved under complex access situations.

Description

A load balancing method and system for a WEB server cluster

technical field

The invention relates to the field of computer server clusters, in particular to a load balancing method and system for a WEB server cluster.

Background technique

With the rapid development of the Internet, the scale of various networked information systems is getting larger and larger, and the access traffic is increasing exponentially, which puts forward higher requirements for the working performance of the server. When a single web server handles a large number of service requests, there will be problems such as excessive workload, long service request response time, and large feedback delay. In severe cases, there will be a single point of failure, and the user experience will be poor. Therefore, server cluster technology is used to improve system response efficiency and reliability, and to provide users with concurrent services. The structure and performance of individual servers in the server cluster are different, and client access is random and sudden, so the existing system generally configures a load balancer. According to the maximum processing capacity of a single server and the current load status, the load balancer distributes load tasks according to the load scheduling strategy, reasonably utilizes the resources of each server to avoid overloading, and improves the stability and response speed of the server cluster.

For the load balancing research of Web server clusters, there have been many achievements in the aspects of model, method and application realization. There are two types of load balancing methods: static and dynamic. Static load balancing methods include Random Random (RAN) method, Round-Robin Round Robin (RR) method and Static Weight Round Robin (SWRR) method. The random method is to randomly select a server from the server cluster to respond to the service request; the polling method is to send a new access request to each server in a poll; the static weighted polling method is to give each Each server is assigned different weights so that it can accept service requests with corresponding weights. The static load balancing method saves the task distribution ratio of each server as a system parameter in the load balancer, regardless of the real-time changes in server load. The static load balancing method is an open-loop load adjustment method, which is difficult to meet the performance requirements of the new generation of server clusters. Dynamic load balancing methods include Least Connection (LC) method, Weighted Least Connection (WLC) method and Dynamic Weighted Round Robin (DWRR) method. The load balancer periodically collects the load information of each server, and calculates the task distribution weight distribution tasks according to the real-time load level of the server. In the case of different types of request tasks and large differences in workload, the performance of dynamic load balancing is better than that of static loads. Balancing method, but the dynamic load balancing method requires additional data processing and transmission. The traditional dynamic load balancing method has limitations. For example, the least connection method uses the number of connections to measure the current load of the server. Because the performance of the server is different and the server resources consumed by the dynamic page access task are different, the same number of connections does not mean the load on the server. The load is the same, so the traditional dynamic load balancing method cannot accurately measure the current load of the server.

Contents of the invention

The first purpose of the present invention is to overcome the shortcoming and deficiency of prior art, provide a kind of load balancing method of WEB server cluster, this method can realize the load balancing of server cluster under mixed request service, compared with traditional load balancing method, This method can make full use of server resources, and can still achieve the effect of load balancing under complex access situations.

The second object of the present invention is to provide a kind of load balancing system for realizing the WEB server cluster of above-mentioned load balancing method.

The first purpose of the present invention is achieved through the following technical solutions: a load balancing method of a WEB server cluster, comprising the steps:

S1. First, specify the server to be allocated for each type of static page access; at the same time, the load rate of each server at the last moment of the next cycle corresponds to the target construction that is closest to the average predicted load rate of the server at the last moment of the next cycle Load balancing objective function, through the objective function to solve the optimal allocation weight of each server's dynamic page access in the next period; and construct the constraints of the load balancing objective function, including: the load rate of each server at the last moment of the next period is less than a certain value And the sum of the dynamic page access allocation weights of each server in the next cycle is 1;

For the page access request received from the client at each moment of the current cycle, the following processing is performed:

S2, when receiving the page access request proposed by the client at each moment of the current cycle, first classify the page requested to be accessed, if it is a dynamic page, then proceed to step S4; if it is a static page, then proceed to step S3 At the same time, when the time reaches the last moment of the current cycle, collect the load rate of each server at the last moment of the current cycle, and calculate the optimal distribution weight of each server's dynamic page access in the next cycle. The specific process is as follows: first, according to The load rate of each server at the last moment of the current cycle and the predicted load rate of each server at the last moment of the current cycle predicted at the last moment of the previous cycle predict the predicted load rate of each server at the last moment of the next cycle, and then Calculate the average value to obtain the average predicted load rate of the server at the last moment of the next cycle, and finally input the average predicted load rate of the server at the last moment of the next cycle into the load balancing objective function constructed in step S1 to obtain the optimal solution, thereby obtaining the next cycle The optimal allocation weight of each server's dynamic page access; when the next cycle comes, return to this step;

S3. For the static page access requests at each moment of the current cycle, first obtain the type of the static page that is requested to be accessed, and then search for the server to be assigned for this type of static page access, and determine whether the server can currently access the page Provide services, if so, assign the static page access to the server; if not, assign the static page access to the server with the smallest load rate at the last moment of the last cycle in the WEB server cluster, and assign this type of static page access Specify that the server to be assigned is updated to this server;

S4. For the dynamic page access requests at each moment of the current cycle, the dynamic page access at each moment of the current cycle is allocated according to the optimal distribution weight of the dynamic page access of each server in the current cycle obtained at the last moment of the previous cycle.

Preferably, in step S1, the server to be assigned is specified for each type of static page access respectively, and is recorded in the hash table; in the step S3, for the static page access request at each moment of the current cycle, the request access information is obtained. After specifying the type of the static page, look up the static page of this type through the hash table to access the specified server to be allocated.

