JP2517852B2 - How to change process status - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process state changing system for changing the state of all processes having a specific ID (unique name) in a multiprocessor system which performs interprocessor processing by asynchronous message communication.

[0002]

2. Description of the Related Art In a multiprocessor system,
In order to speed up processing, a plurality of processes having a specific ID are often distributed to a plurality of processors and executed. In such a case, the same ID may be generated due to external commands from the console or physical restrictions of hardware.
It also causes the state of all processes having

As a conventional process state changing method, there is one described in the following document A. To prepare for the description of the conventional process state changing method, first, the “calculation model” and the “end detection method (WTC method)” used by the conventional process state changing method will be described in this order, and then the conventional process state changing method will be described. The "state change method" will be described.

Reference A; Hideo Kawai, Akihiko Nakase, Akira Imai, Atsuhiro Goto, Kazuaki Muzawa, "Execution control method of KL1 in parallel inference machine-problems and countermeasures of distributed goal management", IPSJ 74 Technical Committee on Computer Architecture 82-2,199
0.4.20.

(1) "Computational model" (1-1) The assumed process model is as follows. Note that FIG. 2 is an explanatory diagram of an assumed process model.
In FIG. 2 and each drawing described later, a control process is shown by a rectangular double solid line having a process pool ID (hereinafter, simply abbreviated as ID) inside, and a child process is shown by a circular solid line having a process ID inside. The processor is shown as a solid line with the process inside.

(1-1-1) One process pool consists of one control process and at least one child process. (1-1-2) There can be a plurality of process pools in the system, and each process pool has a unique ID in the system. (1-1-3) Child process has state. (1-1-4) The child process can terminate at any time. (1-1-5) A child process can always create a child process having the same ID. Also, a process pool with a new ID can be created at any time. (1-1-6) When all the child processes having a certain ID end, the process pool of that ID ends. (1-1-7) Changing the state of the process pool with a certain ID means that the control process changes the state of all child processes having that ID.

(1-2) The above process pool is executed in the multiprocessor system in the following format.
Note that FIG. 3 is an explanatory diagram of the relationship between the processors and the process pools, and includes three processors PEi, PEj, and P.
Ek and two process pools with IDs A and B are shown. Here, the processor is C in terms of hardware.
The basic configuration is a PU, a memory, and a communication configuration with another processor.

(1-2-1) Among processes existing in one processor, those having the same ID are called "subpool". (1-2-2) The subpool ends when all the child processes that compose it have finished. (1-2-3) The processor can detect the end of the subpool in its own processor. In addition, it is possible to change the states of all child processes belonging to the subpool having the specific ID in the own processor. Such state change processing may be realized by any of software, hardware, and firmware. (1-2-4) A processor can send a child process existing in its own processor to another processor. Sending a child process is specifically sending a message that carries the child process. (1-2-5) The control process and each processor (in this specification, the end processing unit of the sub-pool and the state changing unit of the sub-pool are also simply called processors) can communicate with each other by asynchronous messages. For example, a processor that detects the end of a subpool can send a message to that effect to the control process. Further, the control process can send a "message for changing the states of all the child processes belonging to the subpool of the specific ID to the designated state" to the processor. (1-2-6) A message takes an arbitrary time from being sent to being received. Therefore, at any time,
"Message sent but not yet received"
(Hereinafter referred to as a message being communicated) may exist.

(2) "End detection method (WTC method)" Detecting the end of all child processes is a very basic and important function in the above-mentioned calculation model.
The state change method described in Document A uses a process end detection method called WTC method. This WTC method will be described below. The WTC method is described in Document B shown below.

Reference B; K. Rokusawa, N. Ichiyoshi et al.
“An Efficient Termination Detection and Abortio
n Algorithm for Distributed Processing Systems ”,
InProceedings of the 1988 International Conference
on Parallel Processing, Vol.1, pp.18-22, August 1
988.

This system is a system in which a "weight" is given to each of a control process, a subpool, and a message during communication. The weight given to the control process (negative) and the weight given to the subpool and the message (positive) are positive and negative integers. Then, the control is performed so that “the total weight is 0”. The weight of the control process is controlled to be 0 only when all the child processes are completed. In FIG. 4, the weights of 180, 250, and 400 are assigned to all (3) subpools, the weight of 20 is assigned to the communicating child process, and the weight of the control process is −.
In this example, a weight of 850 is added.

In order to manage such weights, each processor is provided with a control process weight table and a sub-pool weight table. In the control process weight table, the IDs of all control processes of the processor and the weights of the control processes are stored in association with each other.
In FIG. 5, there are control processes with IDs 0, 1, 35, and 40, which are -250, -450, -5014, and -4, respectively.
9 shows a table with a weight of 002. In the sub-pool weight table, the IDs of all the sub-pools that the processor has and the weights of the sub-pools are stored in association with each other. FIG. 6 shows a table in which control processes with IDs 21, 22, and 50 exist, and have weights of 800, 132, and 95, respectively. Since the subpool ID is stored in the subpool weight table, it is used to search for a subpool existing in the processor and to check whether a certain subpool exists in the processor.

