JP2001229145A - Method and device for data communication between processors - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the reliability of the communications of an inter-processor data communication device. SOLUTION: The inter-processor data communication device which sends and/or receives unit data by using a communication function having a non- exclusive transfer instruction by which an instruction other than a transfer instruction is executed during the execution of the transfer instruction is equipped with a time-out monitor means which monitors a time-out state as to the transmission and/or reception of the unit data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inter-processor data communication method, which can be used, for example, when data is transferred between a plurality of processors using a non-blocking transfer instruction of an MPI (Message Passing Interface) function. It is.

[0002] The present invention also relates to an inter-processor data communication apparatus using such an inter-processor data communication method.

[0003]

2. Description of the Related Art MPI is a message passing library whose specifications are being examined by the MPI Forum.
MP from a program written in n language or C language
It is called as a function of I or a subroutine. M
PI includes one-to-one communication function, collective communication (collection / distribution / integration operation) function, derived data type function, communicator (Communicator)
icator) function, process topology (Process Topolog)
y) There is an advantage that it can support a function, a dynamic process generation / management function, a one-side communication function, an extended collective communication (collective distribution / integrated operation) function, and a parallel input / output function.

[0004] Among them, the communicator function can realize a multiplex communication space which has been difficult to realize with a conventional message passing library, and the process topology function allows a programmer to define a process in an easy-to-understand form. .

Such an MPI function has a non-blocking transfer instruction as another feature.

[0006] The non-blocking transfer instruction is, for example, an arithmetic processing requiring an enormous amount of arithmetic processing is shared by a plurality of processors in a network, and distributed processing (or parallel processing) is performed.
This is advantageous in cases such as processing.

[0007] Due to the non-blocking nature of the transfer instruction,
Each of the processors involved in the distributed processing can execute other processing even before the started transfer instruction is completely completed. By applying the calculation shared by the processors among them, each processor can execute the message transfer processing even while executing the calculation, and can execute the calculation even while executing the message transfer processing.

FIGS. 2A and 2B show a flowchart of only the communication portion for executing the non-blocking transfer instruction. In this transfer instruction, as shown in FIG. 3, message transfer between the processor 10 and the processor 11 in the network 8 is executed.

FIG. 2A shows the operation in the processor 10, and FIG. 2B shows the operation in the processor 11. The flowchart of FIG. 2A includes steps S1 to S5, and the flowchart of FIG.
To S9.

In order to transmit (or receive) a message in the processor 10, as shown in FIG. 2A, for example, function names MPI_Isend, MPI_Irecv
Is executed (S1), the processor 11 also receives (or transmits) the message, so that the function names MPI_Irecv, MPI_Isend
(S6).

Next, the processor 10 executes another process different from the non-blocking transfer instruction, for example, the above-mentioned shared operation (S2). Similarly, the processor 11 performs another process different from the non-blocking transfer instruction, for example, The assigned calculation is executed (S7).

Then, step S3 following step S2
Then, the processor 10 executes, for example, the function name MPI_Tes.
In step S8 following step S7, the processor 11 checks, for example, the function name MPI
Wait for transfer completion corresponding to _Wait.

In the transfer confirmation processing in step S3, the MPI
When the function of _Test is called, a status indicating whether or not the transfer is completed is set, and the function immediately returns to the function call side. If the function of MPI_Wait is called here, as soon as the transfer is completed, the function returns to the function calling side.

MPI is an application interface and MPI for applications that use MPI.
Merely specifies the function of each function in the function library that should be prepared, so various contents may be considered as the specific contents of each process such as the transfer confirmation process as long as they do not violate the specifications. But not mentioned here.

When there is no problem in the operation of the network and each processor, and there is no failure, the processor 10 performs a status check following a one-shot transfer confirmation (S3) (S4), and completes the communication part normally. This is performed (S5). Even after the normal termination, the processing such as the above-mentioned shared operation can be continuously executed.

The status check (S4) performed after the transfer confirmation (S3) is performed because the transfer confirmation (S3) simply outputs whether or not the transfer has been completed.
To know if the termination was successful,
This is because it is necessary to confirm the contents of the status by the status check.

Similarly, in the normal state, the processor 11
However, after a short transfer completion wait (S8), normal termination of the communication part is performed (S9).

Subsequently, when a message other than the above message is transferred, the same processing as that in the flowcharts of FIGS. 2A and 2B is repeated in each of the processors 10 and 11.

If it is assumed that a transfer instruction or the like has a blocking property, each processor completely occupies the hardware resources for the operation until the operation of the shared operation is completed. Then, no other processing can be executed. That is, the transfer instruction cannot be executed until the operation to be shared is completed, and the operation cannot be executed until the transfer instruction is completely completed.

From this, the usefulness of the non-blocking transfer instruction can be understood.

[0021]

When there is a problem in the operation of the network or each processor and a failure occurs, the step S
The transfer confirmation of No. 3 may be sorry only for one shot, that is, once, and the transfer completion wait (S8) may be sorry in a short time.

In the non-blocking transfer of the MPI function, there is no concept of a timeout, and transfer time management and time monitoring cannot be performed. In the worst case, the transfer confirmation in the step S3 is repeated endlessly. A situation in which the waiting for the transfer completion in step S8 is unlimitedly long may occur.