Preferably, in step S2, the formula for calculating the load rate of each server at each moment in the current period is:

Where R _i (t) is the load rate of the i-th server at time t of the current period, is the static page load to be processed by the i-th server at time t of the current period, is the dynamic page load to be processed by the i-th server at time t in the current cycle, and C _i is the maximum processing capacity of the i-th server; when t is (k-1)T, the current cycle is calculated by the above formula (k-1) The load rate of the i-th server at the last moment (k-1)T of the cycle, k=2,3,...;

The formula for calculating the static page load to be processed by each server at each moment in the current cycle is:

Among them, c _i,1 (t), c _i,2 (t), ... c _{i, j} (t) respectively correspond to the first, second, ..., j-type static files held by the i-th server at the moment of the current cycle t number of visits; is the load generated by accessing the jth type of static page, S _j is the size of the jth type of static page, λ is the resource utilization coefficient for reading the static page file and processing and sending, T is the cycle length; when t is taken as (k-1 )T, the static page load to be processed by the i-th server at the last moment (k-1)T of the current period (k-1)th period is calculated by the above formula

The formula for calculating the dynamic page load to be processed by each server at each moment in the current cycle is:

in is the uncompleted dynamic page load of the i-th server before time t in the current cycle, M _i is the performance constant of the dynamic page generated by the i-th server, w _i (k-1) is the current cycle obtained in the previous cycle, i.e. the (k-1) The optimal allocation weight of the i-th server's dynamic page access in the cycle; is the number of dynamic page visits that need to be buffered and forwarded to the i-th server at time t of the current period, Corresponding to time t of the current cycle The load generated by the number of dynamic page visits on the i server; when t is (k-1)T, the current cycle is calculated by the above formula, that is, the last moment (k-1) of the (k-1)th cycle The amount of dynamic page load to be processed by the i-th server at time T

The load rate of each server at the last moment of the current cycle:

Wherein R _i ((k-1)T) is the load rate of the i-th server at the last moment (k-1)T of the current cycle, that is, the (k-1)th cycle.

Furthermore, the value of the resource utilization coefficient λ for reading static page files and processing and sending is as follows:

When the server reads the static page file from the disk and processes it, take λ=λ ₁ , where λ ₁ is the resource utilization coefficient for the server to read the static page file from the disk and process it,

When the server reads the static page file and processes it from the cache, get λ=λ ₂ ; λ ₂ is the resource utilization coefficient that the server reads the static page file from the cache and processes and sends it;

where λ ₁ >λ ₂ .

Furthermore, the maximum processing capacity C _i of the i-th server is:

Among them, k ₁ ~ k ₄ are the influence coefficients of each index on the maximum processing capacity of the server; μ _i is the CPU quantity of the i-th server, f _i is the CPU frequency of the i-th server, G _i is the memory capacity of the i-th server, and D _i is The I/O rate of the i-th server, N _i is the network throughput of the i-th server.

Preferably, in step S2, according to the load rate of each server at the last moment of the current cycle and the predicted load rate of each server at the last moment of the current cycle predicted at the last moment of the previous cycle, the following is predicted by the exponential smoothing prediction method The predicted load rate of each server at the last moment of a cycle:

Where R _i ((k-1)T) is the load rate of the i-th server at the last moment (k-1)T of the current period (k-1)th period, The last moment (k-1)T of the current period (k-1)th period predicted by the last period of the current period (k-1)th period, i.e. the last moment of the k-2th period The predicted load rate of each server at time, is the predicted load rate of the i-th server at time kT at the last moment of the k-th cycle in the next cycle; α is the smoothing coefficient;

According to the predicted load rate of each server at the last moment of the next cycle, the average predicted load rate of the server at the last moment of the next cycle is obtained:

in It is the average predicted load rate of the server at kT time at the last moment of the next period, that is, the kth period; n is the total number of servers in the WEB server cluster.

Preferably, the load balancing objective function constructed in step S1 is:

The constraints of the load balancing objective function constructed in step S1 are:

R _i (kT)≤0.9, i=1,2,3,...n;

Among them, R _i (kT) is the load rate of each server at kT at the last moment of the k-th cycle in the next cycle, is the average predicted load rate of the server at the last moment of the next cycle, that is, the kth cycle, _n is the total number of WEB server cluster servers; Optimal distribution of weights, where i=1,2,...n; T is the cycle length.

Furthermore, in step S2, the optimal solution in the load balancing objective function, that is, the optimal distribution weight is obtained through Cauchy's inequality and exhaustive method.

Preferably, in step S3, when server receives static page visit, at first whether cache has this static page, if not, then read from disk;

In step S2, for the dynamic page access requests at each moment of the first cycle, the optimal allocation weights of each server dynamic page access in this cycle are set to be the same.

The second object of the present invention is achieved through the following technical solutions: a load balancing system for a WEB server cluster for implementing the above load balancing method, including a client and a WEB server cluster, and the WEB server cluster includes a plurality of servers; It is characterized in that it also includes a load balancer, and the load balancer is set between the client and the WEB server cluster;

The client is configured to send a page access request to the load balancer through the Internet;

The load balancer is used to specify the server to be allocated for each type of static page access; it is used to construct a load balancing objective function and constraints; The load rate corresponds to the target that is closest to the average predicted load rate of the servers at the last moment of the next cycle. The constraints include that the load rate of each server is less than a certain value at the last moment of the next cycle and the dynamic page access allocation of each server in the next cycle The sum of the weights is 1;

It is used to receive the page access request from the client at each moment of the current cycle. After receiving the page access request, first classify the page that is requested to be accessed; if it is a static page, first obtain the type of the static page that is requested to be accessed, and then search This type of static page access specifies the server to be allocated, and judges whether the server can currently provide services for page access. If so, assign the static page access to the server; if not, assign the static page access to the WEB In the server cluster, the server with the smallest load rate at the last moment of the previous cycle, and the server to be allocated for this type of static page access is updated to this server; if it is a dynamic page, according to the obtained at the last moment of the previous cycle The optimal allocation weight of each server's dynamic page access in the current cycle is used to allocate the dynamic page access at each moment in the current cycle;

