JP3502009B2 - Load distribution control device and method, and recording medium - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used in a device for allocating access to a group of servers accessed by a client so that loads are balanced.

[0002]

2. Description of the Related Art There is known a load distribution control device for allocating access to a group of servers accessed by a client so that loads are balanced. The following are typical methods for distributing the load on the conventional server. (1) Round Robin Method <Method A> This is a method in which all servers are registered, and each time a request from a client is accepted, the server is cyclically (round robin) assigned.
The most typical implementation is to register a plurality of IP addresses for one domain. Round Robin 1
There is also a weighted round-robin method in which the number assigned in a cycle differs depending on the server specifications. (2) Load measurement method <Method B> This is a method in which the load (CPU usage rate, processing data amount, etc.) of all servers is measured periodically or event-driven, and the server with the least load is assigned.
It is also possible to monitor the server failure and stop the allocation to the failed server.

Further, the hash function is applied to the partial data of the data request packet of the client, and the allocation number of the hash value is determined based on the load condition of the server (depending on the average connection time to the server and the number of connections per unit time). Japanese Patent Laid-Open No. Hei 1 (1999) -1 which aims at load distribution by dynamically changing
There is a method disclosed in Japanese Patent Publication No. 1-96128 <Method C
>. In this method, since the information that can be acquired in the load balancer, such as the average connection time to the server and the number of connections per unit time, is used, there is no control overhead of the load status. Therefore, the measurement of the load condition itself does not increase the load on the server or the network.

As a method of increasing the load distribution to the distributed cache server group and the reuse rate of cache data, there is a method disclosed in Japanese Patent Laid-Open No. 11-224219 <Method D>. In this method, a hash value is calculated for the identification information of the data requested by the client, and the cache server corresponding to the hash value is accessed. There is a synergistic effect that the load is distributed to the cache server by the hash value and the reuse rate of the cache data is improved because the same data is handled by the same server.

[0005]

However, the conventional method has the following problems. In method A, it is hard to say that the cyclic server allocation realizes highly accurate load distribution control without considering the load status of each server. Therefore, there will be a difference in accessibility of the clients. Further, it is difficult to dynamically detect the failure status of each server, and there is a risk that serviceability will be greatly impaired by continuously allocating even servers that cannot provide information. Further, when the server functions as a cache server, the server is assigned without considering the type of request data and the accumulated data of the server, which causes a problem that the reuse rate of the cache data becomes low.

In method B, the accuracy of load balancing control and the ability to isolate a fault are high, but the type of request data is not distinguished as in method A, and the reusability of cache data on the server is low. In addition, there is a problem that the measurement of the load condition itself gives a load to the server and wastes network resources.

Also in the method C, the types of request data are not distinguished, and the reusability of cache data on the server is low.

In the method D, under the condition that the processing load of each request data and the access frequency are uniform, the reuse of the load balancing cache data works well. However, when handling various multimedia contents such as texts, still images, audio data, and moving images, the processing load of each request data greatly differs. Further, considering the characteristic of the Internet in which access is remarkably concentrated on popular contents, accurate load distribution cannot be achieved only by applying a hash to the identification information of request data. Such deterioration of load balancing accuracy may cause problems such as variations in access performance and deterioration of service reliability.

The present invention has been made against such a background. In consideration of the identification information of the request data, the reuse ratio of the cache data is increased and the load balancing control with high accuracy is realized. As well as ensuring the health of network services, the control process is simple,
An object of the present invention is to provide a load balancing control apparatus and method having a low control load and having scalability capable of handling the requests of many clients.

[0010]

The first aspect of the present invention is to:
It is a load balancing control device having means for accommodating clients and controlling access to n servers from the clients.

Here, a feature of the present invention is that the means for controlling access measures a predetermined load measurement item for each of the n servers each time a specific cycle or a request from a client is accepted. Means, a means for calculating the average load value and the respective load values of the n servers according to the measurement result, and a means for calculating a hash value for the identification information of the request data every time a request from the client is accepted. ,
Server S (0 ≦ S ≦ n−1) corresponding to this hash value
And a load balancing execution coefficient α (> 1) and a load balancing permission coefficient β (α ≧ β ≧
1) means for deciding to access the server S if the load value of the server S is less than or equal to α times the average load value, and server S + i mod if the load value is greater than α times the average load value. n (i = 1, 2, ..., N-1)
Means for sequentially searching whether the load value of is less than or equal to β times the average load value, and means for deciding to access this server as soon as a server satisfying the condition is found according to the search result. is there.