Next, a change in weight when transferring a child process between processors will be described. The processor weights and sends the child process (the message that carries it).
At that time, the weight of the subpool is reduced by the amount added to the child process. On the other hand, when the processor receives a child process,
Based on the sub-pool weight table, it is determined whether or not the sub-pool with that ID exists. If it exists, the attached weight is added to the subpool having the same ID as the child process which received the weight. If it does not exist, a new subpool with that ID is newly created, and the weight attached to the received child process is used as the weight of that subpool.

FIG. 7 shows that the sending processor PEi assigns a weight of 40 to a child process having an ID of A and sends the child process, thereby changing the weight of the subpool from 420 to 380.
On the other hand, in the processor PEj that has received this child process, this indicates that the weight of the existing subpool with ID A has changed from 220 to 260 by this reception. In FIG. 8, the sending processor PEi changes the weight of the subpool from 420 to 380 by sending the child process with ID A with weight 40, and on the other hand, in the processor PEj that receives this child process, Is
There is no subpool with ID A, so the weight 40
Indicates that an ID with has generated a subpool of A.

Next, a change in weight when communicating a termination message from the subpool to the control process will be described. When the subpool is finished, send a termination message to the control process. All the weights of the finished subpools are attached to the finish message. Also, the completed subpool is deleted from the subpool weight table. When the control process receives the end message, the control process adds the given weight to its own weight. The weight of the added result is 0
When, the end of the process pool (all child processes have ended) is detected.

In FIG. 9, the subpool with ID A is completed,
An end message with its own weight 340 is sent to the control process, so that the weight of the control process is -14.
An example in which the value is changed from 00 to -1060 is shown. FIG.
Is the weight 1000, 800, 700 that the sub-pool has in sequence from each of all the sub-pools to the control process to which the weight -2500 is initially assigned.
The end message to which is added is given, the weight of the control process becomes 0, and the end of the process pool is detected.

(3) "State change method" In order to change the state of all child processes in the process pool, the control process "message to change the state of all child processes belonging to the subpool having a specific ID".
(Hereinafter referred to as a state change message) needs to be transmitted to all the processors. However, sending this state change message alone cannot completely change the state of the child process. That is, since the child processes are communicated by using the asynchronous message communication, as shown in FIG. 11, when the state change message is given to all the subpools, the communication is not performed in any subpool. There can be child processes and the state of this communicating child process cannot be changed. Note that, in FIG. 11, the letters “a” and “b” relating to the control process and the child process represent states.

According to the above-mentioned document A, when a processor creates a subpool, the fact is notified to the control process so that the control process recognizes the existence of the communicating process and the state of the child process in the communicating process. The method that can be changed is also described. Below, the conventional state change method that can change the state of the child process even if there is a child process in communication when the state change message is sent,
An explanation will be given separately for a system without message passing and a system with message passing.

(3-1) When there is no message overtaking When a subpool is created, the processor sends a message (hereinafter referred to as a creation message) that conveys the creation of the subpool to the control process. The control process remembers the originating processor when it receives the create message and deletes it when it receives the end message. There is always a subpool in the processor stored in the control process. Each processor has a subpool table for storing the processor that sent the generated message. As shown in FIG. 12, the sub-pool table is made up of the ID of the sub-pool and the sender processor of the generated message.

(3-1-1) Processing of control process The processing of the control process at the time of state change using this subpool table is as follows. Send the status change message only to the processors registered in the subpool table. When the generation message is received after such processing, the transmission source processor is stored and the state change message is transmitted. The state change message is also weighted, and the weight of the control process is reduced accordingly.

(3-1-2) Processing of Processor The processor that receives the status change message first searches the weight table for subpools to determine whether or not there is a subpool with the same ID as the ID indicated by the status change message. Check in.

If the subpool exists, the state of all child processes belonging to the subpool is changed to the state designated by the state change message. In addition, the weight held by the status change message is added to the weight of the subpool.

On the other hand, when the sub-pool does not exist, the status change message has a message by sending back the weight of the status change message (hereinafter referred to as a weight return message) to the control process. The weight that was used is returned to the control process.

(3-1-3) State Change The child process existing in the subpool (registered in the subpool table) recognized by the control process at the start of the state change is the first state change message described above. The state is changed by sending. On the other hand, the state of the child process that was in communication at the start time is changed by the state change message retransmitted after receiving the generation message. This is because the communicating child process arrives at some processor within a finite time, a subpool is created, and a create message is sent to the control process.