Used to collect the load rate of each server at the last moment of the current cycle in the WEB server cluster when the time reaches the last moment of the current cycle;

When the time reaches the last moment of the current period, it is used to calculate the optimal distribution weight of each server's dynamic page access in the next period. The specific process is as follows: first, according to the load rate of each server at the last moment of the current period and the last period of the previous period The predicted load rate of each server at the last moment of the current cycle is predicted at one moment, and the predicted load rate of each server at the last moment of the next cycle is predicted, and then the average value is obtained to obtain the average predicted load rate of the server at the last moment of the next cycle. Finally, the average predicted load rate of the server at the last moment of the next cycle is input into the load balancing objective function to find the optimal solution, so as to obtain the optimal allocation weight of the dynamic page access of each server in the next cycle;

Each server in the WEB server cluster is used to calculate the load rate therein when the time reaches the last moment of the current cycle, and provide the load balancer for periodic collection; for receiving the static pages assigned to it by the load balancer Access and dynamic page access; used to respond to the client with the corresponding service based on the static page access and dynamic page access assigned to it.

Compared with the prior art, the present invention has the following advantages and effects:

(1) In the present invention, at first for each type of static page access, specify the server to be allocated respectively, and at the same time, the load rate of each server at the last moment of the next cycle corresponds closest to the average predicted load of the server at the last moment of the next cycle Construct a load balancing objective function with the target rate, and set constraints; when receiving a page method request, first classify it, and when it is a static page access request, obtain the type of static page that is requested to be accessed, and then allocate the static page access The designated server to be allocated for this type of static page access that can still provide access services, if the server can no longer provide services, the static page access will be allocated to the server with the smallest load rate at the last moment of the previous period, and the Each type of static page access specifies that the server to be allocated is updated to this server; when it is a dynamic page access request, the optimal allocation weight of each server's dynamic page access in the current cycle obtained in the previous cycle has a significant impact on the current cycle at each moment. Dynamic page access is allocated; when the time reaches the last moment of the current cycle, the average predicted load rate of the server at the last moment of the next cycle is input into the load balancing objective function to obtain the optimal solution, so as to obtain the dynamic status of each server in the next cycle The optimal allocation weight of page access; As can be seen from the above, the present invention proposes different processing methods for static pages and dynamic pages, so that server resources can be fully utilized, and the effect of load balancing can still be achieved under complex access situations.

(2) In the present invention, at first for each type of static page access, specify its server to be allocated respectively, when receiving the static page access request, obtain the type of the static page that requests to visit, then assign the static page access to another This type of static page that can provide access services is accessed to the designated server to be allocated. If the server can no longer provide services, the static page access will be allocated to the server with the smallest load rate at the last moment of the previous cycle, and this type Static page access specifies that the server to be assigned is updated to the server. The present invention makes full use of the feature that the server can cache static pages, and first specifies the server to be allocated for each type of static page access, and performs real-time update; when receiving a static page access request, it first finds out the type Static page access specifies the server to be assigned to it, and if the server is still able to provide services, assign the requested static page access to the server, and the server is assigned to the server for this type of static page access , so it is likely to have received the visit of the static page, so it is likely to cache the static page provided when the static page was visited last time, which can greatly improve the response speed and improve the user experience.

(3) In the present invention, each type of static page access specifies its server to be allocated respectively, and is recorded in the hash table; After the type, look up the static page of this type through the hash table to access the specified server to be allocated. By above-mentioned restraint, the method of the present invention can search through the hash table the server that certain type of static page access designation will distribute, has improved search rate.

(4) In the present invention, the load rate of each server at the last moment of the current cycle is obtained by the sum of the dynamic page load and the static page load, so the current load of the server can be measured more accurately, greatly improving the effect of load balancing .

(5) In the present invention, it is only necessary to collect the load rate of each server at the last moment of the current cycle and obtain the predicted load rate of each server at the last moment of the current cycle to predict the load rate of each server at the last moment of the next cycle. Predict the load rate, and then the average predicted load rate of the servers at the last moment of the next cycle can be obtained by the load balancing objective function to obtain the optimal distribution weight of each server dynamic page access in the next cycle. It can be seen that the load balancing process of the present invention does not need The load rate of each server is collected in real time, which greatly reduces the calculation amount of the load rate of each server. In addition, the present invention predicts the predicted load rate of each server at the last moment of the next cycle through the exponential smoothing prediction method; it can be seen that the present invention considers that the load parameters may fluctuate greatly in a short period of time, compared with traditional load balancing The method only judges according to the real-time server load when processing load distribution. The present invention predicts the load rate of each server at the last moment of the next cycle through the exponential smoothing prediction method, which can further improve the effect of load balancing.

(6) In the present invention, a load balancer is arranged between the client and the WEB server cluster, and the load balancer respectively collects the load rate of each server at the last moment of the cycle at the last moment of the current cycle, that is, load balancing The server periodically collects the load rate of each server in the WEB server cluster; and according to the collected load rate of each server at the last moment of the current cycle, the predicted load rate of each server at the last moment of the current cycle, and the objective function can finally be calculated The optimal distribution weight of each server's dynamic page access in the next cycle is a kind of dynamic feedback task distribution strategy, which can effectively improve the load balancing effect.

Description of drawings

FIG. 1 is a flowchart of a load balancing method for a WEB server cluster in the present invention.

