JP2006228115A - Load distribution prediction device, load distribution control method, program thereof, load distribution control system, and load distribution control method for this system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a load distribution control system which is capable of furthermore reducing processing for load distribution of a load distribution device itself by using statistical methods. <P>SOLUTION: A load distribution prediction device 4 generates a prediction table showing whether a spare server 3 is to be used or not, on the basis of statistics of presence/no presence of load of processing in a load distribution device 1, and the load distribution device 1 distributes load using the prediction table. The load distribution prediction device 4 corrects the prediction table in accordance with a probability that use/non-use of the space server 3 which is shown by the prediction table coincide with expected use/non-use. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a load distribution prediction apparatus, a load distribution control method, a program, a load distribution control system, and a load distribution control method in the system, which distribute processing to spare servers when the processing load increases.

2. Description of the Related Art Conventionally, in a communication network technique, a load distribution technique that distributes processing to a plurality of servers is used so that the processing load is not concentrated on a specific server. In this load distribution technique, for example, when a load distribution apparatus receives information via a communication network and distributes the information to a processing server, the amount of the received information is not less than a predetermined amount per unit time. In this case, there is a technique in which information is transferred to other spare servers of the processing server and information processing is performed by a plurality of servers. As a load balancing technique, Patent Literature 1 is disclosed.
JP 2004-147060 A

  Here, the above-described technique for load distribution is a technique for distributing information to the spare server at the time of high load. Repeat the process of connecting and disconnecting. For example, when this repetitive process occurs in a short time, the load is further increased by the repetitive process, causing a reduction in the process. Further, since load balancing processing is not accumulated in the spare server, statistical use / non-use information of the spare server is not used, which is inefficient.

  Therefore, the present invention provides a load distribution prediction apparatus, a load distribution control method, a program thereof, a load distribution control system, and the system that can further reduce the load distribution process of the load distribution apparatus itself by using a statistical method. The purpose of this is to provide a load balancing control method.

  The present invention was made to solve the above-described problem, and is information on the presence / absence of a load used for determining whether or not to use a spare server used only for load reduction during load balancing processing. And load information receiving means for receiving, as needed, load presence / absence information indicating whether or not the load exceeds a preset load threshold, and information on the load presence / absence for each small period divided by a predetermined interval And the use of the spare server in each future small period divided in the same manner as the predetermined period and the small period based on the probability of the presence or absence of load for each of the small periods over a plurality of the predetermined periods. Based on prediction table generation means for generating non-use information as a prediction table, and information on the presence / absence of the load received after the generation of the prediction table, expected use / non-use information of the spare server for each small period Inspection Then, it is determined whether or not the spare server use / non-use information for each small period in the prediction table matches, and if the determination matches, the prediction table is positive and the determination does not match In some cases, the correct / incorrect accumulation table generating means for generating the correct / incorrect accumulation table indicating that the prediction table is erroneous for each of the small periods over the predetermined period, and the correct / incorrect accumulation table includes a plurality of the predetermined periods. A load comprising: a prediction table correcting unit that corrects the use / non-use information of the preliminary server shown in the prediction table based on the probability of correctness of the prediction table for each small period It is a distributed prediction device.

  Further, the present invention determines that the spare server is used when the load presence / absence information indicates that there is a load, and when the load presence / absence information indicates that there is no load, the spare server The information is determined to be non-use.

  In the invention, it is preferable that the load threshold value is at least one or more values of a communication data amount, a CPU load factor, and a memory area usage rate.

  In the invention, it is preferable that the predetermined period is one week and the small period is one day.

  According to the present invention, in the above-described load distribution prediction apparatus, a prediction table transmission unit that transmits the prediction table to a load distribution apparatus that distributes processing to the spare server based on use / non-use indicated by the prediction table; It is characterized by providing.

  The present invention is also a load distribution control method in a load distribution prediction apparatus, wherein the load information receiving means of the load distribution prediction apparatus uses a spare server that is used only for load reduction during load distribution processing. The load presence / absence information used to determine whether or not the load exceeds a preset load threshold is received as needed, and the load distribution prediction apparatus generates a prediction table. The means is based on the information on presence / absence of the load for each small period obtained by dividing the predetermined period at a predetermined interval and the probability of the presence / absence of the load for each of the small periods over a plurality of the predetermined periods. And the use / non-use information of the spare server in each future small period divided into the same period as the small period as a prediction table, and a correct / incorrect storage table generating means of the load distribution prediction device includes: Based on the information on the presence / absence of the load received after the generation, the expected use / non-use information of the spare server for each small period is detected, and the use / non-use of the spare server for each small period in the prediction table The prediction table is positive if the determination matches, and if the determination does not match, the correct / incorrect storage table indicating that the prediction table is incorrect is The prediction table correction means of the load distribution prediction device generates for each predetermined period, and the prediction table correction means of the load distribution prediction apparatus is one of the correctness and incorrectness of the prediction table for each of the small periods over the predetermined period in the correct / incorrect storage table. The load distribution control method is characterized in that, based on the probability, the use / non-use information of the spare server in the previous period indicated by the prediction table is corrected.

