JP3964896B2 - Resource allocation apparatus and resource allocation method - Google Patents

Description

The present invention relates to an information processing system in which a server load balancer and a plurality of servers are incorporated, and in particular, a resource for controlling the operation (allocation number) of a server (resource) according to a load state in the information processing system. The present invention relates to an allocation device and a resource allocation method .

  For example, as shown in FIG. 5, in an information processing system 2 adopted for a website published on a corporate network (Internet) 1, the same IP address (URL) from a number of clients (customer consumers) 3 is used. In order to respond in a short time to an access request (processing request) that designates (), one server load balancer (load balancer) 4 and a plurality of servers 5 are incorporated. Each server 5 incorporates a Web processing unit 6 for realizing a processing request such as a homepage browsing request from the client (customer consumer) 3.

  The server load balancer 4 allocates a plurality of access requests (processing requests) specifying the same IP address (URL) input from each client 3 via the network (Internet) 1 to each server 5. The server 5 to which the processing request is assigned executes processing such as homepage transmission corresponding to homepage browsing specified by the processing request in its own Web processing unit 6.

  Each server 5 can simultaneously perform a plurality of processes. If M servers can be processed at the same time, if n servers 5 are incorporated in the information processing system 2, processing requests from (M × n) clients 3 at the same time in the information processing system 2 It is possible to respond.

  In such an information processing system 2, the number of processing requests input to the server load distribution device 4 is not constant and varies greatly according to time and day of the week. Accordingly, all the servers 5 are not always in an operating state, and the number of servers E to be operated is determined as follows in consideration of the margin for one unit according to the processing request number A at that time.

E = (A / M) +1
Only the servers 5 corresponding to the number of servers E are set in the operation (operation) state. Then, the server load distribution device 4 allocates each processing request input from each client 3 via the network 1 to each operating server 5.

Patent Document 1 discloses a technique for calculating a load factor of each load unit (server) and transmitting a new load request to a load unit (server) having a small load factor.
Japanese Patent Laid-Open No. 2001-75934

  However, the information processing system 2 that obtains the number E of servers to be operated according to the number A of processing requests at that time by the above-described method still has the following problems to be solved.

  That is, as shown in FIG. 6, in each server 5, in addition to the Web processing unit 6 for realizing the processing request such as the homepage browsing request from the client 3 described above, a mail response request from the client 3, etc. Assume that the mail processing unit 7 is provided to realize the processing request.

  In this case, each server 5 receives two types of processing requests, a Web processing request and a mail processing request. Each server 5 can perform M processes simultaneously. However, the processing burden on the CPU when executing one mail process is much larger than the processing burden on the CPU when executing one Web process, and as a result, the processing time becomes longer.

  Therefore, the processing load of the server 5 when the M processing requests input to the server 5 are all mail processing requests is the server load when the M processing requests input to the server 5 are all Web processing requests. Compared with the processing burden of 5, it becomes much larger. As a result, the entire processing time of the corresponding server 5 becomes longer than that of the other servers 5. In this case, the processing of the Web processing request input to the server 5 with many mail processing requests is also delayed.

  This has a problem that a large difference occurs in the processing time required for each processing request input from each client 3 to the information processing system 2 via the network 1. In addition, a large non-uniform processing load occurs between the servers 5.

The present invention has been made in view of such circumstances, and even if the processing requests input from the outside are a plurality of types of processing requests having different processing loads, the number of processing requests input from the outside is not limited. To provide a resource allocation device and a resource allocation method that can maintain the processing load of each server almost evenly even if it fluctuates in time, and that can normally execute processing specified by a processing request input from the outside without time delay With the goal.

The present invention is a resource allocation apparatus in the information processing system having a plurality of servers, a server load balancer that allocate processing requests of a plurality and plurality of types of input from a client to a server operating state.