Fig. 2 is a structural block diagram of the load balancing system of the WEB server cluster of the present invention.

detailed description

The present invention will be described in further detail below in conjunction with the examples and accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example

This embodiment discloses a load balancing method for a WEB server cluster, and the specific process is as follows:

S1. First, specify the server to be allocated for each type of static page access, and record it in the hash table of the load balancer; among them, static pages are classified according to file size, and static pages of a certain size range are defined as one Type; in this embodiment, according to the size of the static page file, the static page is classified into 4 types, wherein the file size of the first type of static page is 0-50KB, and the file size of the second type of static page is 51-75KB , the file size of the third type of static page is 76-100KB; the file size of the fourth type of static page is larger than 100KB.

At the same time, the load rate of each server at the last moment of the next period corresponds closest to the target average predicted load rate of the server at the last moment of the next period. Build a load balancing objective function, and use the objective function to solve the dynamic page access of each server in the next period. The optimal allocation weights; and construct the constraints of the load balancing objective function, including: the load rate of each server is less than a certain value at the last moment of the next cycle and the sum of the dynamic page access allocation weights of each server in the next cycle is 1;

Wherein the next cycle in this embodiment refers to the next cycle of the current cycle;

For the page access request received from the client at each moment of the current cycle, the following processing is performed:

S2, when receiving the page access request proposed by the client at each moment of the current cycle, first classify the page requested to be accessed, if it is a dynamic page, then proceed to step S4; if it is a static page, then proceed to step S3 Processing; at the same time in this embodiment, when the time reaches the last moment of the current cycle, the load rate of each server at the last moment of the current cycle is collected, and the optimal distribution weight of each server dynamic page access in the next cycle is calculated, specifically The process is as follows: First, predict the load rate of each server at the last moment of the next cycle based on the load rate of each server at the last moment of the current cycle and the predicted load rate of each server at the last moment of the previous cycle. Predict the load rate, then calculate the average value to obtain the average predicted load rate of the server at the last moment of the next cycle, and finally input the average predicted load rate of the server at the last moment of the next cycle into the load balancing objective function constructed in step S1 to obtain the optimal solution, Thereby obtain the optimal distribution weight of each server dynamic page access in the next cycle; When the next cycle arrives, return to this step; In this embodiment, the cycle is set for the dynamic page access requests at each moment of the first cycle The optimal allocation weights of each server's dynamic page access are the same. That is, w ₁ (1)=w ₂ (1), . . . ,=w _n (1).

The load rate of each server at the last moment of the current cycle is calculated by each server itself, and this embodiment periodically collects the load rate of each server, that is, in this embodiment, each The load rate of the server. In the present embodiment, if the page file of the current request to visit ends with .html, then the page is defined as a static page, and if the page file of the current request to visit ends with .jsp, then the page is defined as a dynamic page.

S3. For the static page access requests at each moment of the current cycle, first obtain the type of the static page that is requested to be accessed, and then search for the server to be assigned for this type of static page access, and determine whether the server can currently access the page Provide services, if so, assign the static page access to the server; if not, assign the static page access to the server with the smallest load rate at the last moment of the last cycle in the WEB server cluster, and assign this type of static page access Specify that the server to be allocated is updated to this server; the server with the smallest load rate can be obtained by comparing the load rate of each server at the last moment of the previous cycle; in addition, when the server receives a static page access, whether to cache first If there is this static page, if not, it will be read from disk. In this embodiment, it is judged whether the server can also provide services for page access by detecting whether the load borne by the server solves its maximum processing capacity.

S4. For the dynamic page access requests at each moment of the current cycle, the dynamic page access at each moment of the current cycle is allocated according to the optimal distribution weight of the dynamic page access of each server in the current cycle obtained at the last moment of the previous cycle.

In the above-mentioned step S2 of the present embodiment, the calculation formula of the load rate of each server at each moment of the current cycle is:

Where R _i (t) is the load rate of the i-th server at time t of the current period, is the static page load to be processed by the i-th server at time t of the current period, is the dynamic page load to be processed by the i-th server at time t in the current cycle, and C _i is the maximum processing capacity of the i-th server; when t is (k-1)T, the current cycle is calculated by the above formula (k-1) The load rate of the i-th server at the last moment (k-1)T of the cycle, k=2,3,...;

The formula for calculating the static page load to be processed by each server at each moment in the current cycle is:

Among them, c _i,1 (t), c _i,2 (t), ... c _{i, j} (t) respectively correspond to the first, second, ..., j-type static files held by the i-th server at the moment of the current cycle t number of visits; is the load generated by accessing the jth type of static page, S _j is the size of the jth type of static page, λ is the resource utilization coefficient for reading the static page file and processing and sending, T is the cycle length; when t is taken as (k-1 )T, the static page load to be processed by the i-th server at the last moment (k-1)T of the current period (k-1)th period is calculated by the above formula

Among them, the value of the resource utilization coefficient λ for reading static page files and processing and sending is as follows:

When the server reads the static page file from the disk and processes it, take λ=λ ₁ , where λ ₁ is the resource utilization coefficient for the server to read the static page file from the disk and process it,

When the server reads the static page file and processes it from the cache, get λ=λ ₂ ; λ ₂ is the resource utilization coefficient that the server reads the static page file from the cache and processes and sends it;

where λ ₁ >λ ₂ .

