JPH0644090A - Master/slave computer system and master selection control method - Google Patents

Abstract

PURPOSE:To prevent the useless master re-selection processing associated with competition occurrence. CONSTITUTION:A same processing is executed in parallel by plural computers (CPU), one of them is adopted as the main one (M) and the others are the stand-by ones (S) and also one of CPUs being S is adopted as M at the time of a fault in M in a system. In the system, respective CPUs are provided with a role storage means 11 storing role information which shows whether CPU itself is M or not, a time storage means 12 (role renewal processing decision time storage part) storing the renewal time of the contents of the role storage means as time information, a selecting means 14 which executes an inquiry to respective CPUs at the time of the fault in M, receives role information and time information of respective CPUs, which are obtained by the result, so as to select the optimum one as M, renews the role decision time, adopts time information after renewal as new time information and outputs role information and new time information of respective CPUs, which are decided by selection, a responding means 13 responding the present state of itself adding role information and time information for the inquiry, a means 15 (updating part) which receives the report of role information and new time information for self CPU which is decided by the selecting means and updates the storage contents of the role storage means and the time storage means of the self CPU and a recognizing means 16 executing a re-inquiry.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a master selection method for deciding which computing entity outputs a system as a whole in a system in which a plurality of computing entities simultaneously perform the same processing to improve reliability. The present invention relates to a master / slave computer system and a master selection control method for the master / slave computer system, which can avoid waste of the master re-election process due to the occurrence of conflict due to the existence of an old master and enable efficient operation. .

[0002]

2. Description of the Related Art With the downsizing, cost reduction and higher performance of electronic computers, the range of applications has become widespread, and various ways of using them have been proposed. It became so. Distributed systems are one of the most promising technologies for the future. One of the reasons is the high reliability due to the existence of multiple pieces of hardware.

For example, by causing a plurality of computers to perform the same processing at the same time, even if a failure occurs in one of the computers, the processing can be continued by another computer. The improvement of reliability due to such multiplexing has already been used in many cases in a system requiring high reliability, and has been effective.

By the way, in a computer system, its input and output are usually the output and input of another system. For example, in a computer system that controls a robot, based on the robot's current state such as the position of the target robot's arm and the position of the material set on that robot, that robot's next action to take. , Which is the output for that robot and will be sent directly to the robot.

In such a system, the output to the robot is 1 even if the system performs the same processing by a plurality of computers to improve reliability.
It is desirable to limit the number to one. This is because the robot needs to unify a plurality of messages, which means that the versatility of the robot is reduced. For this reason, only one of the plurality of existing computers performs the actual output processing, and the other computers do not perform the actual output. If, for some reason, the computer that should output originally cannot output. When that happens, a function is required to detect that and select a new representative, and the selected representative outputs.

[0006]

In order to realize such a function, each computer examines the situation and makes a judgment so that output can be performed regardless of which computer is the last one. In many cases, the method of notifying computers to each other is used. Further, it is absolutely necessary for the computer selected as the master to confirm that there is no other master after a certain period of time in order to ensure the reliability of the operation.

Here, it is assumed that an abnormality occurs in this new master until the new master is selected and it is confirmed that no other master exists, and another computer is selected as the latest master.

[0008] Then, when the first selected master confirms that there is no other master, it will confirm itself as the only master. But,
In reality, this computer is abnormal, and the selection process must be carried out immediately.

Therefore, there is a problem that useless processing is repeatedly executed forever. Also, in order to avoid this, if the confirmation after a certain time is omitted, there may be two or more masters this time, and there is a problem that the certainty of the operation must be sacrificed. Occurs.

The present invention has been made in view of the above circumstances, and an object thereof is to determine the validity of a received role notification by the number of times of master selection. Thus, it is possible to confirm whether or not there is another master after a certain time has elapsed after the master is selected, and improve the reliability of the operation while preventing unnecessary master selection processing from being repeated.