  The present invention is also a program to be executed by a computer of a load distribution prediction apparatus, which is information on the presence / absence of a load used for determining whether or not to use a spare server that is used only for load distribution processing. And a process for receiving load information indicating whether or not the load exceeds a preset load threshold as needed, and information on the load presence / absence for each small period obtained by dividing a predetermined period by a predetermined interval. And the use of the spare server in each future small period divided in the same manner as the predetermined period and the small period based on the probability of the presence or absence of load for each of the small periods over a plurality of the predetermined periods. Based on the process of generating non-use information as a prediction table and the load presence / absence information received after the generation of the prediction table, information on the expected use / non-use of the spare server for each small period And whether the spare server usage / non-use information for each of the small periods in the prediction table matches, and if the determination matches, the prediction table is positive, and the determination is In the case where they do not match, a process for generating a correct / incorrect accumulation table indicating that the prediction table is incorrect for each of the small periods over the predetermined period, and in the correct / incorrect accumulation table, the small and long accumulation tables over a plurality of the predetermined periods. This is a program for causing a computer to execute a process of correcting information on the use / non-use of the preliminary server indicated by the prediction table based on either probability of correctness of the prediction table for each period.

  The present invention is also a load distribution control system including a load distribution prediction device and a load distribution device, wherein the load distribution prediction device uses a spare server that is used for the purpose of reducing the load only during load distribution processing. Load information reception information that is used to determine whether or not the load is received, and information on whether or not the load has exceeded a preset load threshold is received from the load balancer at any time. Based on the means, information on the presence / absence of the load for each small period obtained by dividing the predetermined period at a constant interval, and the probability of the presence / absence of the load for each of the small periods over a plurality of the predetermined periods. And a prediction table generating means for generating, as a prediction table, use / non-use information of the spare server in each future small period divided into periods similar to the small period, and the use / Based on the prediction table transmission means for transmitting the processing to the spare server based on the usage, and the load presence / absence information received from the load distribution device after the generation of the prediction table, it is expected for each small period. The spare server use / non-use information is detected, and it is determined whether or not the spare server use / non-use information for each small period in the prediction table matches. A correct / incorrect accumulation table generating means for generating a correct / incorrect accumulation table indicating that the prediction table is positive and the prediction table is incorrect when the determinations do not match, over the predetermined period for each of the small periods; Prediction that corrects the use / non-use information of the preliminary server shown in the prediction table based on the probability of correctness of the prediction table for each small period over a plurality of the predetermined periods in the correct / incorrect storage table Table repair And means, wherein the prediction table transmitting means is a load distribution control system and transmits the prediction table said corrected to the load balancer.

  The present invention is also a load distribution control method in a load distribution control system comprising a load distribution prediction device and a load distribution device, wherein the load information receiving means of the load distribution prediction device reduces the load only during load distribution processing. Load presence / absence information used for determining whether or not to use a spare server used as a purpose, and whether or not the load exceeds a preset load threshold, The prediction table generating means of the load distribution prediction device receives the information on the presence / absence of the load for each small period obtained by dividing the predetermined period at a predetermined interval, and the presence / absence of the load for each of the small periods over a plurality of the predetermined periods. Based on one of the probabilities, the spare server use / non-use information is generated as a prediction table in each future small period divided into the predetermined period and the small period, and the load distribution is performed. The prediction table transmission means of the measuring device transmits the prediction table to a load distribution device that distributes the processing to the spare server based on use / non-use indicated by the prediction table, and generates a correct / incorrect storage table of the load distribution prediction device The means detects the expected use / non-use of the spare server for each small period from the information on the presence / absence of the load received after the generation of the prediction table, and the information for each small period in the prediction table It is determined whether the information matches the usage / non-use information of the spare server. If the determination matches, the prediction table is positive, and if the determination does not match, the prediction table indicates that the prediction table is incorrect. A correct / incorrect accumulation table is generated for each of the small periods over the predetermined period, and the prediction table correcting means of the load distribution prediction apparatus predicts for each of the small periods over the predetermined period in the correct / incorrect accumulation table. Any of table correctness Based on the probability, the use / non-use information of the preliminary server shown in the prediction table is corrected, and the prediction table transmission means of the load distribution prediction device transmits the corrected prediction table to the load distribution device This is a load distribution control method.