The formula for calculating the dynamic page load to be processed by each server at each moment in the current cycle is:

in is the uncompleted dynamic page load of the i-th server before time t in the current cycle, M _i is the performance constant of the dynamic page generated by the i-th server, w _i (k-1) is the current cycle obtained in the previous cycle, i.e. the (k-1) The optimal allocation weight of the i-th server's dynamic page access in the cycle; is the number of dynamic page visits that need to be buffered and forwarded to the i-th server at time t of the current period, Corresponding to time t of the current cycle The number of dynamic page visits is the load generated by the i server; when t is (k-1)T, the current cycle is calculated by the above formula, that is, the last moment (k-1) of the (k-1)th cycle The amount of dynamic page load to be processed by the i-th server at time T which is:

Get the load rate of each server at the last moment of the current cycle:

Wherein R _i ((k-1)T) is the load rate of the i-th server at the last moment (k-1)T of the current cycle, that is, the (k-1)th cycle;

In this embodiment, the maximum processing capacity Ci of the _i -th server is:

Among them, the influence coefficients of the indicators k ₁ to k ₄ on the maximum processing capacity of the server, in this embodiment, the values of k ₁ to k ₄ are 0.2, 0.3, 0, and 0.5 respectively; μ _i is the number of CPUs of the i-th server, f _i The CPU frequency of the i-th server, G _i is the memory capacity of the i-th server, D _i is the I/O rate of the i-th server, N _i is the network throughput of the i-th server.

In the above step S2 of this embodiment, according to the load rate of each server at the last moment of the current cycle and the predicted load rate of each server at the last moment of the current cycle predicted at the last moment of the previous cycle, it is predicted by the exponential smoothing prediction method The predicted load rate of each server at the last moment of the next cycle:

Where R _i ((k-1)T) is the load rate of the i-th server at the last moment (k-1)T of the current period (k-1)th period, The last moment (k-1)T of the current period (k-1)th period predicted by the last period of the current period (k-1)th period, i.e. the last moment of the k-2th period The predicted load rate of each server at time, is the predicted load rate of the i-th server at time kT at the last moment of the next cycle, that is, the kth cycle; α is the smoothing coefficient, and the smaller α is, the stronger the smoothing effect is. In this implementation, α is 0.2; the current cycle is In the first period, since there is no predicted load rate of each server at the last moment of the first period, the load rate of each server at the last moment of the first period is taken as the forecast of each server at the last moment of the second period Load rate, that is, when k=2,

According to the predicted load rate of each server at the last moment of the next cycle, the average predicted load rate of the server at the last moment of the next cycle is obtained:

in It is the average predicted load rate of the server at kT time at the last moment of the next period, that is, the kth period; n is the total number of servers in the WEB server cluster.

The load balancing objective function constructed in the above step S1 of this embodiment is:

The constraints of the constructed load balancing objective function are:

R _i (kT)≤0.9, i=1,2,3,...n;

Among them, R _i (kT) is the load rate of each server at kT at the last moment of the k-th cycle in the next cycle, is the average predicted load rate of the server at the last moment of the next cycle, that is, the kth cycle, _n is the total number of WEB server cluster servers; Optimal distribution of weights, where i=1,2,...n; T is the cycle length. The load rate of each server at each period and time should be lower than 0.9. If it exceeds 0.9, it means that the load on the server is close to saturation, and the request task should not be allocated to the server.

In step S2 of this embodiment, the optimal solution in the load balancing objective function, that is, the optimal distribution weight, is obtained through Cauchy's inequality and exhaustive method. details as follows:

For the above load balancing objective function, according to Cauchy's inequality:

if and only if That is, when the load rates of all servers are equal at the last moment of the k-th cycle in the next cycle, the equal sign is established. At this time, f(w ₁ (kT),…,w _n (kT)) obtains the minimum value, and the equations can be obtained:

Using the above formula, the optimal allocation weight w _i (k) of the i-th server in the k-th cycle of the next cycle can be obtained; there is no qualified solution, and it is necessary to find a suitable w _i (k) through experiments so that the objective function can Possibly small.

This embodiment also discloses a load balancing system of a WEB server cluster for realizing the above load balancing method, as shown in Figure 2, including a client, a load balancer and a WEB server cluster, and the WEB server cluster includes multiple servers, They are server 1, server 2, ..., server n, that is, the first server to the nth server; the load balancer is set between the client and the WEB server cluster; where:

The client is used to send a page access request to the load balancer through the Internet;

The load balancer is used to specify the server to be allocated for each type of static page access; it is used to construct the load balancing objective function and constraints; the load balancing objective function is the load rate of each server at the last moment of the next cycle Corresponding to the target that is closest to the average predicted load rate of the server at the last moment of the next cycle, the constraints include the load rate of each server at the last moment of the next cycle being less than a certain value and the dynamic page access allocation weight of each server in the next cycle sum to 1;

It is used to receive the page access request from the client at each moment of the current cycle. After receiving the page access request, first classify the page that is requested to be accessed; if it is a static page, first obtain the type of the static page that is requested to be accessed, and then search This type of static page access specifies the server to be allocated, and judges whether the server can currently provide services for page access. If so, assign the static page access to the server; if not, assign the static page access to the WEB In the server cluster, the server with the smallest load rate at the last moment of the previous cycle, and the server to be allocated for this type of static page access is updated to this server; if it is a dynamic page, according to the obtained at the last moment of the previous cycle The optimal allocation weight of each server's dynamic page access in the current cycle is used to allocate the dynamic page access at each moment in the current cycle;

It is used to collect the load rate of each server at the last moment of the current cycle in the WEB server cluster when the time reaches the last moment of the current cycle; that is, the load balancer periodically collects the load rate of each server. In this embodiment, the load balancer Collect the load rate of each server in the WEB server cluster at the last moment of each cycle, and what is collected is the load rate of each server at the last moment of each cycle;