[0011]

In order to achieve the above object, the present invention is configured as follows. That is, firstly, the same processing is performed in parallel on a plurality of computers, and
In a system in which one of them is used as a master and the other is used as a spare, and when a master fails, one of the spare computers operates as a master, and each computer has its own master. Role storage means for storing information on whether or not there is a role, number storage means for storing the number of times of renewal of the role stored in the role storage means as number information, and inquiry to each computer in case of master failure Based on the above-mentioned role information and frequency information of each computer obtained as a result, an optimal one is selected as a master, and the number of times of role determination is renewed and the frequency information after renewal is renewed. And a selection means for outputting the role information and new count information of each computer determined by the above selection, and its own current state including the role information and the count information in response to the inquiry. Reply means for replying, change means for updating the contents stored in the role storing means and the number-of-times storing means of the own computer upon receiving notification of role information and new number of times information for the own computer determined by the selecting means, When the own computer is selected as a new master, a confirmation means is provided to re-execute the inquiry.

Secondly, a plurality of computers perform the same processing in parallel, and one of them is used as a master, and the other is used as a spare. In a system in which one of the above is used as a master, each computer has a role storage step of storing information on the role of whether or not it is a master,

The number of times of renewal of the role stored in the role storage step is stored as the number of times information, and in case of master failure, each computer is inquired, and the information of the role of each computer obtained as a result is obtained. And a number of times information to select an optimal one as a master, and the number of times of role determination is updated to obtain new number of times information, and the role information of each computer determined by the above selection and the above The step of outputting the updated number-of-times information, the step of responding to the inquiry about the current state of the machine including the role information and the number-of-times information, the role information and the new information of the determined and output self-computer. The change step of receiving the number of times information and updating the stored information of the role information of the own computer and the number of times information, and the own computer is selected as a new master. When it is, providing a confirmation step for executing the query again.

[0014]

In such a configuration, each computer stores information on the role of whether or not it is a master, and
In storing the role information, the number of renewals performed by the role renewal determination process is stored as the number of times information, and the current state of the user including the role information and the number of times information can be returned to the inquiry.

In the case of a master failure, the computer that knows this makes an inquiry to each computer, receives the information on the role of each computer and the frequency information obtained as a result, and selects the optimum one as the master. To do.

Then, the number of times of role determination is updated to obtain new number of times information, the role information of each computer determined by the above selection and the updated number of times information are output, and each computer is also output. Then, the stored contents of the role information of the own computer and the number of times information are updated in response to the information of the role and the new number of times for the own computer determined and output. Also, when your computer is selected as a new master,
Confirm the master conflict by executing the above inquiry again.

As described above, in the present invention, each time the role is renewed,
By updating the number of role renewals and holding it as a cumulative value, it is possible to avoid waste of master re-election processing due to conflicts due to the existence of old masters by comparing this renewal count information on each computer. Therefore, the efficiency of the system can be improved.

[0018]

The present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram showing an example of a system to which the present invention is applied. In the figure, 1A, 1B, to 1n are electronic computers, 2 is an intra-system communication line, and 3
Is an output line.

The computers 1A, 1B, to 1n are connected to each other by an intra-system communication line 2 and an output line 3. In this example, each computing entity is one computer, but if each computing entity has a similar configuration, the same thing can be realized independently of the number of computers. The electronic computers 1A, 1B, to 1n to be used can be realized not only by a large-scale computer but also by a microprocessor or the like.

Computers 1A, 1B, ...
n are connected by an intra-system communication line 2, and various communications within the system are performed using this intra-system communication line 2.
Communication in the master selection described here also uses the intra-system communication line 2. Separately from this, an output line 3 for outputting to another system is connected to each of the computers 1A, 1B, to 1n similarly to the intra-system communication line.