Alternatively, when the n servers have different performances, the maximum performance values of the n servers are given in advance, and the means for controlling the access is controlled by a specific cycle or by a client. Means for measuring a predetermined load measurement item for each of the n servers each time the request is received,
A means for calculating the normal average load value and the respective normal load values by interrupting the respective load values of the n servers by the respective maximum performance values according to this measurement result, and a request for each request from the client. A means for calculating a hash value for the identification information of the data and a server S (0 ≦ S ≦ n− corresponding to this hash value.
1) and a predetermined load balancing execution coefficient α (> 1) and a load balancing permission coefficient β (α ≧ β
≧ 1) means for deciding to access the server S if the regular load value of the server S is less than or equal to α times the regular average load value, and the regular load value is more than α times the regular average load value. If it is larger, the server S + i mod n (i ==
1, 2, ..., N-1) means for sequentially searching whether the load value is β times or less of the average load value, and access this server as soon as a server satisfying the condition is found according to the search result. And means to decide what to do.

A second aspect of the present invention is a load balancing control method for controlling access from a client to n servers.

Here, a feature of the present invention is that predetermined load measurement items are measured for each of the n servers each time a specific period or a request from a client is received, and the load measurement items of the n servers are measured. The load values are obtained and the average load values thereof are calculated, and the hash value for the identification information of the request data is calculated each time the request from the client is received, and the server S (0≤S≤ n-1), and a predetermined load balancing execution coefficient α (> 1)
And the load balancing permission coefficient β (α ≧ β ≧ 1), it is determined to access the server S if the load value of the server S is equal to or less than α times the average load value, and if not, the server S + i. mod n (i = 1, 2, ..., N-
Whether or not the load value of 1) is less than or equal to β times the average load value is sequentially examined, and as soon as a server satisfying the condition is found, it is decided to access this server and the process is terminated.

Alternatively, when there is a difference in performance among the n servers, the maximum performance values of the n servers are given in advance, and the n values are given each time a specific period or a request from a client is accepted. Each of the predetermined load measurement items for each server is obtained, the load value of each of the n servers is obtained, and each of them is interrupted by the maximum performance value to obtain a normal load value and based on them. A normal average load value is calculated, a hash value for the identification information of the request data is calculated each time a request from the client is received, and the server S (0 ≦ S
≤n-1) is determined, and a predetermined load balancing execution coefficient α (> 1) and load balancing permission coefficient β (α ≧ β ≧
According to 1), if the regular load value of this server S is less than or equal to α times the regular average load value, it is decided to access this server S, and if not, the server S + i mod n (i
, 1, ..., N-1) are sequentially checked to see if the load value is less than or equal to β times the average load value, and as soon as a server satisfying that condition is found, access to this server is determined and processed. Is about to end.

According to a third aspect of the present invention, a computer device provided with predetermined hardware and predetermined basic software installed in the hardware further installs the computer device on the load of the present invention. It is a recording medium in which software is recorded as a device corresponding to the dispersing device.

Thus, the reuse rate of the cache data is increased in consideration of the identification information of the request data, and
It realizes highly accurate load distribution control, ensures the soundness of network services, and because the control process is simple, it has a low control load and has scalability to handle the requests of many clients.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION The configuration of a load balancer according to an embodiment of the present invention will be described with reference to FIGS. 1 and 2 are diagrams for explaining the operation of acquiring request data by the load balancer according to the embodiment of the present invention. FIG. 3 is a block diagram of a main part of the load balancer according to the embodiment of the present invention.

The present invention, as shown in FIG. 1 and FIG.
Load distribution control devices 110 and 111 that accommodate clients 100 to 103 and control access from the clients 100 to 103 to the servers 120 to 123.