According to the present invention, a prediction table indicating whether or not a spare server is to be used is generated based on statistics on the presence or absence of processing load in the load distribution device, and the load distribution device performs load distribution using the prediction table. Further, the prediction table is modified according to the probability of whether the use / non-use of the spare server indicated in the prediction table matches the expected use / non-use.
As a result, the prediction table is changed according to the statistics of the presence / absence of processing load in the load balancer, and load distribution processing is performed on the spare server in a time zone in which it is assumed that there is a load in advance according to the prediction table. Compared to the technology that performs load distribution processing after the occurrence of a load, the load distribution processing can be performed smoothly during a time period when the load suddenly increases, and the load distribution processing itself can be performed by the load distribution processing itself. It is possible to reduce the decrease in the information transfer process.
In the case of a technique for determining whether or not to use a spare server depending on whether or not the load threshold is exceeded, when the threshold value is raised or lowered in a short time, processing of using or not using the spare server repeatedly occurs. However, in this process, whether or not to use the spare server is determined in advance in the prediction table, so that it is possible to reduce the degradation of the processing of the load balancer due to repeated use / nonuse of the spare server in a short time. it can.
Moreover, since the prediction table is modified according to the prediction table, even if a change in the time zone with or without load occurs with the passage of time, the change can be dealt with.

  Hereinafter, a load distribution control system according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a load distribution control system according to the embodiment. In this figure, reference numeral 1 denotes a load balancer that distributes received information to each server for distribution. Reference numeral 2 denotes a main server that always receives information from the load balancer 1 and performs predetermined processing. Further, when the amount of information processed per unit time in the load balancer 1 exceeds a predetermined threshold, 3 receives information distributed to the main server 2 from the load balancer 1 like the main server 2. This is a spare server that performs the same processing as the main server 2. Reference numeral 4 denotes a prediction table used by the load balancer 1, and generates a prediction table used to determine whether or not to distribute information (that is, load distribution) to the spare server 3 in the load distribution process. It is a load distribution prediction device. The load balancer 1, the main server 2, the spare server 3, and the load balance prediction device 4 are connected via an internal communication network. Further, the load balancer 1 receives information transmitted from the external communication network to a server group including the main server 2 and the spare server 3 and performs processing for transferring the information to the main server 2 or the spare server 3. .

  In this load distribution control system, the load distribution apparatus 1 receives the prediction table generated by the load distribution prediction apparatus 4 from the load distribution prediction apparatus 4 and reserves it before becoming high load according to the information shown in the prediction table. Connection to the server 3 and distribution of information to the spare server 3 are performed. Although details of the prediction table will be described later, for example, a table that holds information indicating whether or not to connect to the spare server 3 in a short period of 4 hours over a plurality of days (that is, from 0 to 4 o'clock). Whether to connect to the prediction server 3, whether to connect to the prediction server 3 between 4 o'clock and 8 o'clock, ... whether to connect to the prediction server 3 between 20 o'clock and 24 o'clock Data). The prediction table is appropriately corrected in the load distribution prediction device 4, and each time the correction table is corrected, the prediction table is transmitted to the load distribution device 1. When the load distribution device 1 receives a new prediction table, the load distribution device 1 connects to the spare server 3 and distributes the information to the spare server 3 before the load becomes high according to the information shown in the new prediction table. Do.

FIG. 2 is a functional block diagram showing functional configurations of the load distribution device and the load distribution prediction device.
Next, functional configurations of the load distribution device and the load distribution prediction device will be described with reference to FIG.
As shown in FIG. 2, first, in the load distribution apparatus 1, reference numeral 11 denotes a communication processing unit that transmits and receives information to and from other apparatuses via an internal / external communication network. A control unit 12 controls each processing unit in the load distribution apparatus 1. Reference numeral 13 denotes a load detection unit that detects whether or not the load of information processing has exceeded a load threshold value and transmits the load presence / absence information to the load distribution prediction device 4. A load distribution processing unit 14 determines whether or not to distribute information to the spare server 3 based on the prediction table, and distributes information to the main server 2 and the spare server 3. A database 15 stores various data.