In order to solve the above-described problem, the resource allocation device of the present invention includes a correction value storage unit that stores a correction value corresponding to the time required to execute the type of processing for each type of processing request, and a client. The processing request number calculating means for calculating the number of processing requests per unit time for each type inputted, and the processing request number per unit time for each type calculated by the processing request number calculating means are stored in the correction value storage means. It was appropriate as the processing request count correcting means for classification for each of multiplying the correction value for correcting the number of processing requests per unit of each type time, corrected per unit time per corrected type in this processing request number correction means by adding the number of processing requests, a processing request adding means for obtaining the number of corrected total processing required for the input processing request from the client, that can be processed simultaneously in the addition number was corrected total processing requesting server punished Is divided by the maximum processable number indicating the number of requests, a server number calculating means for determining a number of servers necessary to process the input processing request number obtained by adding a margin number to the quotient from the client, the server Server control means for increasing or decreasing the number of servers in the operating state based on the required number of servers calculated by the number calculation means.

  In the resource allocation device in the information processing system configured as described above, the number of processing requests for each type input from the outside is calculated, and the calculated number of processing requests for each type corresponds to the processing time for each type. It is corrected with the correction value. Therefore, the number of processing requests for each type after correction is the number of processing requests corresponding to (proportional to) the time required for processing. Therefore, the required number of servers can be accurately obtained from the added value of the number of processing requests for each type after correction.

  As a result, even if the processing request input from the outside is a plurality of types of processing requests with different processing loads, and the number of processing requests input from the outside fluctuates, the processing load of each server is almost the same. It can be maintained uniformly.

Another invention includes a plurality of servers, a resource allocation method in an information processing system having a server load balancer that allocate the server process requests the operating status of the plurality and plurality of types input from the client.

In order to solve the above problem, the resource allocation method according to the present invention is calculated by a processing request number calculating step for calculating a processing request number per unit time for each type input from a client, and a processing request number calculating step. A processing request number correction step of correcting the processing request number per unit time for each type by multiplying the processing request number per unit time for each type by a correction value corresponding to the processing time of the processing request of the corresponding type ; A processing request addition step for obtaining a corrected total processing request for the processing request input from the client by adding the corrected processing request number per unit time for each type corrected in the processing request number correction step, and adding Divide the corrected total processing request count by the maximum processing count that indicates the number of processing requests that can be processed simultaneously by the server, and add the surplus number to the division value. Server control step of increasing or decreasing the number of servers calculating step of obtaining a number of servers required, the number of servers operating state based on the required number of servers that are calculated by the server number calculating step for processing the processing request inputted from bets And.

  In the resource allocation method in the information processing system configured as described above, it is possible to achieve substantially the same operational effects as the resource allocation device in the information processing system described above.

  The present invention calculates the number of processing requests for each type input from the outside, corrects the calculated number of processing requests for each type with a correction value corresponding to the processing time for each type, and processes the processing request for each type after correction. The required number of servers is obtained from the sum of the numbers.

  Therefore, even if processing requests input from the outside are multiple types of processing requests with different processing loads, and even if the number of processing requests input from outside fluctuates over time, the processing load on each server is reduced. The process can be maintained almost uniformly, and the process specified by the process request input from the outside can be executed normally without time delay.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 is a schematic diagram showing a schematic configuration of an information processing system in which a resource allocation device employing a resource allocation method according to an embodiment of the present invention is incorporated. The same parts as those of the conventional information processing system 2 shown in FIG. 5 and FIG.

  For example, the information processing system 11 employed in a Web site published on the corporate network (Internet) 1 responds to each access request (processing request) from a large number of clients (customer consumers) 3 in a short time. For this purpose, one server load balancer (load balancer) 12, n servers 5, and one resource allocation device 13 are incorporated.

  Each server 5 includes a Web processing unit 6 for realizing a processing request such as a homepage browsing request from the client (customer consumer) 3 and a processing request such as a mail response request from the client (customer consumer) 3. In order to realize the above, two types of processing units with the mail processing unit 7 are incorporated. The CPU processing load, that is, the processing time when the mail processing unit 7 executes one mail process, for example, needs to be twice as long as the processing time when the Web processing unit 6 executes one Web process. To do. Each server 5 can perform M processes simultaneously. Each server 5 is ON / OFF controlled by the resource allocation device 13.