FIG. 2 is a block diagram showing the internal structure of each of the computers 1A, 1B, to 1n. As shown in the figure, each of the computers 1A, 1B, to 1n has a role storage unit 11, a role renewal process determination number storage unit 12, a response unit 13, and a selection unit 1 respectively.
4, a changing unit 15 and a confirming unit 16, and in addition to this, the processing unit 1 for performing actual processing and detecting a master failure.
There is 7.

Of these, the role storage unit 11 is for storing and saving the current role such as whether or not the own computer is a master as information, and the role renewal decision processing frequency storage unit 12 is for the role This is a storage unit for updating and storing the number of times each time a new assignment is made, and stores the number of role assignment updates up to the present of the own computer as number information. Further, the reply unit 13 returns a reply to the inquiry, and inquires about the current role of the own computer and the number of renewal of the role assignment (that is, the value of the role renewal decision processing frequency storage unit 12). It has a function of replying to the selecting unit 14 of the computer.

The selection unit 14 issues a "state inquiry" to the response unit 13 of each computer other than itself, and the intra-system communication line 2
In addition to sending out via this, upon receipt of the response to this inquiry, it finds the one with the maximum number of times X as the number of times information, and it is the smallest among the computers that returned the response with the own computer. Of the computer having the identification number of M, and the changing unit 1 of the computer M having the smallest identification number
5 has a function of controlling to notify "role = master" and "number of times = X + 1". Further, the selection unit 14 notifies the computers other than M including itself of “role = NOT master” and “number of times = X + 1”.

Further, when the changing unit 15 receives role and frequency information from another computer and sets the content in the role storage unit 11 and the role renewal decision processing frequency storage unit 12 of its own computer according to the received content. The above-mentioned processing is performed.

When the confirmation unit 16 is newly elected as the master, after a certain time has elapsed, it issues an inquiry to the response unit 13 to check the status of each of the other computers, waits for a response, and obtains the inquiry. The presence / absence of a reply having the same role as the self and the number of times of information is checked from the reply information, and the presence / absence of conflict, that is, whether or not the self is the only master is checked. If there is a conflict, compare the number of times information to see if the master role was assigned this time or in the past, and cancel it if the master role assignment was done in the past. Then, the master role is assigned to the master role assigned this time. Further, when there is a conflict between the master roles assigned this time, the selecting unit 14 is activated again to restart the master selection. The confirmation unit 16 has such a function. Since the processing unit 17 is a unit that performs actual processing and outputs as described above, and detects a failure of the master, it is connected to the intra-system communication line 2 and the output line 3.

If the content of the role storage unit 11 of the computer itself is "master", the processing unit 17 outputs from the output line 3 to another system at the time of output. Further, in the normal processing, when communication with another computer is required, the communication is performed by the intra-system communication line 2. If the content of the role storage unit 11 is "NOT master" and it is found that the output to another system is necessary, the output line 3 is monitored, and the master which should exist in the system outputs. Make sure you do. If no output is made, the self-selection section 1
4 is started and the operation of selecting a new master is performed.

FIG. 3 shows the operation procedure of the selection unit 14. In this example, it is assumed that each computer 1A, 1n has a unique identification number in the system, but of course, if there is a mechanism that can recognize each computer 1A, 1n without such an identification number, It doesn't matter.

When the master fails, the processing unit 17 of its own computer detects it and activates the selecting unit 14 as described above. However, it does not matter even if another computer detects a failure of the master and the selection unit 14 of the own computer is activated accordingly.

When activated, the selecting unit 14 first makes a "state inquiry" to the response unit 13 of each computer other than itself.
Is transmitted via the intra-system communication line 2 (S1). In the computer which has made the "state inquiry", the selection unit 14 waits for a fixed time for the reply to the inquiry to be returned (S2), and then receives and checks them (S3).