Here, the feature of the present invention is that the load balancing control devices 110 and 111 of the first embodiment of the present invention.
Measures a predetermined load measurement item for each of the servers 120 to 123 each time a specific period or a request from the client 100 to 103 is accepted,
The server load status measuring unit 1 that calculates the average load value and the respective load values of the servers 120 to 123 according to the measurement result, and the hash value for the identification information of the request data each time the request from the client is accepted,
Server S (0 ≦ S ≦ n−1) corresponding to this hash value
The load value of the server S is α times the average load value according to the original server determining unit 2 that determines the load balancing execution coefficient α (> 1) and the load balancing permission coefficient β (α ≧ β ≧ 1). If it is below, it is decided to access this server S, and if the load value is larger than α times the average load value, the server S + i mod n (i = 1, 2, ..., N−).
An optimum server determination unit 3 that sequentially searches whether the load value of 1) is less than or equal to β times the average load value and determines to access this server as soon as a server satisfying the condition is found according to the search result. Is included.

Alternatively, when there is a difference in performance among the servers 120 to 123, the maximum performance values of the servers 120 to 123 are given in advance, and the load balancing control devices 110 and 111 specify the specific period or the clients 100 to 123. Each time a request from the server 103 is accepted, a predetermined load measurement item is measured for each of the servers 120 to 123, and the load value of each of the servers 120 to 123 is interrupted by each maximum performance value in accordance with the measurement result, so that the regular load measurement items are properly specified. A server load status measuring unit 1 that calculates the average load value and each regular load value, and a hash value for the identification information of the request data every time a request from the client 100 to 103 is received, and a server corresponding to this hash value S (0 ≦ S ≦
The normal load value of the server S is normal based on the original server determining unit 2 that determines n-1) and the predetermined load balancing execution coefficient α (> 1) and load balancing permission coefficient β (α ≧ β ≧ 1). If it is less than α times the average load value, this server S
Access to the server, and if the normal load value is greater than α times the normal average load value, the server S + i mod
It is sequentially searched whether the load value of n (i == 1, 2, ..., N-1) is less than or equal to β times the average load value, and according to the search result, this server that meets the conditions is found. And an optimum server determining unit 3 that determines to access the.

The load balancing control devices 110 and 111 according to the embodiment of the present invention are further installed in a computer device provided with predetermined hardware and predetermined basic software installed in the hardware, thereby the computer device. The personal computer device is used as a device corresponding to the load balancing devices 110 and 111 by reading the recording medium on which software is recorded, which is a device corresponding to the load balancing device of the present invention, into the personal computer device.

Alternatively, in the form of FIG. 1, the middleware functions on the clients 100 to 103 and the server 120 to
It can also be implemented as one function of 123, a function of an application switch (layer 7 switch) on the network, or a primary proxy server function. Further, in the form of FIG. 2, it can be implemented as a name service application function on the clients 100 to 103, one function of the servers 120 to 123, and an additional function of a name server on the network.

The embodiments of the present invention will be described in more detail. Reference numerals 100 to 103 in FIGS. 1 and 2 denote clients that issue data acquisition requests. Reference numerals 110 to 111 denote load distribution control devices according to the embodiment of the present invention, and a plurality of load distribution control devices may exist. 1 and 2, the load balancing control device 1
10 accommodates the clients 100 and 101, and the load balancing control device 111 controls the clients 102 and 103.
Indicates that the Reference numerals 120 to 12
3 is an individual server, and reference numeral 140 is a server group.

The information providing function of the server group 140 provides original contents, and those contents can be obtained from any of the servers 120 to 123. In addition, the cache function of the server group 140 is provided by the remote server 130 via the Internet or the like.
˜132, instead of the clients 100 to 103, the data is acquired and stored in the cache areas of the servers 120 to 123, and at the same time, the data is provided to the clients 100 to 103.

The accumulated cache data is reused and provided to the clients 100 to 103 without being reacquired from the remote servers 130 to 132 when the same data acquisition request is made again from the clients 100 to 103. , Remote servers 130-132 and server 120
1 to 123, the consumption amount of network resources is reduced and the accessibility of the clients 100 to 103 is improved.

Load balancing control devices 110 and 1 of the present invention
11 can be applied when the servers 120 to 123 have the above-mentioned two functions, or when only one of the two functions is available.
Is most effective when has a cache function.