Accordingly, there are two types of processing requests input from each client 3 to the information processing system 11, that is, a web processing request (type W ₁ ) and a mail processing request (type W ₂ ).

  The server load balancer (load balancer) 12 includes a server processing capacity memory 14, a server status memory 15, a server load status memory 16, a processing request reception unit 17, a processing type determination unit 18, and a processing request distribution unit 19. .

In the server processing capacity memory 14, as shown in FIG. 2A, a maximum processable number M (in this embodiment, M = 50) indicating the number of processing requests that can be processed simultaneously in each server 5, The number of independently processable N ₁ and N ₂ (in this embodiment, N ₁ = 50, N ₂ = 25) for each processing type W ₁ and W ₂ is stored. Each of these independently processable numbers N ₁ and N ₂ is the number of processing requests whose processing time ends within the reference time when only the processing of the corresponding types W ₁ and W ₂ is executed in the server 5.

  In the server status memory 15, as shown in FIG. 1-No. The state of whether each of the n servers 5 is operating (ON) or stopped (OFF) at the current time is stored.

The server load status memory 16 stores the number of processing requests X ₁ and X ₂ for each type W ₁ and W ₂ of processing requests being processed at the current time of each of the n servers 5.

The processing request reception unit 17 receives a plurality of access requests (processing requests) input from each client 3 via the network 1 and sends them to the next processing type determination unit 18. The process type determination unit 18 determines the types W ₁ and W ₂ of each input process request, attaches the types W ₁ and W ₂ to each process request, and sends them to the next process request distribution unit 19. .

The processing request distribution unit 19 assigns the processing requests of various types W ₁ and W ₂ that are sequentially input to the active servers 5 set in the server state memory 15 in order, for example, in a round robin manner. (Send it). Then, the allocation (transmission) result is registered (written) in the server load status memory 16.

If the processing request number S = X ₁ + 2X ₂ already allocated by the allocation destination server 5 in the server load status memory 16 exceeds the maximum processable number M, new processing for this server 5 is performed. A new processing request is allocated (transmitted) to the next server 5 without allocating (transmitting) the request. Further, the processing request distribution unit 19 deletes the processing request whose information processing has been completed by the server 5 from the server load status memory 16.

FIG. 3 is a block diagram showing a schematic configuration of the resource allocation device 13.
The processing request number calculation unit 20 calculates the number of processing requests c ₁ per unit time of the Web processing request (type W ₁ ) output from the processing type determination unit 18 of the server load distribution device 12 and determines the next processing request. It is sent to the number correction unit 21. On the other hand, the processing request number calculation unit 22 calculates the processing request number c ₂ per unit time of the mail processing request (type W ₂ ) output from the processing type determination unit 18 of the server load balancer 12, and This is sent to the processing request number correction unit 23.

In the correction value memory 24, correction values B ₁ and B ₂ corresponding to the time required to execute the processing of each processing request are stored for each of the above-described types W ₁ and W ₂ of processing requests. In this embodiment, since the processing time of the mail processing request (type W ₂ ) is _twice the processing time of the Web processing request (type W ₁ ), the correction values are B ₁ = 1 and B ₂ = 2. Is set to

The processing request number correction unit 21 corrects the processing request number of the Web processing request (type W ₁ ) by multiplying the processing request number c ₁ by the correction value B ₁ = 1, and the corrected processing request number c _1A. Is sent to the effective server number calculation unit 25. Similarly, the processing request number correction unit 23 multiplies the processing request number c ₂ by the correction value B ₂ = 2 to correct the processing request number of the mail processing request (type W ₂ ), and the corrected processing request. The number c _2A is sent to the effective server number calculation unit 25.

c _1A = c ₁ × B ₁ = c ₁ × 1 = c ₁
c _2A = c ₂ × B ₂ = c ₂ × 2 = 2c ₂
Valid server number calculation formula memory 26, calculates the maximum processable number described above M, and the processing request number c _1A after correction, using c _2A, the server number D ₁ required in anticipation of one minute margin The calculation formula to be stored is stored.