Here, it is assumed that the computer, which is the failed master, does not reply because it has failed. Election section 14
Searches for a response having the maximum number of times X among the received responses (S4). Also, the computer that has the smallest identification number is searched for among the computers that have returned a reply from the self computer (S5). Then, the selecting unit 14 causes the changing unit 15 of the computer M having the smallest identification number to "role = master",
The number of times = X + 1 is notified. In addition, “role = NOT master”, “number of times = X + 1” for computers other than M including self
Is notified and the process ends (S6, S7).

FIG. 4 shows the operation procedure of the response section 13 and its order relationship. When the reply section 13 receives an inquiry, it sends a reply to the computer that issued the inquiry. Upon receiving the inquiry (S11), the response unit 13 of each of the computers 1A to 1n determines the current role of itself and the number of times the role is determined by the number of times the role assignment has been updated by the computer. (That is, the value of the role renewal determination processing frequency storage unit 12) is returned to the selecting unit 14 of the computer that made the inquiry (S12). At this time, an identifier that can identify the own computer as a response source is added. If it corresponds to the example of FIG. 4, this is the only identification number in the system assigned to the own computer.

FIG. 5 shows an operation procedure of the changing unit 15 and its order relation. The changing unit 15 receives a role and the number of times to be set in its own role storage unit 11 and role update determination processing number storage unit 12 from any of the computers, and changes the role according to the contents. When the role and the number of times are received (S21),
It is determined whether or not this received content meets the conditions (S22).
If the content of the number-of-times information is not larger than the value of the own role renewal decision processing number storage unit 12 among the received information, it is the notification by the selection made in the past than the selection which determines the current one's role. Therefore, the changing unit 15 ends the process without doing anything.

On the other hand, as a result of the judgment in the step S22, if the received number is larger than the value of the own role renewal decision processing number storage section 12, it is judged from the selections executed after the present selection of the own role. Therefore, the latest role is changed so that the contents of its own role storage unit 11 and role update determination processing number storage unit 12 are changed to those received (S23). Then, it is checked whether or not the role is the master (S24), and if it is not the master, the process ends. However, if the role is the master, the confirmation unit 16 is activated after a certain period of time, and then the process ends (S25). ).

FIG. 6 shows the operation of the confirmation section 16. Confirmation part 1
6 performs almost the same operation as the selecting unit 14 described in FIG. That is, when the content of the role storage unit 11 is not “role = master”, the confirmation unit 16 ends the process, and when the content is “role = master”, the confirmation unit 16 proceeds to the process of step S32 and subsequent steps (S31). That is, in step S32, the new master computer is newly elected as a master, and after a lapse of a certain time, the new master computer sends an inquiry to the state of each computer by the response unit 1
Send to No. 3 and wait for a reply (S33). Here again, the selection unit 14
Similarly, no response will be returned from the failed master.

When a reply is received (S34), in the reply,
There is a number whose number is larger than the value of its own role renewal decision processing number storage unit 12, or the number of times is the same as the value of its own role renewal decision processing number storage unit 12, and the role becomes the master. It is confirmed whether or not there is a response i having the content (S35).

As a result, if there is no such response i, the self is the latest and the only master, so the process proceeds to step S38, and "role = NOT master", "number = own role" is given to another computer. The change unit 15 is notified of the "value of the update determination processing number storage unit" and the processing is ended (S38).

On the other hand, when there is a reply i satisfying the above conditions, it is checked whether or not the number of times is the maximum and is the only master in the number of times. As a result, if so, the change section 15 is notified to the computers other than the one that has returned the response i including self as “role = NOT master” and “number = number of times of response i”, and the processing ends (S39). If this condition is not satisfied, the master selection unit 14 is activated and terminated (S37).

FIG. 7 shows an embodiment of the master selection procedure according to the method of the present invention. FIG. 7A is an operation transition diagram,
(B) shows the contents such as the state and operation at each time point. It is assumed that the system is composed of four computers A, B, C and D. Further, it is assumed that the computer D is the master computer at first and the master computer D fails at the time t0. The failure of the master computer D is detected by the processing units 17 of the A, B, and C computers (states A0, B0, C0). Then, each computer activates its own selection unit 14, and the selection unit 14 inquires the response unit 13 of each of the other computers about the current state (role and the number of selections when the role is determined).