Next, the flow of data acquisition by the client 101 will be described with reference to FIG. The example of FIG. 1 illustrates a case where the client 101 acquires data from the server 120 via the load balancing control device 110.
The client 101 issues a data acquisition request to the load balancing control device 110. The load balancing control device 110 selects one server 120 and issues a data acquisition request to the server 120. Here, if the server 120 has already accumulated the requested data, the process proceeds to. The server 120 issues a data acquisition request to the remote server 130 that is the source of the request data. The remote server 130 sends the data to the server 120, and the server 120 stores the acquired data in the cache area. The server 120 transmits data to the load balancing control device 110. Load balancing control function 110
Transmits the acquired data to the client 101 as it is.

The flow in which the client 101 of FIG. 2 acquires data will be described. The example of FIG. 2 illustrates a case where the client 101 acquires data directly from the server 120 by the access control of the load distribution control device 110. The client 101 inquires of the load balancing control device 110 about the addresses of the servers 120 to 123. The load balancing control device 110 selects one server 120,
The address is returned to the client 101. The client 101 again issues a data acquisition request to the server 120. If the server 120 has already stored the requested data, the process proceeds to. The server 120 issues a data acquisition request to the remote server 130 that is the source of the request data. The remote server 130 sends the data to the server 120, and the server 120 stores the acquired data in the cache area. The server 120 sends the data to the client 101.

Next, as the upper application, WW
Taking the W service as an example, the operation of the load balancing control apparatus according to the embodiment of the present invention will be specifically described with reference to FIG. The “identification information of request data” 4 in FIG. 3 is information for uniquely identifying the request data, and in the WWW, the character string of the URL (for example, http://www.XXX.XXX/index.html) is used. Equivalent to. The “access destination server” 5 is the IP address of the access destination server determined by the load balancing control device (for example,
123.45.67.89). "Network service server information" 6, "Load measurement item" 7, "Load distribution execution coefficient α" as information that the administrator should give in advance
8 and “load distribution permission coefficient β” 9.

In the "network service server information" 6, as shown in FIG. 4, "the number n of network service servers" and the serial number (# 0- # n-) of each server.
1), IP address, and maximum performance value are described. However, if there is no difference in the processing capabilities of the servers, the information on the maximum performance value is unnecessary.

In "load measurement item" 7, an item serving as a reference for load distribution is given. This depends on the application, network environment, and traffic conditions,
It should give a bottleneck factor of 0-123. For example, the CPU usage rate, the data processing frequency, the data processing amount per unit time, and information obtained by combining these values and the like can be cited. Among these, when information such as the data processing frequency is given as the “load measurement item” 7, it can be calculated in the load balancing control devices 110 and 111 without communicating with the servers 120 to 123. However, when there are a plurality of load balancing control devices 110 and 111, the states of the other load balancing control devices 110 and 111 are not considered,
It is sufficient to calculate independently. If normal load distribution is performed in the single load distribution control devices 110 and 111, even if the entire service is a combination of these,
This is because normal load distribution is performed.

When the load distribution control is performed by giving the CPU usage rate or the like as the "load measurement item" 7, the server load status measuring unit 1 communicates with each server periodically or every time a request from the client is accepted. Get measurements.

The "load distribution execution coefficient α"(> 1) 8 is a value that defines the accuracy of load distribution. The “load distribution permission coefficient β” (1 ≦ β ≦ α) 9 is a value for preventing data requests from being distributed to a server having a relatively high load.

Next, a set of internal variables will be described. The internal variables include “(normal) load value of network service server” 10 and “(normal) average load value” 11. “(Regular) load value of network service server” 10 is
It is a quantity value related to load measurement items. The present invention determines the normal state or overloaded state of the server based on these values and performs load balancing control. When there is a difference in the performance of the servers 120 to 123 and it is necessary to obtain the regular load value, it can be obtained by (quantity value related to load measurement item) / (maximum performance value described in "Network service server information"). Good. “(Regular) average load value” 11 is all “(normal) load value of network service server”
It is an average value of 10.