When the time reaches the last moment of the current period, it is used to calculate the optimal distribution weight of each server's dynamic page access in the next period. The specific process is as follows: first, according to the load rate of each server at the last moment of the current period and the last period of the previous period Predict the predicted load rate of each server at the last moment of the current cycle measured at the last moment of the current cycle at one moment, predict the predicted load rate of each server at the last moment of the next cycle, and then calculate the average value to obtain the last moment of the next cycle The average predicted load rate of the server, and finally input the average predicted load rate of the server at the last moment of the next cycle into the load balancing objective function to find the optimal solution, so as to obtain the optimal distribution weight of the dynamic page access of each server in the next cycle;

Each server in the WEB server cluster is used to calculate the load rate of each server when the time reaches the last moment of the current cycle, for the load balancer to collect periodically; it is used to receive the static data allocated to it by the load balancer Page access and dynamic page access; used to respond to the corresponding service to the client according to the static page access and dynamic page access assigned thereto.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (10)

1. a kind of load-balancing method of web server cluster, it is characterised in that comprise the following steps：

S1, respectively specify that its server to be distributed first for the access of each type static page；Meanwhile with next cycle most The load factor of each server of later moment in time is corresponding closest to last moment server consensus forecast load factor of next cycle Load balancing object function is built for target, solving each Java Server Page of next cycle by object function accesses most Optimal sorting matches somebody with somebody weight；And load balancing bound for objective function is built, including：Each service of last moment in next cycle It is 1 that device load factor, which is less than each Java Server Page of certain value and next cycle to access distribution weight sum,；

The accessing page request that the client of each reception of current period is proposed is directed to, is handled as follows：

S2, in the accessing page request that each reception of current period is proposed to client, first to request access page Face is classified, and if dynamic page, then carries out step S4 processing；If static page, then step S3 processing is carried out；Together When, when the time reaching last moment of current period, the load factor of each server of last moment of current period is gathered, and And the optimum allocation weight that next each Java Server Page of cycle accesses is calculated, detailed process is as follows：First according to current The current period that the load factor of each server of last moment in cycle and upper last moment in a cycle are predicted to obtain is last The prediction load factor of each server of one moment predicts the prediction load factor of next each server of last moment in cycle, then Average to obtain last moment server consensus forecast load factor of next cycle, finally service last moment in next cycle Optimal solution is asked in the load balancing object function built in device consensus forecast load factor input step S1, so as to obtain next week The optimum allocation weight that each Java Server Page of phase accesses；When next cycle arrives, this step is returned；

S3, the static page access request for being directed to current period each moment, the class for the static page that request accesses is obtained first Type, then search this type static page and access the server to be distributed specified, and judge that the server is currently It is no to provide service to page access, if so, static page then is accessed into distribution to the server；If it is not, then by static page Interview asked distribution into web server cluster in the server of a upper cycle last moment load factor minimum, and by this kind It is the server that type static page access, which specifies the server update to be distributed,；

S4, the dynamic page access request for being directed to current period each moment, according to accessed by upper last moment in a cycle The optimum allocation weight that accesses of each Java Server Page of current period the dynamic page at current period each moment is accessed It is allocated.

2. the load-balancing method of web server cluster according to claim 1, it is characterised in that be every in step S1 Type static page accesses and respectively specifies that its server to be distributed, and is recorded in Hash table；In the step S3 The static page access request at current period each moment is directed to, after the type for obtaining the static page that request accesses, passes through Kazakhstan Uncommon table search this type static page accesses the server to be distributed specified.

3. the load-balancing method of web server cluster according to claim 1, it is characterised in that step S2, current week Phase at each moment load factor calculation formula of each server is：

<mrow> <msub> <mi>R</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <msub> <mi>L</mi> <msub> <mi>S</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>L</mi> <msub> <mi>D</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> <msub> <mi>C</mi> <mi>i</mi> </msub> </mfrac> <mo>;</mo> </mrow>

Wherein R_i(t) it is the load factor of i-th server of current period t,For i-th service of current period t Device static page load capacity to be processed,For the dynamic page load to be processed of i-th server of current period t Amount, C_iFor the maximum processing capability of i-th server；Wherein when t takes (k-1) T, current period is calculated i.e. by above formula The load factor of last moment (k-1) i-th server of T moment of (k-1) individual cycle, k=2,3 ...；

Wherein each server of current period each moment static page load capacity calculation formula to be processed is：

<mrow> <msub> <mi>L</mi> <msub> <mi>S</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mi>T</mi> </mfrac> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <mrow> <msub> <mi>c</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mrow> <msub> <mi>c</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>...</mn> </mtd> <mtd> <mrow> <msub> <mi>c</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <msub> <mi>&amp;rho;</mi> <msub> <mi>S</mi> <mn>1</mn> </msub> </msub> </mtd> </mtr> <mtr> <mtd> <mo>.</mo> </mtd> </mtr> <mtr> <mtd> <mo>.</mo> </mtd> </mtr> <mtr> <mtd> <mo>.</mo> </mtd> </mtr> <mtr> <mtd> <msub> <mi>&amp;rho;</mi> <msub> <mi>S</mi> <mi>j</mi> </msub> </msub> </mtd> </mtr> </mtable> </mfenced> <mo>;</mo> </mrow>

<mrow> <msub> <mi>&amp;rho;</mi> <msub> <mi>S</mi> <mi>j</mi> </msub> </msub> <mo>=</mo> <msub> <mi>&amp;lambda;S</mi> <mi>j</mi> </msub> <mo>;</mo> </mrow>