Thereafter, in order to avoid making the figure complicated,
Only the exchange of messages centered on computer B will be described. In the computer B, when the reply from each of the computers A and C arrives, the selection unit 14 selects a new master on the basis of the reply and the role of the change unit 15 of each of the computers A, B and C including itself. Is notified (state B1).

Here, it is assumed that the computer B is elected as the master. The changing unit 15 of the computer B, which has received the notification to newly become the master, sets its own role as “master” and activates its own confirmation unit 16 after a preset time Tc (time T3). The activation of the confirmation unit 16 causes the confirmation unit 16 to send an inquiry again to another computer.

It is now assumed that the computer D is restored before the time T3. In this case, the response unit 13 of the computer D responds to the inquiry sent by the confirmation unit 16 of the computer B with its own role "master", and the computer B receives this together with the response from other computers. become.

Here, the role of the reply from the computer D is "master", but the number information in the role renewal decision processing number storage unit 12 of the computer D has not been renewed due to the occurrence of a failure and remains as before. Is. Therefore, the count information of the role update determination processing count storage unit 12 of the computer D is smaller than the current latest count information, so that the computer B recognizes that it should be the master, and The change unit 15 of each computer is notified of “NOT master”.

As described above, the present invention allows a plurality of computers to execute the same processing in parallel, and one of them is used as a master and the other is used as a spare. In a system in which one computer is used as a master, each computer determines the role storage unit that stores whether or not it is the master and the role that is stored in the role storage unit. If the master memory is unable to output for some reason, it stores the role of each calculation subject at that time and the role of each computing entity in the number of times of selection. The appropriate number of 1 is obtained by inquiring the other calculation subject about the number of times indicating
A selection unit that selects one as a master, a response unit that replies the status of itself to the above inquiry, and a change unit that changes the role of itself upon receiving a notification of the selection result of the selection unit,
It is characterized by performing a master determination process with a confirmation unit which again inquires and confirms the state of each calculation subject that the calculation subject that has become the master by the above selection unit does not really exist another master after a certain period of time elapses. To do.

In such a configuration, each computer stores information on the role of whether or not it is a master in the role storage unit, and the number of renewal times for which the role renewal determination processing is performed for storing this role information. Is stored as the number of times information in the number of times storage section, and it is possible to reply to the current state of itself including the role information and the number of times information in response to an inquiry. The known computer makes an inquiry to each computer, receives the information on the role of each computer and the information on the number of times obtained as a result, and selects the optimum one as the master. And
The number of times of role determination is updated to obtain new number of times information, and the role information of each computer determined by the above selection and the updated number of times information are output, and each computer outputs the determined number of times. The role information and the new count information of the own computer are received, and the stored contents of the role information and the count information of the own computer are updated. Also, when the self-computer is selected as a new master, the above-mentioned inquiry is executed again to confirm the master conflict.

As described above, according to the present invention, every time the role of another computer is renewed due to a master failure, the number of renewal of roles is updated and held as an accumulated value, so that the information on the number of renewals of each computer is compared. By doing so, it is possible to avoid waste of master re-election processing due to occurrence of competition due to the existence of an old master, and it is possible to improve the efficiency of the system.

[0046]

As described above, according to the present invention, when an abnormality occurs in a master and another master is selected before the confirmation processing is executed after the master selection, the original computer is confirmed by the confirmation processing. Does not become a master, and unnecessary selection work can be omitted while guaranteeing the reliability of operation.

[Brief description of drawings]

FIG. 1 is a block diagram showing the overall configuration of an embodiment of the present invention.

FIG. 2 is a block diagram of each computer 1A, 1B, ˜1 in FIG.
The block diagram which shows the internal structure of n.