  In the load distribution prediction apparatus 4, reference numeral 21 denotes a communication processing unit that transmits and receives information to and from the load distribution apparatus 1. A control unit 22 controls each processing unit of the load distribution prediction apparatus 4. Reference numeral 23 denotes a prediction table generation unit that performs processing for generating a prediction table. Reference numeral 24 denotes a correct / incorrect accumulation table generator for performing a process of generating an accurate / incorrect accumulation table used for correcting the prediction table. Reference numeral 25 denotes a prediction table correction unit that corrects the prediction table using the correct / incorrect accumulation table. A database 26 stores various data.

Next, the processing flow of the load distribution control system will be described sequentially.
FIG. 3 is a diagram showing a processing flow until the load distribution prediction apparatus generates a prediction table.
First, the load detector 13 of the load distribution apparatus 1 detects whether or not the information processing load exceeds the load threshold (step S101). Here, the information processing load is the processing load on the computer system, such as the amount of information received via the external network (for example, the amount of information received per unit time), the load factor of the CPU, and the memory area usage rate. The load detection unit 13 uses one or more of these pieces of information in detecting the load. In the detection, a predetermined load threshold in the received amount of information per unit time, CPU load factor, memory area usage rate, and the like is compared with the current processing load information, and the current processing is performed. If the load exceeds the threshold, it is determined that there is a load, and if it does not exceed the threshold, it is detected that there is no load. In addition, the load detection unit 13 always detects information indicating the presence or absence of a load, and performs a process of transmitting information about the presence or absence of a load in response to an inquiry from the load distribution prediction device 4.

  The prediction table generation unit 23 of the load distribution prediction device 4 determines whether or not the prediction table has already been recorded in the database when the computer of the load distribution prediction device 4 is started for the first time or at the first stage in the load distribution control system. Is detected (step S102). Then, when the prediction table is not recorded in the database 26 or the like, the generation process of the prediction table is started (step S103). If a prediction table has already been generated and recorded in the database 26, a new prediction table is generated, so the existing prediction table is deleted from the database 26 (step S104). Here, the prediction table generation unit 23 starts generating a prediction table according to the parameters set by the user. The parameters are n (predetermined period) = 1 day, t (small period) = 4 hours, N (prediction). Table generation correction interval) = 10 days, T (predetermined period elapsed detection time) = 24 hours. These parameters are recorded in the memory and the database 26.

  First, the prediction table generation unit 23 sends inquiry information about the presence / absence of load to the load balancer 1 every t = 4 hours to inquire about the information on presence / absence of load (step S105). This inquiry information about the presence / absence of the load is transmitted to the load balancer 1 every t = 4 hours even after the generation of the prediction table. The load detection unit 13 of the load distribution apparatus 1 instructs the communication processing unit 11 to transmit the load presence / absence information at that time to the load distribution prediction apparatus 4 each time the load presence / absence inquiry information is received. Thereby, the load balancer 1 transmits the load presence / absence information to the load balance prediction device 4 every t = 4 hours (step S106). Then, the prediction table generation unit 23 sequentially accumulates the load presence / absence information received every t = 4 hours in the load data time series accumulation table of the database 26 (step S107). At this time, the prediction table generation unit 23 determines that one day has passed when the load presence / absence information is received six times, and when it detects that N = 10 days have passed, the load presence / absence information received over 10 days. A prediction table is generated from the load data time series accumulation table created based on the above.

FIG. 4 is a diagram showing a processing image when a prediction table is generated from the load data time series accumulation table.
Now, in the load data time series accumulation table as shown in FIG. 4, ◯ indicates that there is a load, and × indicates that there is no load. Then, use / non-use of the spare server 3 for each time period of t = 4 hours is determined according to the probability of having / no load for each time period of t = 4 hours over N = 10 days. To do. For example, when the load probability pX in N = 10 days in the time zone t1 from 0:00 to 4:00 is 70% or more, the spare server 3 is set in the time zone from 0:00 to 4:00. Decide to use. It is detected whether use / non-use of the spare server 3 is determined in all t = 4 hours (step S108). If not determined, similarly, from 4 o'clock to 8 o'clock, from 8 o'clock 12:00 ... It is determined whether or not the spare server 3 is to be used in all t = 4 hours from 20:00 to 24:00. And thereby, the prediction table | surface which shows whether the spare server 3 is utilized in each time slot | zone for every t = 4 hours is produced | generated (step S109).