D ₁ = [(c _1A + c _2A ) / M] +1
The effective server number calculation unit 25 calculates the server number D ₁ necessary for realizing the corrected processing request numbers c _1A and c _2A for the processing request using the above-described calculation formula, and sends the calculation request to the comparison unit 27. Send it out.

The server state reading unit 28 reads the number of active servers D ₀ in the server state memory 15 of the server load distribution device 12 and sends it to the comparison unit 27. Furthermore, the server status reading unit 28 reads the operating (ON) or stopped (OFF) status of each server 5 at the current time point and sends it to the status change server determining unit 30.

The comparison unit 27 compares the calculated necessary number of servers D ₁ with the number of servers D ₀ currently operating, and activates the server change number calculation unit 29 only when there is a mismatch. The server change number calculation unit 29 calculates the server change number (D ₁ -D ₀ ) indicated by the difference between the required server number D ₁ and the active server number D _0, and sends it to the state change server determination unit 30. Send it out.

  The state change server determination unit 30 determines, for example, the servers 5 to be newly started or stopped in ascending order of the server numbers based on the operating (ON) or stopped (OFF) state of each server 5 at the current time point. The server status change unit 31 changes the operating (ON) or stopped (OFF) status of the server 5 in the server status memory 15 of the server load distribution device 12.

  Further, the state change server determination unit 30 actually starts or stops the determined server 5 to be started or stopped via the server start / stop control unit 32.

FIG. 4 is a flowchart showing the overall operation of the resource allocation device 13 configured as described above. When the specified time T _S determined by the control cycle elapses (step S1), per unit time of the Web processing request (type W ₁ ) and the mail processing request (type W ₂ ) input from each client 3 to the information processing system 11 The processing request numbers c ₁ and c ₂ are read (S2). In order to make the read processing request counts c ₁ and c ₂ uniform conditions, the correction values B ₁ and B ₂ are multiplied and corrected (S3).

From the corrected number of processing requests c _1A and c _2A , the necessary number of servers D ₁ in anticipation of a margin for _one is calculated using the calculation formula D ₁ = [(c _1A + c _2A ) / M] +1 (S4 ). The number of servers D ₀ currently operating is read (S5). Compared calculated and server number D ₁ required a number of servers D ₀ in operation at the present time (S6), if there is a match (S7), since it is not necessary to change the number of servers during operation, doing anything Instead, the process returns to S1 and waits until the next control cycle.

Calculated with the server number D ₁ and required number of servers D ₀ in operation at the present time is not matched only (S7), and calculates the server change number (D ₁ over D ₀₎ (S8). Then, the server 5 to be started or stopped is determined (S9). The state of operation (ON) or stop (OFF) of the server 5 in the server state memory 15 of the server load balancer 12 is changed (S10). Finally, the determined server 5 to be started or stopped is actually started or stopped via the server start / stop control unit 32 (S11). When the above process ends, the process returns to S1 and waits for the next control cycle.

In the resource allocation device and the resource allocation method in the information processing system configured as described above, a Web processing request (type W ₁ ) and a mail processing request (type W ₁ ) input from each client 3 to the information processing system 11 via the network 1 ( The number of processing requests c ₁ and c ₂ per unit time with the type W ₂ ) is calculated, and the calculated number of processing requests c ₁ and c ₂ for each type W ₁ and W ₂ corresponds to the processing time for each type. The correction values B ₁ and B ₂ are corrected. Accordingly, the number of processing requests c _A1 and c _{A2 for} each type after correction is the number of processing requests corresponding to (proportional to) the time required for processing. Therefore, the required number of servers D ₁ = [(c _1A + c _2A ) / M] +1 can be accurately obtained from the added value (c _A1 + c _A2 ) of the number of processing requests for each type after correction.

As a result, processing requests per unit time of Web processing requests (type W ₁ ) and mail processing requests (type W ₂ ) with different processing loads input from each client 3 to the information processing system 11 via the network 1. Even if the ratio between the numbers c ₁ and c ₂ fluctuates with time, the processing load on each server 5 can be maintained substantially uniformly. Therefore, the processing specified by the processing request input from each client 3 to the information processing system 11 via the network 1 can be normally executed without a time delay.