Wherein c_i,1(t), c_i,2(t) ... c_i,j(t) the i-th server of current period t is held the 1st is corresponded to respectively, 2 ..., j class static file access number；It is to access to load caused by jth class static page, S_jFor jth class static page Size, λ are the resource utilization coefficients for reading static page file and processing transmission, and T is Cycle Length；When t takes (k-1) T, Last i-th server of moment (k-1) T moment of current period i.e. (k-1) individual cycle, which is calculated, by above formula to be handled Static page load capacityK=2,3 ...；

Wherein each server of current period each moment dynamic page load capacity calculation formula to be processed is：

<mrow> <msub> <mi>L</mi> <msub> <mi>D</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mover> <mi>L</mi> <mo>~</mo> </mover> <msub> <mi>D</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>M</mi> <mi>i</mi> </msub> <mo>+</mo> <mfrac> <mrow> <msub> <mi>&amp;rho;</mi> <msub> <mi>D</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <msub> <mi>c</mi> <msub> <mi>D</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> <mi>T</mi> </mfrac> <msub> <mi>w</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>;</mo> </mrow>

WhereinDynamic page load capacity, M are not completed for i-th server before current period t_iFor i-th server Generate the performance parameters of dynamic page, w_i(k-1) current period accessed by a upper cycle is (k-1) individual cycle i-th The optimum allocation weight that platform Java Server Page accesses；Need to buffer for current period t and be forwarded to i-th The dynamic page access number of server,It is corresponding for current period tIndividual dynamic page access number is i-th Load capacity caused by platform server；When t takes (k-1) T, current period i.e. (k-1) individual cycle is calculated most by above formula Later moment in time (k-1) T moment i-th server dynamic page load capacity to be processedK=2,3 ...；

Last moment of current period load factor of each server is：

<mrow> <msub> <mi>R</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mo>(</mo> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mi>T</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <msub> <mi>L</mi> <msub> <mi>S</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mo>(</mo> <mi>k</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mi>T</mi> <mo>)</mo> <mo>+</mo> <msub> <mi>L</mi> <msub> <mi>D</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <mo>(</mo> <mi>k</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mi>T</mi> <mo>)</mo> </mrow> <msub> <mi>C</mi> <mi>i</mi> </msub> </mfrac> <mo>;</mo> </mrow>

Wherein R_i((k-1) T) is that current period is bearing for last i-th server of moment (k-1) T moment of (k-1) individual cycle Load rate.

4. the load-balancing method of web server cluster according to claim 3, it is characterised in that read static page The resource utilization coefficient lambda value that file and processing are sent is as follows：

When server reads static page file and processing from disk, λ=λ is taken₁, λ₁Static page is read from disk for server The resource utilization coefficient that face file and processing are sent,

When server reads static page file and processing from caching, λ=λ is taken₂；λ₂Static page is read from caching for server The resource utilization coefficient that face file and processing are sent；

Wherein λ₁>λ₂。

5. the load-balancing method of web server cluster according to claim 3, it is characterised in that i-th server Maximum processing capability C_iFor：

<mrow> <msub> <mi>C</mi> <mi>i</mi> </msub> <mo>=</mo> <mrow> <mo>(</mo> <msub> <mi>k</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>k</mi> <mn>2</mn> </msub> <mo>,</mo> <msub> <mi>k</mi> <mn>3</mn> </msub> <mo>,</mo> <msub> <mi>k</mi> <mn>4</mn> </msub> <mo>)</mo> </mrow> <mo>&amp;CenterDot;</mo> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <msub> <mi>&amp;mu;</mi> <mi>i</mi> </msub> <msub> <mi>f</mi> <mi>i</mi> </msub> </mtd> </mtr> <mtr> <mtd> <msub> <mi>G</mi> <mi>i</mi> </msub> </mtd> </mtr> <mtr> <mtd> <msub> <mi>D</mi> <mi>i</mi> </msub> </mtd> </mtr> <mtr> <mtd> <msub> <mi>N</mi> <mi>i</mi> </msub> </mtd> </mtr> </mtable> </mfenced> <mo>;</mo> </mrow>

Wherein k₁~k₄Influence coefficient of each index to server maximum processing capability；μ_iFor i-th server CPU quantity, f_iI-th Platform server cpu frequency, G_iFor the memory size of i-th server, D_iFor the I/O speed of i-th server, N_iFor i-th clothes The network throughput of business device.

6. the load-balancing method of web server cluster according to claim 1, it is characterised in that in step S2, according to The load factor of each server of last moment of current period and last moment in a cycle is predicted thereon current period are most The prediction load factor of each server of later moment in time, each clothes of last moment in next cycle are predicted by index smoothing forecasting method The prediction load factor of business device：

<mrow> <msub> <mover> <mi>R</mi> <mo>~</mo> </mover> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mi>T</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>&amp;alpha;R</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mo>(</mo> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mi>T</mi> <mo>)</mo> </mrow> <mo>+</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>&amp;alpha;</mi> <mo>)</mo> </mrow> <msub> <mover> <mi>R</mi> <mo>~</mo> </mover> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mo>(</mo> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mi>T</mi> <mo>)</mo> </mrow> <mo>,</mo> <mi>k</mi> <mo>=</mo> <mn>3</mn> <mo>,</mo> <mn>4</mn> <mo>,</mo> <mo>...</mo> <mo>;</mo> </mrow>

<mrow> <msub> <mover> <mi>R</mi> <mo>~</mo> </mover> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mi>T</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>R</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mo>(</mo> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mi>T</mi> <mo>)</mo> </mrow> <mo>,</mo> <mi>k</mi> <mo>=</mo> <mn>2</mn> <mo>;</mo> </mrow>