  In the prediction table of FIG. 4, ◯ indicates that the spare server 3 is used, and × indicates that the spare server 3 is not used. Then, the prediction table generation unit 23 records the generated prediction table in the database and instructs the communication processing unit 21 to transmit the prediction table to the load distribution apparatus 1 via the communication network. When the load balancer 1 receives the prediction table, the load distribution processing unit 14 records the prediction table in the database 15 and determines whether the use / non-use is performed according to the information on the use / non-use of the spare server 3 in the prediction table. = Detected every 4 hours and when the spare server 3 is used, the information received from the external communication network is transferred to the spare server 3 connected to the internal network. As a result, the load distribution apparatus 1 performs load distribution processing that distributes information processing to the spare server 3 for each time period in which the load is predicted to be large according to the prediction table.

FIG. 5 is a diagram showing a processing flow for generating a errata accumulation table.
FIG. 6 is a diagram showing a errata accumulation table.
Next, the correct / incorrect accumulation table generation processing will be described with reference to FIGS.
First, after the prediction table is generated, information is sequentially stored in the load data time-series storage table in the prediction table generation unit 23 by the same processing as described above. Then, the right / wrong accumulation table generation unit 24 stores the ◯ × information of the load data time-series accumulation table held in the database 26 (that is, information about whether the load exceeds the threshold) and the ◯ × information ( That is, the spare server 3 is used / not used (step S201). Here, it is expected that the spare server 3 is used when the load exceeds the threshold. Further, when the load does not exceed the threshold, it is expected to use the main server 2 without distributing the load. Therefore, it is determined whether the load data time series accumulation table and the prediction table are both ◯ or both X in the same time period of t = 4 hours (step S202). In other words, when the load distribution measure performs load distribution processing using the prediction table as expected), positive information is recorded in the recording area of the correct / incorrect accumulation table corresponding to the time zone (step S203). . Also, in the load data time series accumulation table and the prediction table, when the information of ○ and × are opposite in the same time period of t = 4 hours (that is, the load distribution measure is loaded using the prediction table as expected). When the distributed processing is not performed), erroneous information is recorded in the recording area of the correct / incorrect accumulation table corresponding to the time zone (step S204). Then, the right / wrong accumulation table generation unit 24 sequentially performs right / wrong determination for each time slot of t = 4 hours, similarly to the prediction table. As a result, correct / incorrect information in the prediction table is sequentially stored in the correct / incorrect storage table (step S204). In FIG. 6, the correct / incorrect storage table shows □ as positive information and ■ as incorrect information.

FIG. 7 is a diagram showing an image of processing when the prediction table is corrected from the correct / incorrect accumulation table.
FIG. 8 is a diagram showing a flow of correction processing for a prediction table.
Next, a processing flow for correcting the prediction table from the errata accumulation table will be described with reference to FIGS.
First, the prediction correction unit 25 of the load distribution prediction apparatus 4 starts correct / incorrect determination, and then correct / error information is recorded in each time zone of t = 4 hours over N = 10 days in the correct / incorrect storage table. Then, the correction process of the prediction table is started. In FIG. 7A, when n different from day to day n1 to n10, N = n1 to n10 N = 10 days over the same time zone (for example, 4 hours from 0 o'clock to 4 o'clock) correct / wrong information Is shown. Similarly, (b) is correct / incorrect information in the same time zone (for example, 4 hours from 0 o'clock to 4 o'clock) over N = 10 days from n2 to n11 with the next day after the processing of (a) being performed. Is indicated by a solid frame. Similarly, (c), (d), and (e) are n3 to n12, and n4 to n13, n5 based on the next day after the processing of (b), (c), and (d), respectively. The correct / incorrect information in the same time period (for example, 4 hours from 0 o'clock to 4 o'clock) over N = 10 days from to n14 is indicated by a solid line frame. In this way, the process of correcting the correct / incorrect information in the prediction table in the same time zone is performed while sequentially shifting the sub-period n over N = 10 one day at a time as the day changes.

  Then, in the right / wrong accumulation table, according to any probability of correctness / incorrectness in the same time zone over N = 10 days, the server ○ × (use / non-use) information indicated in the prediction table for the same time zone is displayed. It is determined whether or not to correct. For example, in the same time zone of N = 10 days in the right / wrong accumulation table, if erroneous information is recorded more than a predetermined statistical probability set, use / non-use of the prediction table in the time zone Correct usage. In the same time zone over N = 10 days in the right / wrong accumulation table, if erroneous information is not recorded more than the set statistical predetermined probability, use / non-use of the prediction table in that time zone Do not correct.