In addition, this invention is not limited to embodiment mentioned above. In the embodiment, the case where each server 5 is provided with two types of processing units corresponding to the processing requests of the two types W ₁ and W ₂ of the Web processing unit 6 and the mail processing unit 7 is shown. The number of types of processing requests and the number of processing copies of each server 5 are not limited to two.

  Also, each processing unit provided in each server 5 is not limited to a Web processing unit or a mail processing unit.

The schematic diagram which shows schematic structure of the information processing system incorporating the resource allocation apparatus as which the resource allocation method of one Embodiment of this invention was employ | adopted. The figure which shows the memory content of the server processing capacity memory and the server status memory provided in the information processing system The block diagram which shows schematic structure of the resource allocation apparatus of the embodiment A flowchart showing the operation of the resource allocation device of the same embodiment Schematic diagram showing the schematic configuration of a conventional information processing system Similarly, a schematic diagram showing a schematic configuration of a conventional information processing system

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Network, 3 ... Client, 5 ... Server, 6 ... Web processing part, 7 ... Mail processing part, 11 ... Information processing system, 12 ... Server load distribution apparatus, 13 ... Resource allocation apparatus, 14 ... Server processing capacity memory, DESCRIPTION OF SYMBOLS 15 ... Server state memory, 16 ... Server load condition memory, 17 ... Processing request receiving part, 18 ... Processing type determination part, 19 ... Processing request distribution part, 20, 22 ... Processing request number calculation part, 21.23 ... Processing request Number correction unit, 24 ... Correction value memory, 25 ... Effective server number calculation unit, 26 ... Effective server number calculation formula memory, 27 ... Comparison unit, 28 ... Server state reading unit, 29 ... Server change number calculation unit, 30 ... State Change server determination unit, 31 ... server state change unit, 32 ... server start / stop control unit

Claims (2)

A plurality of servers, keep the information processing system having a server load balancer that allocate processing requests of a plurality and plurality of types of input from a client to a server operating state,
Correction value storage means for storing a correction value corresponding to the time required to execute the type of processing for each type of processing request;
Processing request number calculating means for calculating the number of processing requests per unit time for each type input from the client ;
The number of processing requests per unit time calculated by the processing request number calculating means is multiplied by the correction value for each corresponding type stored in the correction value storage means by the number of processing requests per unit time calculated by the processing request number calculating means. A processing request number correcting means for correcting;
Processing request addition means for adding the corrected number of processing requests per unit time for each type corrected by the processing request number correction means to obtain a corrected total processing request number for the processing requests input from the client ;
The number of corrected total processing requests added is divided by the maximum number of processing requests indicating the number of processing requests that can be processed simultaneously by the server, and the number obtained by adding the surplus number to the divided value is the processing request input from the client. Server number calculating means for obtaining the number of servers necessary for processing
A resource allocation device comprising server control means for increasing or decreasing the number of servers in an operating state based on the required number of servers calculated by the server number calculation means. A plurality of servers, in the information processing system having a server load balancer that allocate processing requests of a plurality and plurality of types of input from a client to a server operating state,
A processing request number calculating step of calculating the number of processing requests per unit time for each type input from the client ;
The number of processing requests per unit time for each type calculated in the processing request number calculation step is multiplied by a correction value corresponding to the processing time of the processing request for the corresponding type to correct the number of processing requests per unit time for each type. A processing request number correction step to be performed;
A processing request addition step for adding the corrected processing request count per unit time for each type corrected in the processing request count correction step to obtain a corrected total processing request count for the processing request input from the client ;
The number of corrected total processing requests added is divided by the maximum number of processing requests indicating the number of processing requests that can be processed simultaneously by the server, and the number obtained by adding the surplus number to the divided value is the processing request input from the client. Calculating the number of servers as the number of servers required to process
A resource allocation method comprising: a server control step for increasing / decreasing the number of servers in an operating state based on the required number of servers calculated in the server number calculation step.

Images