Wherein R_i((k-1) T) is that current period is bearing for last i-th server of moment (k-1) T moment of (k-1) individual cycle Load rate,For current period be (k-1) individual cycle a upper cycle i.e. -2 last moment in cycle of kth institute it is pre- The prediction load factor of last each server of moment (k-1) T moment of the current period measured i.e. (k-1) individual cycle,It is the prediction load factor of k-th of kT moment at last moment in cycle, i-th server for next cycle；α is smooth system Number；

Last moment server of next cycle is asked for according to the prediction load factor of each server of last moment in next cycle Consensus forecast load factor：

<mrow> <mover> <mi>R</mi> <mo>&amp;OverBar;</mo> </mover> <mrow> <mo>(</mo> <mi>k</mi> <mi>T</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mi>n</mi> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mover> <mi>R</mi> <mo>~</mo> </mover> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mi>T</mi> <mo>)</mo> </mrow> <mo>;</mo> </mrow>

WhereinIt is k-th of cycle last moment kT moment server consensus forecast load factor for next cycle；N is WEB Total number of units of server in server cluster.

7. the load-balancing method of web server cluster according to claim 1, it is characterised in that built in step S1 Load balancing object function be：

<mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <mi>f</mi> <mo>(</mo> <mrow> <msub> <mi>w</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>w</mi> <mi>n</mi> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>=</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <mfrac> <msqrt> <mrow> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>R</mi> <mi>i</mi> </msub> <mo>(</mo> <mi>k</mi> <mi>T</mi> <mo>)</mo> <mo>-</mo> <mover> <mi>R</mi> <mo>&amp;OverBar;</mo> </mover> <mo>(</mo> <mi>k</mi> <mi>T</mi> <mo>)</mo> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msqrt> <mrow> <mi>n</mi> <mo>-</mo> <mn>1</mn> </mrow> </mfrac> <mo>)</mo> </mrow> <mo>;</mo> </mrow>

The load balancing bound for objective function built in step S1 is：

R_i(kT)≤0.9, i=1,2,3 ... n；

<mrow> <mn>1</mn> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>w</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> <mo>,</mo> <mi>k</mi> <mo>=</mo> <mn>2</mn> <mo>,</mo> <mn>3</mn> <mo>,</mo> <mo>...</mo> <mo>;</mo> </mrow>

Wherein R_i(kT) for next cycle be k-th of kT moment at last moment in cycle each server load factor, It is k-th of cycle last moment kT moment server consensus forecast load factor for next cycle, n takes for web server cluster Total number of units of business device；w_i(k) the optimum allocation weight accessed for i-th Java Server Page of i.e. k-th of cycle in next week, its Middle i=1,2 ... n；T is Cycle Length.

8. the load-balancing method of web server cluster according to claim 7, it is characterised in that in step S2, pass through Cauchy inequality and the method for exhaustion try to achieve the optimal solution in load balancing object function, i.e. optimum allocation weight.

9. the load-balancing method of web server cluster according to claim 1, it is characterised in that server in step S3 When receiving static page access, the static page whether is cached with first, if it is not, then being read from disk；

In step S2, the dynamic page access request at period 1 at each moment is directed to, each server dynamic of the cycle is set The optimum allocation weight of page access is identical.

A kind of 10. load for being used to realize the web server cluster of load-balancing method any one of claim 1 to 9 Equalizing system, including client and web server cluster, the web server cluster include multiple servers；Its feature exists In, in addition to load equalizer, the load equalizer be arranged between client and web server cluster；

The client, for sending accessing page request to load equalizer by internet；

The load equalizer, its server to be distributed is respectively specified that for being accessed for each type static page；For Build load balancing object function and constraints；The load balancing object function is with each clothes of last moment in next cycle The load factor of business device it is corresponding closest to last moment server consensus forecast load factor of next cycle be target, the constraint Condition is moved including each server load rate of last moment in next cycle less than each server of certain value and next cycle State page access distribution weight sum is 1；

The accessing page request proposed for receiving each moment client of current period, after accessing page request is received, The page accessed first request is classified；If static page, the type of the static page accessed is asked in acquisition first, so This type static page is searched afterwards and accesses the server to be distributed specified, and judges that the server currently whether can also Service is provided to page access, if so, static page then is accessed into distribution to the server；If it is not, then static page is accessed Distributed into web server cluster in the server of last moment load factor minimum of the upper cycle, and this type is quiet It is the server that state page access, which specifies the server update to be distributed,；If dynamic page, according to a upper cycle last The optimum allocation weight that each Java Server Page of current period accessed by moment accesses is to current period each moment Dynamic page is accessed and is allocated；

For when the time reaching last moment of current period, gathering last moment of current period in web server cluster The load factor of each server；

Accessed most for when the time reaching last moment of current period, calculating each Java Server Page of next cycle Optimal sorting matches somebody with somebody weight, and detailed process is as follows：First according to the load factor of each server of last moment of current period and upper one week Phase at last moment predicts the prediction load factor of obtained each server of last moment of current period, predicts next cycle The prediction load factor of each server of last moment, it is averagely pre- then to average to obtain next cycle last moment server Load factor is surveyed, will finally be asked in next cycle last moment server consensus forecast load factor input load equalization target function Optimal solution is taken, so as to obtain the optimum allocation weight that each Java Server Page of next cycle accesses；

Each server in the web server cluster, for when the time reaching last moment of current period, calculating it Middle load factor, periodically gathered for load equalizer；Distributed for balancing received load device to static page therein Access and dynamic page accesses；For according to being assigned to, static page therein to be accessed and dynamic page is accessed to client end response Corresponding service.