  For details of the correction process, first, the prediction correction unit 25 creates a probability-tolerance table in advance (step S301). This is a table as shown in FIG. 9. The probability p that the prediction table is positive, the probability q that the prediction table is incorrect, N that is the prediction table generation correction interval, and the error of the same time zone in N = 10 days The allowable number of occurrences cnt is recorded in association with each other, and a statistical method is used to calculate the number of “errors” (cnt) in the parameter indicated by N, and the relationship between the positive and incorrect probabilities. It is shown for each. Then, the prediction correction unit 25 counts how many pieces of erroneous information are recorded in the same time period for the first N = 10 days (step S302), and sets this value as cnt.

  Next, the prediction table correction unit 25 extracts the range of the probability p associated with the set cnt value and recorded in the probability-tolerance table, and sets the minimum value as pN (step S303). As a result, it is possible to specify by pN how many probabilities "correct" must be when there are the number of cnts set as "false". In addition, a probability pA in which N, which must be “positive”, set in advance by the user as a parameter is recorded in the database 26. The prediction table correction unit 25 reads the probability pA from the database 26 (step S304), and compares pN and pA (step S305). Here, if pN is equal to or greater than pA, the prediction table can be assumed to be more correct than the probability pA expected by the user. On the other hand, if pN is lower than pA, the prediction table is not correct beyond the probability expected by the user, and therefore the accuracy of the prediction table is poor. Therefore, only when pN <pA, the information on whether the spare server 3 is used / not used in the prediction table in the corresponding time zone is reversed (step S306).

  When the prediction table correction unit 25 inverts the information on the use / non-use of the spare server 3 in the prediction table in the correction process of the prediction table, the correction is performed as shown in the process of FIG. All the information in the correct / incorrect storage table used in the process (that is, the error information in the same time period of t = 4 hours over the period of N = 10) is changed to positive (step S307). This avoids using inaccurate information when correcting the prediction table for the same time period of t = 4 hours over a period of N = 10 from the next day after the correction processing is performed. Because. Then, the prediction table correction unit 25 determines whether or not the prediction table correction processing has been performed for every time period of t = 4 hours in one day (step S308). And the prediction table correction | amendment part 25 repeats the correction process of the prediction table of the same time slot | zone of t = 4 hours 6 times (that is, 0-4 o'clock, 4 o'clock-8 o'clock, 8 o'clock-12 o'clock, 12 o'clock- 16:00, 16:00 to 20:00, and 20:00 to 24:00, each t = 4 hours, the correction process was repeated 6 times, and the prediction table was corrected. In this case, a process for instructing the communication processing unit 21 to transmit the corrected prediction table to the load balancer 1 is performed (step S309). As a result, the corrected prediction table is transmitted to the load distribution apparatus 1, and the load distribution apparatus 1 can perform load distribution processing based on a highly accurate prediction table. After one day has passed, a number of erroneous information is recorded in the first t = 4 hours (for example, 0 to 4 o'clock) of N = 10 days starting from the next day. Is counted (step S310), and this value is set to cnt. Then, step S303 to step S310 are repeated.

As described above, the embodiment of the present invention has been described. As a result, the prediction table is changed according to the statistics on the presence / absence of processing load in the load balancer, and the time that is assumed to have a load in advance according to the prediction table. Since load balancing processing is performed on spare servers during the period, load distribution processing can be smoothly transferred during times when the load suddenly increases compared to technologies that perform load balancing processing after a load has occurred. In addition, it is possible to reduce a decrease in information transfer processing in the load distribution apparatus by processing of the load distribution processing itself.
In the case of a technique for determining whether or not to use a spare server depending on whether or not the load threshold is exceeded, when the threshold value is raised or lowered in a short time, processing of using or not using the spare server repeatedly occurs. However, in this process, whether or not to use the spare server is determined in advance in the prediction table, so that it is possible to reduce the degradation of the processing of the load balancer due to repeated use / nonuse of the spare server in a short time. it can.
Moreover, since the prediction table is modified according to the prediction table, even if a change in the time zone with or without load occurs with the passage of time, the change can be dealt with.