FIG. 3 is a flowchart showing an operation procedure of a selection unit 14 in FIG.

FIG. 4 is a flowchart showing an operation procedure of a response unit 13 in FIG.

5 is a flowchart showing an operation procedure of a changing unit 15 in FIG.

FIG. 6 is a flowchart showing an operation procedure of the confirmation unit 16 in FIG.

7A and 7B are diagrams illustrating an example of a master selection procedure according to the method of the present invention, in which FIG. 7A is an operation transition diagram, and FIG. 7B is a diagram showing contents such as states and operations at respective time points.

[Explanation of symbols]

1A, 1B, to 1n ... Computer, 2 ... In-system communication line, 3 ... Output line, 11 ... Role storage unit, 12 ... Role renewal process decision number storage unit, 13 ... Response unit, 14 ... Selection unit, 15
... change unit, 16 ... confirmation unit, 17 ... processing unit.

Claims (3)

[Claims] 1. A plurality of computers perform the same processing in parallel, and one of them is used as a master and the other is used as a spare. When the master fails, one of the spare computers is used as a master. In the system that is operated as described above, each computer has a role storage means that stores information on the role of whether or not it is a master, and the number of times the roles are renewed stored in the role storage means. In the case of master failure, the number of times storing means for storing as information is inquired to each computer, and the optimum one is selected as the master by receiving the information of the role of each computer and the number of times information obtained as a result, Further, the number of times of role determination is renewed, the number of times information after renewal is used as new number of times information, and a selection means for outputting the information of the role of each computer determined by the above selection and the new number of times information. In response to the above inquiry, a reply means for replying the present status of itself including the role information and the number of times information, and the notification of the role information and the new number of times information for the own computer determined by the selecting means, A master-slave comprising a changing means for updating the contents stored in the role storing means and the number-of-times storing means, and a confirming means for executing the inquiry again when the own computer is selected as a new master. Computer system. 2. A plurality of computers perform the same processing in parallel, one of them is used as a master and the other is used as a spare, and when the master fails, one of the spare computers is used as a master. In the system that is operated as described above, each computer has a role storage unit that stores information on the role of whether or not it is a master and a role that is stored in the role storage unit. The number of times storage means for storing the number of times information as to whether it has been determined in the role determination process, and in the case of a master failure, inquires of each computer, and the resulting role information and number of times information of each computer is obtained. Upon receipt, the most suitable one is selected as a master according to these information and the priority of each computer, and the number of times of role determination is renewed and the number of times after renewal is used as new number of times information. Together, a selection means for outputting the role information and new count information of each computer determined by the above selection, and a response means for replying to the inquiry about the current status of itself including the role information and the count information. A change means for updating the contents stored in the role storage means and the number-of-times storage means of the own computer upon receiving the notification of the role information and the new number of times information for the own computer determined by the selection means; A master-slave computer system, comprising: a confirmation unit that re-executes the inquiry when the master-slave computer is selected. 3. The same processing is executed in parallel on a plurality of computers, and one of them is used as a master and the other is used as a spare, and when the master fails, one of the spare computers is used as a master. In the system that is operated as described above, each computer has a role storage step for storing role information indicating whether or not it is a master, and the number of renewals of roles stored in the role storage step as number information. The step of storing the number of times to store, and in the case of a master failure, make an inquiry to each computer and receive the above-mentioned role information and number of times information of each computer obtained as a result, and according to this information and the priority order of each computer The step of selecting one as a master, updating the number of times of role determination to obtain new number of times information, and each calculation determined by the above selection Of the role information and the updated number of times information, a reply step of replying to the inquiry about the current state of the device including the role information and the number of times information, and the determined and output own computer A change step of updating the stored contents of the role information and the frequency information of the own computer in response to the role information and the new number information, and a confirmation step of executing the above inquiry again when the own computer is selected as a new master. A master selection control method for a master-slave computer system, comprising: