JP3394437B2 - Processor Decision Method by Load Balancing Multiprocessor Method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for activating a load-balanced multiprocessor, and more particularly to a processor determination method based on a load-balanced multiprocessor method when overloaded.

[0002]

2. Description of the Related Art Conventionally, a processor determination method based on this type of load balancing multiprocessor method is disclosed in, for example, Japanese Patent Laid-Open No.
It is disclosed in Japanese Patent Publication No. 235662. In this conventional system, in a load-distributed multiprocessor system, it is necessary to determine the processor that executes the process by some method, and it is required that each processor be evenly requested to perform the process. To be However, the contents of processing required for each processor are not the same, and the load on each processor differs. As a result, the load on a particular processor may be high. There is a case where the processor is deadlocked due to a load that exceeds the processing capability of the processor that causes the processor with the higher load to execute further processing. In order to prevent this, the processing is not activated for the processor under a certain load, but the processing is controlled to be transferred to another processor.

[0003]

As described above, in the method of the prior art, when a processor having a load higher than a certain level is detected, the corresponding processor is not started for a certain period of time, so that the processing request is not received. The processing power of the processor was rapidly recovering. Furthermore, the load on each processor is not always detected,
It was limited to periodic detection. As a result of the processing of both of these, the processor whose load once rises and is not started up continues from the high load state to the normal load state until it can see the load state even if it reaches a level at which processing can be executed. However, there is a problem that a phenomenon occurs in which a request for processing is not received even though processing is actually possible.

Further, as the processing load on the system increases,
The number of processors that exceed the limit of processable capacity and cannot be activated increases, and the number of processors that can accept a processing request decreases. If the load further increases after this, the processing requests concentrate on a small number of processors, and the processor suddenly falls into an unprocessable state, and no processor accepts the processing. Further, since this phenomenon can be detected only by periodically detecting the load, there is a problem that if the detection timing is delayed, all the processors may be deadlocked.

An object of the present invention is to provide effective use of load-balanced processors.

A further object of the present invention is to prevent a processor deadlock caused by a processing request exceeding the processing limit of the processor.

[0007]

In order to achieve the above object, the present invention provides a plurality of load-balanced first processors, and a second processor for selecting one of the load-balanced first processors. A processor determination method based on a load balancing multiprocessor method including
The processor of (1) reads out and compares the CPU usage rate of the first processor and a preset congestion determination number at regular intervals, and the CPU usage rate of the comparison means is equal to or greater than the congestion determination number. For example, if the CPU utilization rate by the comparing means is less than the determination number, the first processor reduces the number of process executions that can be processed by the task of the first processor in a stepwise manner. a return means for stepwise return until the said processing executable number of tasks on the number that can be maximum processing in normal load conditions, when requested task processing, the task processing
Of the process executable number of the first processor
A determining means for determining whether the remaining number is within an allowable range , and the remaining number of the process executable number is allowable by the determining means.
And determining means for determining the first processor as a processor for performing the task processing when it is determined to be within the range .

Further, the processing can be executed by the judging means.
A discarding means for discarding the request for the task processing when the number of the first processors in normal operation falls below a predetermined number when it is determined that the remaining number of active numbers is not within the allowable range; It is characterized by having.

[0009]

Further, the second processor is the first processor.
It said congestion determining the number of each processor, and the processing executable <br/> number, is characterized in that it comprises a management table including a number that can be the maximum processing.

[0011]

1 is a block diagram of an embodiment of the present invention.

Referring to FIG. 1, the decision processor 10 operating under program control is connected to a plurality of processing processors 20, 30, 40 operating under program control by a bus 80.

The decision processor 10 is a processor for selecting and activating the load-balanced processing processors 20, 30, 40. The processing processors 20, 30 and 40 are processors that perform actual processing, and receive a start instruction from the above-mentioned decision processor 10. A plurality of processing processors 20, 30, 40 are described here to show that the load is distributed, and the processing contents of all the processors are the same.

The decision processor 10 includes a decision unit 101 that decides a pusher to execute a process, a state collection unit 102 that collects the CPU usage rate of each processing processor that is used as a decision material when making a decision, and a state collection unit. And a processing regulation level table 103 for determining increase / decrease in the number of processing processor activations when the CPU usage rate of the processor collected in step 102 exceeds an allowable range. The respective processing processors 20, 30, 40 are processing processors 20, 30, which are collected by the state collecting unit 102 in the decision processor 10.
State determination units 202 and 30 for determining the CPU usage rate of 40
2, 402, and processing execution units 201, 301, 401 that perform processing for the original purpose of the system.

The format of the processing regulation level table 103 has a variable "n" indicating a level variable of a congestion state (described later) for each processing processor and a variable for gradually returning from the congestion state to a state where maximum processing is possible. "M" indicates "maximum number of processing executions" indicating the maximum number that can be processed by the processing processor
And "the number of executable processes" indicating the number of executions that can be processed by the processing processor, and "the number of congestion determinations" indicating the usage rate of the CPU that is determined to be in the congestion state. "N" is set to a value "0" at system startup. The "maximum number of process executions", "number of process executions", "number of congestion judgments", and "m" are set to initial values according to the capability of each processor when the system is started up.

Next, the block diagram of FIG. 1 and FIG.
The operation of an example of the embodiment will be described with reference to the flowchart for determining the processor.

In FIG. 1, processing processors 20, 3
The processors 0 and 40 are load-balanced processors and are installed to perform the same processing. In this case, it is necessary to determine which processing processor should execute the processing when the processing request is issued, and the processor determining unit 101 makes this determination. The processor determining unit 101 basically activates each processing processor at the same rate, but the CPU usage rate of a specific processing processor rises for some reason (hereinafter, referred to as a congestion state) and cannot be activated any more. It is necessary to exclude from the selection target.
At this time, the state collection unit 102 periodically monitors the CPU usage rate of the processing processor by the state determination units 202, 302, 40.
It is realized by inquiring to 2.

The state collection unit 102 periodically grasps the congestion state of all the processing processors, and when it is judged as the congestion state, it judges the regulation level from the previous state of the relevant processor and stores it in the processing regulation level table 103. The collection of the CPU usage rate at this time is as follows.

In each processing processor, a processing request generated from time to time to the CPU is interrupted every cycle (for example, 8 ms), and the driver (control program) waits in a queue (task waiting for processing). Take out from. The ratio of 8 ms of the time required to complete all the individual processing requests is the instantaneous CPU usage rate. The state determination units 202, 302, and 402 use the 8 ms
The CPU usage rate of the unit is measured and added as needed. At this time, the state collecting unit 10 is periodic (for example, every 20 seconds).
According to the request from 2, the average for 20 seconds until then is calculated (ratio to 20 seconds), and the collected value is transferred to the state collection unit 102 as the CPU usage rate.

FIG. 2 shows a processing flow when the state collection unit 102 of the decision processor 10 operates.

In FIG. 2, when an interrupt is taken every 20 ms, the state collection unit 102 selects the first processing processor. Further, the CPU usage rate is read from the state determination unit of the processing processor, the “congestion determination number” in the processing regulation level table 103 is read, and the CPU usage rate and the “congestion determination number” are read.
Is compared (steps S1 and S2).

If the CPU usage rate is equal to or higher than the "congestion determination number", it is determined to be in a congestion state and "process executable number" (y).
Is read from the processing regulation level table 103, and it is checked whether the value is “0”. If the value is not "0", the calculation in step S5 is performed. Note that the expression y = z * (0.
6-0.1 * n + 0.2 * m) is an example of an expression for deciding the number of executions of a task that can be processed step by step according to the congestion state, and does not necessarily have to be this expression. Z in the formula is “maximum number of processing executions” in the processing regulation level table 103.
Indicates. In the processing of the example, the "process executable number" is first set to 50% of the "process execution maximum number", and then gradually increased to 10%.
It is set to be lowered by steps (steps S4 and S5).

If the CPU usage rate is less than the "congestion judgment number", it is judged to be a normal load state, "m" of the processing regulation level table 103 is read, and it is checked whether the value is "0". If m = 0, a process like step S7 is performed.
The formula y = z * (0.6-0.1 * n + 0.2 * m)
Is an example of an expression for stepwise determining the number of executions that can be processed by a task in order to return the congestion state to the maximum processing state, and it is not always necessary to use this expression. In the process of the example, in order to return to the maximum processable state, the "process executable number" is increased by 20% of the "process executable maximum number" (steps S6 and S7). Note that the maximum processable state means that the “process executable number” in the process restriction level table 103 has reached the value of the “process executable maximum number”, that is, the processing of the tasks of each processing processor is processed up to the maximum number. The state that can be done.

If m = 0 is not satisfied, the maximum number of process executions and the maximum number of process executions (confirmation of complete restoration) are compared, and if they match, the process proceeds to step S7 (steps S8, S8).
9).

The designation of the processing processor is performed in rotation (processing processor 20 → 30 → 40) in order, but if the next processor still exists, step S
Fly to 2 If not, the process ends (step S10).

It should be noted that one rotation process is performed every 20 ms.

FIG. 4 shows an example of a method of managing the restriction level of processing requests to the processing processor 20 in the congestion state. The vertical axis represents the level that regulates the processing request per unit time. The horizontal axis shows the passage of time. At this time, the "maximum number of processing executions" is set to 100, and the "congestion determination number" is set to 50% at system startup.

The state collection unit 102 periodically collects the CPU usage rate from the state determination units 202, 302, 402 of each processor.

At this time, when the state collection unit 102 which collects the CPU usage rate from the state determination unit 202 of the processing processor 20 detects the CPU usage rate that exceeds the reference value, it judges that the processor is in a congestion state and the number of process executable Enforce regulations. The restriction on the number of process executables is to reduce the number of process executables significantly (down by 50% according to step S5 in FIG. 2) when the CPU usage rate up to that time is normal, and once the congestion state occurs. If the processing processor has not recovered from the congestion state even in the next collection cycle, the number of processes that can be executed is narrowed (10% is reduced according to step S5 of FIG. 2).
Reduce to. This small reduction of the process executable number is continued until the processing processor in the congested state is in the maximum processable state (according to step S7 of FIG. 2, it is recovered by 20% in the normal load state). ). In this way, the processor determining unit 101 determines the processor to execute the process when the process restriction level table 103 is used.
Is executed to determine whether the process is executable.

Next, FIG. 3 shows an operation example in which the processor determining unit 101 determines a processing processor when a task processing request is made.

In FIG. 3, processing processors are sequentially selected as processing execution candidate processors (step A).
1). It is judged from the regulation level table 103 whether the processing processor selected as a candidate is in the congestion state (step A
3). If the processing processor selected here is not in the congestion state, this processing processor is determined as the processing execution target processor. (Step A3). If it is during congestion, the remaining number of process executables is determined (step 4), and if it is within the process request practicable range (within the process processor allowable range), it is selected as the process execution processor (step A).
5). Further, in that case, the process executable number of the regulation level data is subtracted by one. When the process executable number is 0, another process processor is selected as a process execution candidate processor (step A6), and the process returns to step A2 to determine whether the process is possible.

As described above, since the limit of the number of process executables of each processor is determined based on the CPU usage rate, it is possible to operate without wasting the processing capacity of the processor.

In the above-described example of the embodiment, when a plurality of processors are congested and are excluded from the processing targets, the processing is rapidly concentrated on the remaining few processors, and the concentrated processors are concentrated. However, there is a possibility that the worst processing processor may be in a deadlock state because the required processing amount exceeds the processing capacity at once by eliminating the time margin for obtaining the effects of the invention described above. Therefore, the number of processors in congestion and the number of processors in normal operation are managed, and when the number of processors in normal operation falls below a certain number, the requested processing is not executed. At this time, the “number of processors that can be processed” may be set in the processing regulation control level table 103. This "processable processor number" is initialized when the system is started up, and is decreased by "1" when there is a processor that cannot be processed ("process executable number" = 0). FIG. 5 shows an operation example when the decision processor 10 in this case decides the processing processor.

In FIG. 5, steps A1 to A7 are the same as those described in FIG. In step 5 of FIG. 5, if the allowable number of receptions is out of the range, the value of “the number of processable processors” is checked, and if the value is “0”, the processing request is discarded (steps A11, A1).
2). When the value of “the number of processable processors” is larger than “0”, the next processing processor is set as the processing candidate processor (step A6).

As described above, by limiting the number of processors that can be processed according to the load condition, it is possible to withstand a sudden load and to operate the system stably.

[0036]

As described above, since the number of processes that can be executed by the processor is limited according to the load state, even if one of the load-balanced processors is in the congestion state, it is applicable. There is an effect that it can be operated without wasting the processing capacity of the processor.

Further, since the number of processors that can be processed is limited according to the load state, even if a sudden load increase occurs in a system in which the load is distributed among the processors, the system can be stabilized. There is an effect that the state can be maintained.

[Brief description of drawings]

FIG. 1 is a diagram showing a block configuration of a system in which processors are load-balanced according to an example of an embodiment of the present invention.

FIG. 2 is a flowchart relating to setting of a limit number of tasks that can be processed in the status collecting unit of each processor of FIG.

FIG. 3 is a flowchart regarding determination of a processor that can process a task in a state collection unit of the decision processor 10 of FIG.

FIG. 4 is a diagram showing a stepwise restriction situation of processing requests to a specific processing processor in an example of an embodiment of the present invention.

FIG. 5 is a flowchart relating to determination of load-balanced processor selection in an example of an embodiment of the present invention.

[Explanation of symbols]

10 decision processor 20, 30, 40 Processor 101 processor determination unit 102 status collection unit 103 Regulation level table 201, 301, 401 Process execution unit 202, 302, 402 status determination unit

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 15/16-15/177 G06F 9/46-9/54

Claims (3)

(57) [Claims] 1. A processor determination method according to a load balancing multiprocessor method, comprising: a plurality of load-balanced first processors; and a second processor that selects one of the load-balanced first processors, The second
The processor of (1) reads out and compares the CPU usage rate of the first processor and a preset congestion determination number at regular intervals, and the CPU usage rate of the comparison means is equal to or greater than the congestion determination number. For example, if the CPU utilization rate by the comparing means is less than the determination number, the first processor reduces the number of process executions that can be processed by the task of the first processor in a stepwise manner. Means for gradually returning the process executable number of the tasks until reaching the maximum processable number under the normal load condition, and the first processor for performing the task processing when the task processing is requested. A determining unit that determines whether the remaining number of process executable numbers is within an allowable range; and the determining unit determines that the remaining number of process executable numbers is within an allowable range. Processor decision method according to the load distribution multiprocessor system, characterized in that the processor for performing the task processing when having determining means for determining said first processor. 2. When the number of the first processors in normal operation falls below a predetermined number when the judgment means judges that the remaining number of the process executable numbers is not within an allowable range. 2. The processor determination method according to claim 1, further comprising: a discarding unit that discards the task processing request. 3. The second processor comprises a management table including the congestion determination number, the process executable number, and the maximum processable number for each of the first processors. The processor determination method according to the load balancing multiprocessor method according to claim 1 .