  Note that the above-described load distribution device, main server 2, spare server 3, and load distribution prediction device 4 have a computer system therein. The process described above is stored in a computer-readable recording medium in the form of a program, and the above process is performed by the computer reading and executing this program. Here, the computer-readable recording medium means a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, or the like. Alternatively, the computer program may be distributed to the computer via a communication line, and the computer that has received the distribution may execute the program.

  The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

It is a block diagram which shows the structure of the load distribution control system by one Embodiment of this invention. It is a functional block diagram which shows the function structure of the load distribution apparatus and load distribution prediction apparatus by one Embodiment of this invention. It is a figure which shows the processing flow until the load distribution prediction apparatus by one Embodiment of this invention produces | generates a prediction table. It is a figure which shows the processing image at the time of producing | generating a prediction table from the load data time series storage table by one Embodiment of this invention. It is a figure which shows the processing flow of the right / wrong accumulation table production | generation by one Embodiment of this invention. It is a figure which shows the correct / incorrect accumulation table by one Embodiment of this invention. It is a figure which shows the image of the process at the time of correcting a prediction table from the correct / incorrect accumulation table by one Embodiment of this invention. It is a figure which shows the flow of the correction process of the prediction table by one Embodiment of this invention. It is a figure which shows the probability-tolerance table by one Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Load distribution apparatus 2 ... Main server 3 ... Spare server 4 ... Load distribution prediction apparatus

Claims (9)