13 to 16 show that the child processes belonging to the subpools in the processors PEi and PEj registered in the subpool table have their states changed (from a to b) by the initial state change message, and the first state change. The child process in communication at the time of sending the message changes its state by the state change message sent again by the control process after the destination processor PEk that is not registered in the subpool table sends the create message to the control process. It is meant to be changed.

(3-2) When message is overtaken When message is overtaken (system), a state change message instructing change to a certain state is sent to another state later. There is a possibility that you will be overtaken by the state change message that instructs you to change. Therefore, if the state is simply changed according to the received state change message, the state may not be changed correctly. For example, the control process first changes the state to a state change message (hereinafter, state change message (a)
, But then send a state change message (b) to change to state b, the state of the child process should eventually be set to state b. When the status change message (b) overtakes the status change message (a), the status is finally set to the status a.

Therefore, special processing is required when a message is overtaken. The above-mentioned document A describes a method in which the state to be handled is limited to two states, but a problem does not occur even when a message is overtaken. This method handles two states, stop and start.
This is a method of setting the state correctly by limiting the number of states and counting the number of received state change messages.

A detailed description will be given below. The processor is
A counter is provided for each sub-pool and operated as follows. For example, a counter field is provided in the sub-pool weight table (FIG. 6), and the counter value is operated as follows. It is set to 0 when the subpool is created. + When a status change message (stop) is received
Do 1 When the status change message (start) is received, -1 is set.

When the state change message (stop) or the state change message (start) is received, the state is set as follows after the counter is operated as described above. If the counter value is positive, set the state to stop. If the counter value is 0 or less, set the state to start.

In this way, the status change message (st
Even if the order of arrival of op) and the state change message (start) is exchanged, the state is correctly set by the above operation.

FIG. 17 shows a status change message (stop) that causes the control process to change the status of the subpool to the status stop and a status change message that causes the status to start when the subpool is in the state start and the counter value is 0. (Start) is transmitted in this order, and the final state is correctly set to the state start even if the state change message (start) transmitted later arrives at the processor first.

[0032]

(1) Problems common to cases where there is overtaking and no overtaking, however, according to the conventional process state changing method,
There are the following problems when applied to both a system with overtaking messages and a system without overtaking messages.

It was necessary for each processor to perform a preparation process of transmitting a generation message to the control process when the subpool was generated due to the state change. In practice,
The child processes are often communicated in order to obtain a uniform operation rate of each processor, or in parallel with the process, or in consideration of high-speed processing, and the number of times the generated message is transmitted increases. In the worst case, it will send as many spawned messages as there are child processes. However, as a result, some process pools are not instructed to change the state, communication of generated messages for such process pools is wasteful, and the large number of times is a problem.

When the controlling process receives the create message, it needs to remember its originating processor. Even if only one subpool is stored, a capacity for the number of all processors is required, and when there are a plurality of processors, a capacity corresponding to the number is required, and this table has pressed the memory capacity.

Further, according to the conventional method, the control process cannot confirm that the state change has been completed. That is, when a certain state change message is transmitted, it can be guaranteed that the state changes of all child processes are completed within a finite time, but the completion of the change cannot be detected. Therefore, the state control by the control process may not be properly performed.

(2) Special Issue When There is Overtaking In the case of a system that has an overtaking message, in addition to the above problems, there are the following problems.

A counter for managing the reception of the status change message must be prepared for each subpool, and the operation of this counter is extremely complicated, and this information also causes a pressure on the memory capacity. .

Moreover, the number of states to be handled is limited to two, and it cannot be applied to a system handling more states.

(3) Purpose The present invention has been made in consideration of the above points, and is necessary for the state change processing, which does not require the preparation processing for the state change performed outside the state change processing period. It is an object of the present invention to provide a process state change method which has a small memory capacity and which can detect the completion of the state change by the control process and has an unlimited number of states that can be handled.

[0040]

In order to solve such a problem, according to the present invention, there is provided a process state changing method for a multiprocessor system which performs inter-processor processing by asynchronous message communication. The process state changing method in which the control process changes the state of the subpool existing in each processor of the process pool including the above child processes by controlling the state changing unit of each processor is described below. Realized by processing.

First processing: processing performed by a control process that sends a state change message to all processors.

Second process: This process is performed when the state change unit of each processor receives a state change message, and changes the state of the subpool if the subpool for which the state change is requested exists in the processor. And sending a changed message to the control process and storing the state-changing information, and storing the state-changing information when the subpool for which the state change is requested does not exist in the processor.

Third processing: When a child process is given from another processor in a state where the state-changing information is stored, the state of the received child process is changed and a changed message is sent to the control process. The process performed by the state change unit of each processor.

Fourth processing: A control process, which monitors completion of the change of the state of the process pool based on all the changed messages received and sends a completion message to all the processors when the completion is detected. The process to perform.

Fifth process: This is a process performed by the state changing unit of each processor when a completion message is received. The state changing information is deleted and an arrival confirmation message for the completion message is sent to the control process. processing.

Sixth processing: Processing performed by the control process which terminates the state change processing of the process pool when the arrival confirmation message is received from all the processors.

[0047]

In the present invention, when it is necessary to change the state of a certain process pool, the control process sends a state change message to all processors (first process).

At this time, the state changing unit of each processor
If there is a subpool for which the state change is required for the processor, the state of the subpool is changed, the changed message is sent to the control process, the state change information is stored, and the state change is requested by the processor. When the created sub-pool does not exist, the status change information is stored (second process).

The state changing section of each processor changes the state of the child process when another process gives a child process in a state where the state change information is stored, and the changed message is displayed. Send to the control process (third processing).

After transmitting the state change message, the control process waits for the reception of the changed message. When the control process receives the change message, the control process monitors the completion of the state change of the process pool based on all the received changed messages and confirms the completion. When detected, a completion message is sent to all the processors (fourth processing).

Upon receipt of the completion message, the state change section of each processor deletes the state-changing information and
An arrival confirmation message for the completion message is sent to the control process (fifth processing).

The control process ends the process for changing the state of the process pool by receiving the arrival confirmation message from all the processors (sixth process).

[0053]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the drawings. The calculation model is as described in the section “Prior Art”.

(1) Table In this embodiment, each processor has a control process weight table and a sub-pool weight table as tables. That is, unlike the conventional case, the sub-pool table is not provided.

(1-1) Control Process Weight Table The control process weight table of this embodiment is composed of three fields, as shown in FIG. The ID of the process pool related to the control process existing in the processor is stored in the first field. The weight assigned to the control process is stored in the second field. The third field is used as a work area at the time of the status change processing, and stores the weight of the child process for the status change completed at the time of the status change processing, the number of arrival confirmation messages to be described later not received, and the like. To do.

FIG. 18 shows that the control processes with IDs 0, 12, 3250 and 4000 respectively have a weight of -30.
0, -8200, -5900 and -75 are stored, and I
For the process pool with D of 3250, the state change of the child process is in process, and the state change has been completed for the child process with a weight of -3700.

(1-2) Sub-pool weight table The sub-pool weight table of this embodiment is made up of three fields as shown in FIG. The ID of the process pool related to the subpool existing in the processor is stored in the first field. The weight given to the sub-pool is stored in the second field. The third field is a new state field and stores a new state which is being changed. When a state is written in the new state field, it means that the subpool is undergoing a state change. As will be described later, when a message is overtaken, it may be stored in the new state field indicating that it is waiting to receive a state change message.

FIG. 19 shows that weights 3100 and 3100 are assigned to the subpools with IDs 3, 55, 2300 and 7214, respectively.
55, 209 and 20340 are stored and the ID is 2300
It is indicated that the sub-pool of is being changed to the state b.

Since the subpool ID is also stored in the subpool weight table of this embodiment as described above, it is possible to search for a subpool existing in the processor,
It is used to check whether a certain subpool exists in the processor.

(2) Transmission / reception processing of child process (2-1) Transmission processing of child process When a child process is to be transmitted, the weight table for the sub-pool is searched and the state change processing of the child process of that ID is being performed. It is determined whether or not there is. That is, it is determined whether or not the state is stored in the new state field.
If the state is not being changed, the child process is weighted and transmitted according to the above-mentioned WTC method. In this case,
The weight of the subpool is reduced accordingly. That is, the value of the weight field of the sub-pool weight table is updated. When the weight becomes 0 by this subtraction, the sub pool is deleted from the sub pool weight table. On the other hand, as a result of the search of the weight table for the sub-pool, if the state is being changed, the state change is completed and the normal state (state not being changed) is waited for, and when the normal state is reached, the above-mentioned WTC method is used. Send according to.

(2-2) Processing when a child process is received When a child process is received, to be precise, when a message carrying the child process is received, first, the sub-pool weight table is searched. That is, it is checked whether or not a subpool having the same ID as the received child process already exists, and if it exists, whether the subpool is in the normal state or the state is being changed. Find out.

(2-2-1) If the sub-pool having the same ID as the received child process does not exist, a sub-pool having the ID is generated and registered in the sub-pool weight table. At this time, the received weight of the child process is used as the initial value of the weight of the generated sub-pool.

(2-2-2) When a subpool having the same ID as the received child process exists and the subpool is in the normal state, the weight of the received child process is added to the weight of the subpool. That is, the value of the weight field of the sub-pool weight table is updated to the added value.

(2-2-3) If a subpool having the same ID as the received child process exists and the state of the subpool is being changed, the state of the child process is set to the new state field of the subpool weight table. Change to the new state described in. In addition, the weight of the child process is added to the weight of the subpool. Further, the weight held by the child process is sent to the control process by the changed message. This changed message conveys that "the state has been changed for the child process by this weight". In this case, the information that the status is being changed is continuously stored in the sub-pool weight table.

FIG. 20 shows that a processor having an ID of 8 and a weight of 350, which has a subpool in the process of changing to the state b, has a weight of 50 transmitted from another processor.
Since the child process whose ID is 8 and whose status is a is received, the weight of the sub-pool weight table is updated to 400, and a changed message with a weight of 50 is sent to the control process.

When the control process receives the changed message, it updates the weight of the work area in the control process weight table.

(3) State Change Process Next, a process for changing the state of all child processes in the process pool will be described. The characteristic part of the state change processing is the same in both the system in which there is no message overtaking and the system in which there is message overtaking, but the case will be described separately below.

(3-1) When there is no message overtaking (3-1-1) General processing flow Detailed description will be given of the system state change processing without message overtaking with reference to the flowchart of FIG. To do. Note that FIG. 1A shows the processing of the control process, and FIG. 1B shows the processor (software, hardware,
It may be realized by any of the firmware and should be called a state change unit of the processor, but it is also called a processor hereinafter). Further, in FIG. 1, “message” is abbreviated and represented by “MS”.

When it becomes necessary to change the state of a process pool, the control process starts the process shown in FIG. Then, first, a state change message carrying a new state is transmitted to all the processors (step 100), and the value (weight) of the weight field of the sub-pool in the control process weight table is copied to the work area field ( Step 101). The work area weight indicates the sum of the weights of the child processes whose state has not been changed yet. The control process then proceeds to wait for the modified message (step 102). Note that all the processors are destinations, which is different from the conventional case. Further, it is not necessary to give a weight to the state change message.

Each processor that has received the status change message starts the processing shown in FIG. And first,
Whether there is a subpool for which status change is instructed,
Check using the subpool weight table (step 2
00).

If the subpool exists, the state of all child processes belonging to the subpool is changed to a new state indicated by the state change message (step 201),
The new state is written in the new state field of the weight table for sub-pool (step 202). After that, the changed message with all the weights held by the subpool is returned to the control process (step 203). As described above, the changed message with all the weights held by the subpool means that "the state has been changed for the child processes of this weight". The weights in the sub-pool weight table are retained even after this transmission.

On the other hand, if the sub-pool does not exist, after registering the sub-pool in the weight table for sub-pool (step 204), it is confirmed that the process pool indicated by the status change message is undergoing status change. (Step 205). That is,
The processor writes a 0 in the weight field and a new state in the new state field. After that, the process proceeds to a message reception waiting state (step 206).

As described above, the control process proceeds to the waiting state for receiving the changed message after transmitting the state change message, and when the changed message from any process is received, it is added to the changed message. After adding the existing weight to the weight stored in the work area of the control process weight table (step 10
3) It is determined whether or not the weight after addition is 0, which indicates that the state changes of all the child processes are completed (step 104). If it is not 0, the weight of the work area approaches 0 as the reception of the changed message returns to the waiting state for the changed message, and eventually becomes 0, and a positive result is obtained in step 104.

At this time, the control process sends a completion message to all the processors (step 105). Here, the completion message is a "message which notifies that the state changes of all child processes of that ID have been completed". After that, the number of completed messages sent (equal to the number of processors: represented by T in FIG. 1)
Is written in the work area of the control process weight table (step 106). Then, the process goes to the waiting state for the arrival confirmation message (step 107).

The processor waits for message reception after setting the status change in the sub-pool weight table by storing a new status (step 20).
In 6), when the message is received, the type of the received message is determined (step 207). The message sent to the subpool whose state is changing is a message for carrying a child process or a completion message.

If the message conveys a child process, the processing described in the above-mentioned child process reception processing is performed (step 208). That is, the state of the child process is changed to a new state, the weight held by the child process is added to the weight of the subpool, and the weight held by the child process is sent to the control process by the changed message.
After that, the process returns to the message reception waiting state.

The completion message is always received, regardless of whether the message carrying the child process was received during the state change. At this time, the processor determines whether the weight of the weight field of the sub-pool weight table is 0 (step 209). That is, when the state change message is received, the sub-pool does not exist in the processor, and the sub-pool additionally registered in the weight table for the sub-pool to store the new state (step 204
See).

If the weight is not 0, that is, for the subpool existing in the processor at the time of receiving the state change message, the state stored in the new state field of the subpool weight table is deleted. After the subpool is returned from the state change to the normal state (step 210), an arrival confirmation message for the completion message is returned to the control process and a series of processing is ended (step 211).

On the other hand, if it is determined whether or not the weight is 0, that the weight is 0, the sub pool is deleted from the sub pool weight table (step 212), and then the arrival confirmation message for the completion message is processed by the control process. To end the series of processing (step 21).
1).

When the control process receives the arrival confirmation message from any of the processors, after decrementing the completion message transmission number T stored in the work area field of the control process weight table by 1 (step 108), It is confirmed whether or not the value T is 0 (step 109). If it is not 0, the process returns to the waiting state for receiving the arrival confirmation message. While repeating such processing, arrival confirmation messages from all processors are received and the number of completed message transmissions (arrival confirmation message number) T becomes 0. At this time, the work area field of the control process weight table is reset. A series of processing is ended (step 110).

(3-1-2) State Change Processing and Termination of Subpool The basic processing flow for changing the states of all child processes belonging to the process pool has been described above. A subpool (child process) may end during the process. Below, the relationship between the end of the sub-pool and the state change processing will be explained.

When the subpool whose state is being changed ends, the processor sends the changed message to
Wait for the state change to complete. The completion message is received, the arrival confirmation message is transmitted, and then the termination message is transmitted. This is because when the end message is transmitted first, the subpool is deleted from the subpool weight table, and the processing when the completion message is received cannot be executed.

When the subpool whose status is not being changed is terminated, the processor performs a normal termination process. That is, a termination message is sent to the control process, and the terminated subpool is deleted from the subpool weight table.

When the control process receives the termination message during the state change, the weights of both the weight field and the work area field of the control process weight table are updated by the weight assigned to the termination message.
As a result, even when the weight of the work area field becomes 0, it is determined that the state changes of all the child processes have been completed, and the process proceeds to the completion message transmission process.

(3-1-3) Description of Specific Example Next, the process state changing process will be described using a specific example with reference to FIGS. 21 to 25.

FIG. 21 shows that when the state change message for changing the state from a to b is sent, the number of processors in which the sub-pool of the state change exists is PEi and PEj. An example in which a child process (carrying message) directed to PEi and a child process directed from processor PEj to processor PEk exist is shown.

When the processors PEi and PEj in which the sub-pool exists receive the state change message,
As shown in FIG. 22, after changing the state of the sub-pool from a to b, the changed message with the sub-pool weights 2000 and 1650 is returned to the control process, and the state change is in progress. To store. On the other hand, when the processor PEk that does not have a sub-pool receives the state change message as in the case of other processors not shown, it is registered in the weight table for sub-pool and is in the state change as shown in FIG. Is stored in the sub-pool weight table. Even if the control process receives the changed message transmitted at this stage, the weight of the work area in the control process weight table does not become 0 because there is a child process in communication.

After storing the fact that the state is being changed in the sub-pool weight table, each of the processors PEi and PEk which has received the child process, as shown in FIG.
The state of the received child process is changed to the new state b, and the changed message with the weight of the received child process is sent to the control process. Even after this transmission, the state is being changed. Of course, processor PEi
And PEk respectively update the value of the weight field of the sub-pool weight table when receiving the child process.

When these status change messages are received and the work area weight is updated, the weight becomes zero. At this time, the control process sends a completion message to not only these processors PEi and PEk but also all the processors, as shown in FIG. The control process also stores the number T of completion message transmissions in the work area.

When each processor PEi, PEj, PEk in which the sub-pool to which the child process belongs exists, receives the completion message, controls the arrival confirmation message after returning the state under change to the normal state, as shown in FIG. Reply to the process and end the series of processes. Each of the other processors (not shown) in which there is a subpool with no child process (registered in the table with a weight of 0) receives the completion message, and after deleting the subpool from the table,
The arrival confirmation message is returned to the control process, and the series of processing is terminated. The control process receives the arrival confirmation message by the number T of the completion messages transmitted, and when the value of the work area becomes 0, clears the work area and ends the series of processes.

(3-2) When there is a message overtaking The above-mentioned forced termination method when there is no message overtaking cannot be directly applied when there is a message overtaking. The processing part to be changed will be described below.

The problem of message passing is that
There are cases where the completion message overtakes the state change message (a phenomenon that occurs only for processors without subpools) and cases where the end message overtakes the arrival confirmation message. Since the arrival confirmation messages are the same message overtaking each other,
There is no problem.

(3-2-1) Overtaking of a status change message by a completion message When the processing at the time of receiving an overtaking completion message is not specified, such an overtaking completion message or an overtaken status change message is sent. The reception of the message causes inconvenience such as system runaway. Therefore, when there is an overtaking of a message, it is necessary to clearly specify the processing when the completion message overtakes the state change message.

First, the processing when the processor receives the completion message will be described. When a completion message is received, instead of simply sending an arrival confirmation message, check whether or not the state change message is received, and if it is not received, wait for the state change message to arrive and arrive. You need to send a confirmation message. Whether or not the state change message has been received can be checked by searching the weight table for sub-pool. If the corresponding ID is registered and the state is stored in the new state field (that is, if the state is being changed), it has been received, and in other cases, it has not been received.

Therefore, the processing of the processor is as follows. First, it is checked whether or not the status change message has been received. As a result, when it has been received (when there is no overtaking), it is processed in the same manner as when there is no overtaking. When not received (when there is overtaking), it is determined whether or not the subpool with the ID indicated by the completion message exists. If it exists, the subpool is registered immediately in the table. Later, write "waiting for status change message" in the new status field. When the subpool is not registered, 0 is written in the weight field when registering it.

Next, the processing when the processor receives the status change message will be described. Even when the status change message is received, different processing is performed depending on whether the completion message has been received or not. Whether or not the completion message has been received can be checked by searching the weight table for the sub-pool. That is, the corresponding ID
If "waiting for status change message" is written in the new status field of, it has been received, otherwise it has not been received.

Therefore, the processing of the processor is as follows. First, it is checked whether or not the completion message has been received. As a result, when not received (when there is no overtaking), the same process as when there is no overtaking is performed. When the completion message has been received (when there is overtaking), the processing when the completion message is received when there is no overtaking is performed. That is, if the weight field is not 0, the new state field is cleared to return to the normal state, and the arrival confirmation message of the completion message is sent to the control process. After the deletion, send the arrival confirmation message of the completion message to the control process.

(3-2-2) Overtaking Arrival Confirmation Message by End Message As described above, when the subpool whose state is being changed is terminated, the processor transmits the arrival confirmation message and then transmits the termination message. To do. However, in the case of a system in which a message is overtaken, the termination message may reach the control process first. Depending on the reception of this end message, the sub-pool may be deleted from the control process weight table. Even if the arrival confirmation message is received by the control process after this, it is not clear what process the control process should execute, and there is a risk of runaway. Therefore, when there is an overtaking of a message, it is necessary to clearly specify the processing when the end message overtakes the arrival confirmation message.

In order to confirm the end of the process pool, it is not enough that the weight becomes 0, and it is necessary to confirm reception of all arrival confirmation messages. Since the number of unacknowledged arrival confirmation messages is written in the work area of the control process weight table, reception of all arrival confirmation messages can be confirmed by checking the work area.

First, the processing when the end message is received will be described. The control process updates the weight according to the WTC method by receiving the end message, and when the weight becomes 0, the work area of the control process weight table is checked. When the work area is 0, all the arrival confirmation messages have been received, so the end of the process pool is detected. That is, the process pool is deleted from the table. On the other hand, if it is not 0,
Since there is an arrival confirmation message that has not been received,
The process waits for the arrival confirmation message without any processing, and reserves to detect the end of the process pool even if the weight is 0. That is, the process pool table is not deleted.

Next, the processing at the time of receiving the arrival confirmation message will be described. At this time, the value of the work area is updated to determine whether it is 0 or not. If it is not 0, there is an arrival confirmation message that has not been received, and therefore the state of waiting for the next arrival confirmation message is kept without checking the value of the weight field. On the other hand, when the value of the work area becomes 0 due to the update, all arrival confirmation messages have been received, so it is further checked whether the value of the weight field is 0 or not. If it is 0, the end of the process pool is detected and it is deleted from the table. On the other hand, when the weight is 0, the state change is completed, but the process pool is not completed, so only the work area is cleared.

(4) Effects of the Embodiments According to the above embodiments, the following effects can be obtained.

There is no need to perform processing for changing the state except during the state change processing period. For example, transmission processing of the generated message is unnecessary. That is, although the process of transmitting the generated message to the subpool in which the state is not changed is useless after all, such useless process does not occur.

The table required for the state change processing is only the table used also for the normal termination processing, and the table used only for the state change processing is unnecessary, and effective use of memory resources can be realized.

The present invention can be applied to both a system in which a message is overtaken and a system in which no message is overtaken, and the difference in processing between the two is very small, and the design and configuration of many parts can be made common. In addition, it is easy to change a system that does not overtake a message to a system that overtakes.

Further, according to the above-mentioned embodiment, the number of kinds of states that can be handled is not limited. From this point, the present invention can be applied to a wide range of multiprocessor systems.

Conventionally, the control process could not detect that the states of all the subpools have been changed, but this embodiment can detect it. Therefore, it becomes easy for the control process to perform the processing at an appropriate timing.

(5) Other Embodiments During the status change or waiting for the status change message to be received, a dedicated table or the like may be used for management.

The status change message and the completion message sent by the control process may be sent by using the broadcasting function of the multiprocessor system.

The application and the number of processors of the multiprocessor system to which the process state changing method of the present invention is applied are not limited.

[0111]

As described above, according to the present invention, the state change processing is performed only by transmitting and receiving the messages within the state change processing period. Therefore, the number of arrival confirmation messages received and the changed message Since the control process confirms that sufficient reception is in progress, the processor also changes the state of the child process received during the change and reports it to the control process. As a result, the preparation process for the state change performed outside the state change process period is unnecessary, the required memory capacity is small, the control process can detect the completion of the state change, and the state can be handled. It is possible to realize a process state changing method whose number is not limited.

[Brief description of drawings]

FIG. 1 is a processing flowchart of a control process and a processor (state changing unit thereof) according to an embodiment.

FIG. 2 is a configuration diagram of a process pool.

FIG. 3 is an explanatory diagram of application of a process pool to a multiprocessor system.

FIG. 4 is an explanatory diagram of a WTC method which is a conventional end detection method.

FIG. 5 is a configuration diagram of a conventional control process weight table.

FIG. 6 is a configuration diagram of a conventional sub-pool weight table.

FIG. 7 is an explanatory diagram of transmission / reception processing of a child process when a subpool exists in a conventional transmission destination.

FIG. 8 is an explanatory diagram of transmission / reception processing of a child process when a subpool does not exist in a conventional transmission destination.

FIG. 9 is an explanatory diagram of a conventional sub-pool end process.

FIG. 10 is an explanatory diagram of a conventional process pool end detection process.

FIG. 11 is an explanatory diagram of an influence of a conventional child process during communication on state change processing.

FIG. 12 is a configuration diagram of a sub-pool table that has been conventionally used.

FIG. 13 is an explanatory diagram (part 1) of conventional state change processing.

FIG. 14 is an explanatory diagram (part 2) of conventional state change processing.

FIG. 15 is an explanatory diagram (part 3) of conventional state change processing.

FIG. 16 is an explanatory diagram (part 4) of conventional state change processing.

FIG. 17 is an explanatory diagram of conventional state change processing when a message is overtaken.

FIG. 18 is a configuration diagram of a control process weight table of the embodiment.

FIG. 19 is a configuration diagram of a sub-pool weight table according to the embodiment.

FIG. 20 is an explanatory diagram of a process when a child process is received during the state change of the embodiment.

FIG. 21 is an explanatory diagram (part 1) of the state changing process of the embodiment.

FIG. 22 is an explanatory diagram (part 2) of the state changing process of the embodiment.

FIG. 23 is an explanatory diagram (part 3) of the state changing process of the embodiment.

FIG. 24 is an explanatory diagram (Part 4) of the state changing process of the embodiment.

FIG. 25 is an explanatory diagram (Part 5) of the state changing process of the embodiment.

[Explanation of symbols]

100 ... Transmission process of status change message, 101 to 10
4 ... Status change completion detection processing, 105 ... Completion message transmission processing, 106-109 ... Arrival confirmation message reception completion detection processing, 110 ... Series of processing termination processing,
201 ... Status change processing, 202, 205 ... Storage processing during status change, 203, 208 ... Changed message transmission processing, 204 ... Subpool registration processing in subpool weight table, 210 ... Return to normal status processing, 211
... process of transmitting arrival confirmation message, 212 ... process of deleting sub-pool from weight table for sub-pool.

Claims (2)

(57) [Claims] 1. A method for changing a process state of a multiprocessor system for performing interprocessor processing by asynchronous message communication, which is a subpool of a process pool consisting of one control process and one or more child processes and which exists in each processor. In the process state changing method in which the control process controls and changes the state changing unit of each processor, the first process performed by the control process for transmitting a state change message to all the processors; The process performed when the state change unit of each processor receives a state change message, and when the subpool for which the state change is requested exists in the processor, the state of the subpool is changed and the changed message is displayed. It sends to the control process and stores the status change information, A second process that stores the state-changing information when the subpool for which the state change is requested does not exist in the processor, and a child process is given from another processor in the state where the state-changing information is stored. A third process performed by the state changing unit of each processor for changing the state of the child process received and transmitting the changed message to the control process, and the process based on all the received changed messages A fourth process performed by the control processor, which monitors completion of the change of the pool state and sends a completion message to all the processors when the completion message is detected; Is a process to be performed when a message is received, the above status change information is deleted, and the arrival confirmation for the completion message is received. It comprises a fifth process of transmitting a message to the control process, and a sixth process of the control process that terminates the process pool state change process by receiving an arrival confirmation message from all the processors. A process state changing method characterized by the above. 2. The process state changing method according to claim 1, applied to a multiprocessor system, wherein a completion message may reach the processor before a state change message. This is a process to be executed instead of the fifth process when a completion message is received, and when the above-mentioned state-change information is stored , the process is completed.
Status change message before receiving the end message
Is received, the fifth process is executed , while
If the above status change information is not stored, the status change
Receive a completion message before receiving a change message
And have signal, there in a seventh process of storing reception waiting status information change message, the state changing unit of each processor is performed in place of the second process when it receives the state change message processing If the waiting information for the status change message is not stored, the status is sent before the completion message is received.
The second process is executed assuming that the state change message is received , while the waiting information of the state change message is
Receive status change message if stored
Eighth processing for deleting the status change information and transmitting an arrival confirmation message for the completion message to the control process on the assumption that the completion message is received before the process status. How to change.