Originally, the server determination unit 2 will be described with reference to FIG. From the optimum server determination unit 3, “identification information of request data” 4 (example: http://www.XXX.XXX/index.h
tml). The “network service server information” 6 is accessed and the number n of servers is received. A hash function is applied to "identification information of request data" 4 to obtain hash values # 0 to # n-1. For the hash function, it is necessary to use a method such as MD5 so that the set of "identification information of request data" 4 is uniformly distributed in # 0 to # n-1. The hash value is returned to the optimum server determination unit 3.

The server load condition measuring unit 1 will be described with reference to FIGS. 6 and 7. Here, "Load measurement item"
7 is the data request frequency (the number of requests per unit time). The cumulative processing number up to that point is measured periodically (t1 seconds), and the “load value of the network service server” 10 is given by the number of requests in the past t2 (t2 = a · tl, a is a natural number) seconds. .

To perform the above processing, a data structure as shown in FIG. 6 is given. The array 600 holds the correspondence between the serial numbers of the servers and the counters representing the cumulative number of processes performed so far. When the load balancing control device of the present invention determines that the access destination of the requested data is p (0 ≦ p ≦ n−1), the serial number p
Add 1 to the counter. This counter value is measured every tl seconds and recorded as a record in the database 601. The database 601 holds records for the past t2 seconds, and the difference between the latest record value and the record value before t2 seconds is defined as "load value of network service server" 10.

The flow of the server load status measuring unit 1 will be described with reference to FIG. Here, the measurement timing of each network service server is slid. Set the measurement period tl. The number of servers n is obtained from the “network service server information” 6. Internal variable i is 0
Initialize to. The subprocess L is called with i as an argument. Increment i. However, when i = n, i = 0. Wait tl / n seconds and return to.

The flow of the sub-process L will be described. The counter value of the serial number i in the array 600 is acquired. The value is added as a new record of the serial number i in the database 601. At that time, if a record before t2 seconds exists, it is deleted. For the serial number i, the difference between the latest record value and the record value t2 seconds ago is recorded as the “load value of the network service server i”.

The flow of the optimum server determining unit 3 will be described with reference to FIG. "Identification information of request data" 4 (Example: htt
p: //www.XXX.XXX/index.html). Originally, the server determining unit 2 is called to acquire the original server S for the requested data. The load value of the server S is obtained from “load value of network service server” 10. When the load value of the server S ≦ “average load value” × α, the server S ends as the access destination. If not, proceed to. Initialize the internal variable j = 1. Server S + j
If the load value of mod n ≦ “average load value” × β, the server is determined to be the access destination and the process ends. Otherwise, increment the value of j and repeat. Here, it is clear that there is always one server whose load value is less than or equal to the average load value among the servers, so the above-described iterative process is always terminated.

Here, the load distribution execution coefficient α and the load distribution permission coefficient β will be described. As is clear from the flow of the optimum server determination unit 3, the load balancing execution coefficient α
By defining, the load balancing control is performed so that the highest-loaded server among the servers has a load that is at most α times the average load value. The closer α is to 1, the more accurate the load balancing can be. However, in that case, the probability that the data request is originally not assigned to the server increases. From the viewpoint of the effectiveness of the cache function, when the same data is processed by the same server (that is, the original server), the reuse rate of cache data increases. Reuse of cache data cannot be expected so much except for servers. However, since the distribution destination for the server S is fixedly prioritized as S + immod n (i = 1, 2, ...), the cache data can be reused as the value of i decreases. It can be said that this mechanism enhances the effectiveness of the cache function. For the above reasons, it is desirable to adjust the value of α appropriately.

The load distribution permission coefficient β (1 ≦ β ≦ α) is for preventing the data request from being distributed to the network service server having a relatively high load. This is an idea similar to the line reservation technology in the telephone network. By distributing a certain data request to the server A, the load value of the server A exceeds α, and the data request that should be originally assigned to the server A is Avoid situations where you have to distribute to different servers. Since it is basically issued to the server originally assigned by the hash function,
Originally, the reuse rate of cache data other than the server is not high. As a whole, the less frequently the data requests are distributed, the more effectively the cache can function. When β is 1, the data requests that originally overflowed the allocation to the servers are distributed to the servers having the average load value or less, and the load distribution efficiency is good. However, since the distribution destination server is limited to a server with a light load, reuse of the cache data cannot be expected so much even if the priority order is followed.
For the above reason, it is desirable to adjust the value of β appropriately.

Since the present invention can be realized by a very simple process, the control processing load is relatively low, and the demands of many clients can be handled. Further, it is expected that the hardware can be easily implemented by using a dedicated IC.

As described above, in the load balancing control apparatus according to the embodiment of the present invention, since the identification information of each request data is used to determine the server by the hash function, the same data is handled by the same server. Is high. Therefore, since the reuse rate of cache data can be increased, the effects of reducing the amount of network resources consumed between remote servers and improving the accessibility to clients can be expected. Furthermore, by defining the load distribution execution coefficient α, the load distribution control accuracy is achieved in which the load of an arbitrary server is no more than α times the average load, stabilizing the service quality and improving the reliability of the service. The effect of can be expected. further. Since the load balancing control mechanism is simple, the control processing load is also relatively low,
There is a scalability that can handle the requests of many clients.

[0046]

As described above, according to the present invention,
In consideration of the identification information of the request data, while increasing the reuse rate of cache data, realizing highly accurate load balancing control, ensuring the soundness of network services, and controlling the control process because it is simple, It has low scalability and can handle the requests of many clients.

[Brief description of drawings]

FIG. 1 is a diagram for explaining an operation of acquiring request data by a load distribution control device according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining an operation of acquiring request data by the load distribution control device according to the embodiment of this invention.

FIG. 3 is a block diagram of a main part of the load distribution control device according to the embodiment of the present invention.

FIG. 4 is a diagram showing an example of “network service server information”.

FIG. 5 is a flowchart originally showing the operation of the server determination unit.

FIG. 6 is a configuration diagram of a data array and a database for the server load status measurement unit.

FIG. 7 is a flowchart showing the operation of a server load status measuring unit.

FIG. 8 is a flowchart showing the operation of the optimum server determination unit.

[Explanation of symbols]

1 Server load status measurement unit 2 Originally the server decision unit 3 Optimal server determination unit 4 Request data identification information 5 Access destination server 6 Network service server information 7 Load measurement items 8 Load balancing execution coefficient α 9 Load balancing permission coefficient β 10 (normal) load value of network service server 11 (Regular) average load value 100-103 clients 110, 111 load balancing control device 120-123 servers 130-132 remote server 140 servers 600 arrays 601 database

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-11-224219 (JP, A) JP-A-11-96128 (JP, A) JP-A-11-27320 (JP, A) JP-A-2000-207270 (JP, A) Nakajima Isami Hideaki Horimai Ryutaro Matsumura Hiroka Minami, Server Centralized Traffic Distribution Technology, NTT Technology Journal 1997.11, November 1, 1997, Volume 11, No. 11, p. 27-29 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 13/00 G06F 15/16

Claims (6)

(57) [Claims] 1. A load balancing control device comprising means for accommodating a client and controlling access to n servers from the client, wherein the means for controlling access receives a request from a specific cycle or a client. A means for measuring a predetermined load measurement item for each of the n servers, a means for calculating an average load value and each load value for the n servers according to the measurement result, and a request from a client Means for calculating the hash value for the identification information of the request data and the server S (0 ≦ S ≦ n− corresponding to this hash value.
1) and a predetermined load balancing execution coefficient α (> 1) and a load balancing permission coefficient β (α ≧ β
≧ 1) means for deciding to access the server S if the load value of the server S is less than or equal to the average load value α times, and server S + imod if the load value is greater than the average load value α times. n (i = 1, 2, ..., N-
It includes means for sequentially searching whether the load value of 1) is less than or equal to β times the average load value, and means for deciding to access this server as soon as a server satisfying the condition is found according to the search result. A load balancing control device characterized by the above. 2. A load balancing control device comprising means for accommodating a client and controlling access from the client to n servers, wherein a maximum performance value of each of the n servers is given in advance, and The means for controlling access measures the predetermined load measurement items for the n servers each time a specific period or a request from a client is accepted, and each of the n servers according to the measurement result. Means for calculating the n normal load values and the respective normal load values by interrupting the load value of each by the maximum performance value, and a hash value for the identification information of the request data each time the request from the client is accepted. A means for calculating and a server S (0≤S≤n-1 corresponding to this hash value Means for determining, said by a predetermined load distribution performance factor alpha (> 1) and the load distribution permission coefficient β (α ≧ β ≧ 1) server S
Means for deciding to access this server S if the normal load value of is less than or equal to α times the normal average load value, and if the normal load value is greater than α times the normal average load value, the server S + immod n (i = 1, 2, ..., N-1) means for sequentially searching whether the load value is less than or equal to β times the average load value, and accesses this server as soon as a server satisfying the condition is found according to the search result. And a means for deciding to do so. 3. A load distribution control method for controlling access from a client to n servers, wherein a predetermined load measurement item is measured for each of the n servers each time a specific period or a request from the client is accepted. Then, the load value of each of the n servers is obtained and the average load value thereof is calculated, and the hash value for the identification information of the request data is calculated each time the request from the client is accepted, and the hash value is corresponded to. Which server S (0 ≦ S ≦ n−1) has been determined, and a predetermined load balancing execution coefficient α (> 1)
And the load distribution permission coefficient β (α ≧ β ≧ 1), it is determined to access the server S if the load value of the server S is less than or equal to α times the average load value, and if not, the server S + imdn (I = 1, 2, ..., N-1)
It is sequentially checked whether the load value of is less than β times the average load value,
A load balancing control method characterized in that as soon as a server satisfying the condition is found, it is decided to access this server and the processing is terminated. 4. A load balancing control method for controlling access from a client to n servers, wherein maximum performance values of the n servers are given in advance, and a specific cycle or every time a request from a client is accepted. , The load measurement values of the n servers are respectively measured, the load values of the n servers are respectively obtained, and the load values are interrupted by the individual maximum performance values to obtain the normal load values. The normal load average value of the n units is calculated based on the above, a hash value for the identification information of the request data is calculated every time the request from the client is received, and the server S (0 ≦ S ≦ n-1), and a predetermined load balancing execution coefficient α (> 1)
And the load distribution permission coefficient β (α ≧ β ≧ 1), it is determined that the server S is accessed if the regular load value of this server S is less than or equal to α times the regular average load value, and if not, the server S + imdn (I == 1, 2, ..., N
A load characterized by sequentially checking whether the load value of -1) is less than or equal to β times the average load value, deciding to access this server as soon as a server satisfying the condition is found, and ending the processing. Distributed control method. 5. A specific cycle or client is installed in a computer device having predetermined hardware and predetermined basic software installed in the hardware, thereby installing the specific period or client in the computer device.
Each time it receives a request from Ant,
Load measurement items that are predetermined by
And obtain the load values of the n servers, respectively.
From the client, calculate the average load value
Each time the request of
A server that calculates the hash value and corresponds to this hash value
S (0 ≦ S ≦ n−1) is determined and predetermined
Load balancing execution coefficient α (> 1) and load balancing permission coefficient β
Due to (α ≧ β ≧ 1), the load value of this server S is negative on average.
If it is less than α times the load value, you can access this server S.
And if not, server S + imdn
The load value of (i = 1, 2, ..., N-1) is the average load value β
It is sequentially checked whether it is less than double, and this server that satisfies the condition is
Decided to access this server as soon as it is found
A recording medium on which software for executing a procedure for ending the process is recorded. 6. Predetermined hardware and this hardware
The specified basic software installed on
Installed on the computer device
The maximum number of the above n servers can be
Performance values are given in advance, and the specific cycle
Each time it receives a request from a client,
Predetermined load measurement items for each server
Each is measured and the load value of each of the n servers is measured.
Get and interrupt them by individual maximum performance values
The normal load value is set according to
Calculate the normal average load value of n units,
Each time the request is received, the identification information of the request data
The hash value that is calculated is calculated, and the server corresponding to this hash value is calculated.
S (0 ≦ S ≦ n−1) is determined and determined in advance.
Load balancing execution coefficient α (> 1) and load balancing permission coefficient
The normal load value of this server S is positive due to β (α ≧ β ≧ 1)
If it is less than α times the standard load value, access to this server S
Server S + im otherwise
The load value of odn (i == 1, 2, ..., N-1) is negative in average.
Sequentially check if it is less than β times the load value and satisfy the condition
You can access this server as soon as it is found.
The software that executes the procedure for deciding
A recording medium that records the air .