Whether or not the load has exceeded a preset load threshold, which is information on whether or not to use a spare server that is used as a load reduction purpose only during load distribution processing. Load information receiving means for receiving the load presence / absence information shown at any time;
Based on the information on the presence / absence of the load for each small period obtained by dividing the predetermined period at regular intervals, and the probability of the presence / absence of the load for each of the small periods over a plurality of the predetermined periods, the predetermined period and the small period A prediction table generation means for generating, as a prediction table, information on the use / non-use of the spare server in each future sub-period divided into periods,
Based on the load presence / absence information received after the generation of the prediction table, information on expected use / non-use of the spare server for each small period is detected, and the spare server for each small period in the prediction table is detected. A correct / incorrect accumulation table indicating whether the information matches usage / non-use information, and indicates that the prediction table is positive if the determination matches, and that the prediction table is incorrect if the determination does not match A right / wrong accumulation table generating means for generating the sub-period for the predetermined period;
Prediction that corrects the use / non-use information of the preliminary server shown in the prediction table based on the probability of correctness of the prediction table for each small period over a plurality of the predetermined periods in the correct / incorrect storage table Table correction means;
A load distribution prediction apparatus comprising: The load presence / absence information is determined to use the spare server when the load presence / absence information indicates that there is a load, and when the load presence / absence information indicates that there is no load, the spare server is determined not to be used. The load distribution prediction apparatus according to claim 1, wherein the load distribution prediction apparatus is an information to be performed.   The load distribution prediction apparatus according to claim 1, wherein the load threshold is at least one or more of a communication data amount, a CPU load factor, and a memory area usage rate.   The load distribution prediction apparatus according to any one of claims 1 to 3, wherein the predetermined period is one week and the small period is one day. A prediction table transmitting means for transmitting the prediction table to a load balancer that distributes processing to the spare server based on use / non-use indicated by the prediction table;
The load distribution prediction apparatus according to any one of claims 1 to 4, further comprising: A load distribution control method in a load distribution prediction apparatus,
Load information receiving means of the load distribution prediction device is information on whether or not there is a load used for determining whether or not to use a spare server used only for load reduction during load distribution processing, and is set in advance. The load presence / absence information indicating whether or not the load exceeds the load threshold is received at any time,
The prediction table generation means of the load distribution prediction device is any one of the information on the presence / absence of the load for each small period obtained by dividing the predetermined period at a constant interval, and the presence / absence of the load for each of the small periods over a plurality of the predetermined periods. Based on the probability, the information on the use / non-use of the spare server in each future small period divided into the same period as the predetermined period and the small period is generated as a prediction table,
The correct / incorrect accumulation table generation means of the load distribution prediction device detects information on the use / non-use of the spare server expected for each small period based on the load presence / absence information received after the generation of the prediction table, It is determined whether or not the spare server usage / non-use information for each small period in the prediction table matches, and if the determination matches, the prediction table is positive, and the determination does not match A correct / incorrect accumulation table indicating that the prediction table is erroneous is generated over the predetermined period for each small period,
The prediction table correction means of the load distribution prediction apparatus is configured to provide a preliminary preliminary table indicated by the prediction table based on a probability of correctness of the prediction table for each of the small periods over the predetermined period in the correct / incorrect storage table. A load balancing control method characterized by correcting server use / non-use information. A program to be executed by a computer of a load distribution prediction apparatus,
Whether or not the load has exceeded a preset load threshold, which is information on whether or not to use a spare server that is used as a load reduction purpose only during load distribution processing. A process of receiving information on the presence / absence of the load indicated at any time;
Based on the information on the presence / absence of the load for each small period obtained by dividing the predetermined period at regular intervals, and the probability of the presence / absence of the load for each of the small periods over a plurality of the predetermined periods, the predetermined period and the small period A process for generating, as a prediction table, information on the use / non-use of the spare server in each future sub-period divided into periods,
Based on the load presence / absence information received after the generation of the prediction table, information on expected use / non-use of the spare server for each small period is detected, and the spare server for each small period in the prediction table is detected. A correct / incorrect accumulation table indicating whether the information matches usage / non-use information, and indicates that the prediction table is positive if the determination matches, and that the prediction table is incorrect if the determination does not match Generating for each of the small periods over the predetermined period;
Processing for correcting use / non-use information of the preliminary server shown in the prediction table based on the probability of correctness of the prediction table for each small period over a plurality of the predetermined periods in the correct / incorrect storage table When,
A program that causes a computer to execute. A load balancing control system comprising a load balancing prediction device and a load balancing device,
The load distribution prediction device
Whether or not the load has exceeded a preset load threshold, which is information on whether or not to use a spare server that is used as a load reduction purpose only during load distribution processing. Load information receiving means for receiving, as needed, information from the load balancer indicating whether or not the load is present;
Based on the information on the presence / absence of the load for each small period obtained by dividing the predetermined period at regular intervals, and the probability of the presence / absence of the load for each of the small periods over a plurality of the predetermined periods, the predetermined period and the small period A prediction table generation means for generating, as a prediction table, information on the use / non-use of the spare server in each future sub-period divided into periods,
A prediction table transmitting means for transmitting the prediction table to a load balancer that distributes processing to the spare server based on use / non-use indicated by the prediction table;
Based on the load presence / absence information received from the load balancer after generation of the prediction table, information on expected use / non-use of the spare server for each small period is detected, and for each small period in the prediction table It is determined whether the information matches the use / non-use information of the spare server, and if the determination matches, the prediction table is positive, and if the determination does not match, the prediction table is incorrect A right / wrong accumulation table generating means for generating a right / wrong accumulation table indicating the sub-period for the predetermined period;
Prediction that corrects the use / non-use information of the preliminary server shown in the prediction table based on the probability of correctness of the prediction table for each small period over a plurality of the predetermined periods in the correct / incorrect storage table Table correction means;
With
The prediction table transmission unit transmits the corrected prediction table to the load distribution apparatus. A load balancing control method in a load balancing control system comprising a load balancing prediction device and a load balancing device,
Load information receiving means of the load distribution prediction device is information on whether or not there is a load used for determining whether or not to use a spare server used only for load reduction during load distribution processing, and is set in advance. The load presence / absence information indicating whether or not the load exceeds the load threshold is received at any time,
The prediction table generation means of the load distribution prediction device is any one of the information on the presence / absence of the load for each small period obtained by dividing the predetermined period at a constant interval, and the presence / absence of the load for each of the small periods over a plurality of the predetermined periods. Based on the probability, the information on the use / non-use of the spare server in each future small period divided into the same period as the predetermined period and the small period is generated as a prediction table,
The prediction table transmission means of the load distribution prediction device transmits the prediction table to a load distribution device that distributes processing to the spare server based on use / non-use indicated by the prediction table,
The correct / incorrect accumulation table generation means of the load distribution prediction device detects information on the use / non-use of the spare server expected for each small period based on the load presence / absence information received after the generation of the prediction table, It is determined whether or not the spare server usage / non-use information for each small period in the prediction table matches, and if the determination matches, the prediction table is positive, and the determination does not match A correct / incorrect accumulation table indicating that the prediction table is erroneous is generated over the predetermined period for each small period,
The prediction table correction means of the load distribution prediction apparatus is configured to provide a preliminary preliminary table indicated by the prediction table based on a probability of correctness of the prediction table for each of the small periods over the predetermined period in the correct / incorrect storage table. Modify server usage / non-use information,
The load distribution control method, wherein the prediction table transmission means of the load distribution prediction device transmits the modified prediction table to the load distribution device.

Images