In this case, the next transfer is kept waiting until the transfer is completed, and communication is stagnated. Also,
Regarding the transfer completion wait (S8), when the end status cannot be obtained due to some failure, the program is in a stopped state, and the system is stopped.

After all, the non-blocking transfer command having such a problem has the above-mentioned great advantage, but it can be said that the communication efficiency is low as a whole, the usability is low, and the reliability is low.

[0025]

According to a first aspect of the present invention, there is provided a communication function having a non-exclusive transfer instruction capable of executing an instruction other than the transfer instruction during execution of the transfer instruction. A method for transmitting and / or receiving a unit message by using an inter-processor data communication method, characterized in that a timeout is monitored for the transmission and / or reception of the unit message.

In the second invention, a unit message is transmitted and / or received by using a communication function having a non-exclusive transfer instruction capable of executing an instruction other than the transfer instruction during execution of the transfer instruction. An inter-processor data communication apparatus according to claim 1, further comprising timeout monitoring means for monitoring a timeout with respect to transmission and / or reception of said unit message.

[0027]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (A) Embodiments First and second embodiments of the inter-processor data communication method and apparatus of the present invention will be described below.

(A-1) Configuration of the First Embodiment FIG. 5 shows the internal configuration of the processor 30 of the present embodiment.
The processor 30 is the same as the processor 10 of FIG.
Is the processor corresponding to
Communicate with A network including the processors 30 and 31 is referred to as a network 9.

In FIG. 5, the processor 30 includes a transfer processing unit 21, a timeout monitoring unit 22, and a transfer termination unit 23.

Of these, the transfer processing unit 21 is the same as that shown in FIG.
This is a portion having a function of executing a non-blocking transfer operation on the transmission side similar to the flowchart of FIG.

Note that the transmitting side performs the processing of FIG.
Performing the processing in FIG. 2B on the receiving side is an example of a system in which distributed processing is performed and each processing operates in an input-driven manner, and is not necessarily required in the MPI standard.

When a non-blocking transfer instruction of MPI is performed, activation (MPI_Isend, MPI_Irec
v) and completion of the transfer (MPI_Test or MPI
_Wait), the MPI_Isend transfer confirmation includes the MPI_Te
st may be used, and MPI_Wait may be used.

Similarly, for MPI_Irecv,
Not only wait for transfer completion by MPI_Wait,
Confirmation of completion by I_Test may be performed. Where M
When PI_Test is used, it is equivalent to simply inquiring about the result of the transfer, so that an inquiry (that is, a call to MPI_Test) must be made until the transfer is completed.

If the completion of the transfer cannot be confirmed at the transfer confirmation in step S22 (see FIG. 1) (this process is the same as the transfer confirmation in step S3), the timeout monitor 22 Is the part that executes

In the timeout check process, a difference between the transfer start time and the current time is obtained, and when the difference is equal to or longer than a preset timeout time, it is determined that a transfer timeout has occurred. If it is determined that a transfer timeout has occurred, the timeout monitoring unit 22
Outputs the timeout detection signal TD to the transfer termination unit 23.

The transfer start time required for the timeout monitoring unit 22 to obtain the difference can be obtained by checking the time data 25 included in the data structure 26 shown in FIG.

The data structure 26 shown in FIG. 4 is a handshake structure (here, a handshake structure passed between the data transfer activation process and the data transfer confirmation process in the same processor for associating the transfer activation and the confirmation. "handshake"
Indicates a handshake between programs in the same processor), the request unit 24 and the time data unit 2
And 5.

The request section 24 includes the MPI_I
It is a portion including a data buffer address, a transfer word number (communication word number), a data type, a transfer destination address (MPI rank), a communication request handle, and the like used as arguments of MPI functions such as send and MPI_Irecv.

In the communication using the MPI function, it is necessary to specify the communication partner and the number of communication words for each communication. Therefore, each of the processors 30 and 31 recognizes the address of the communication partner and the number of communication words prior to the communication. However, this recognition is not conscious of the application using the MPI, but should be realized by a method other than the message transfer by the MPI function.

For example, from the processor 30 to the processor 3
For example, when a message is transmitted to the transmission source 1, the transmission function of the transmission source processor 30 performs the handshaking between the processors prior to the transmission of the message, and the address of the transmission destination processor 31 and the address of the transmission destination of the message are transmitted. It is necessary to specify the number of communication words, and to specify the address of the processor 30 as the transmission source and the number of communication words of the message in the reception function of the processor 31 as the transmission destination.
It is considered necessary to perform the method by a method other than the message transfer by the PI function.

If the MPI communication is performed up to the inter-processor handshake, there is a contradiction that further preceding MPI communication is required to transmit the address of the communication partner and the number of communication words which are the premise of the inter-processor handshake. Because.

Accordingly, a specific example of control communication such as handshaking between processors is contrary to the spirit of the present embodiment, which intends to cover the entire range of the MPI standard as it is, and will not be described in detail here.

The transfer processing section 21 of the components of the processor 30 shown in FIG. 5 is a section for taking out the argument from the request section 24, setting the argument in the MPI function, and activating the transfer.

Further, the timeout monitoring unit 22 adds the time data 25 indicating the communication start time corresponding to the argument to the request unit 24 from which the transfer processing unit 21 has extracted the argument, thereby creating the data structure 26. It has the function of generating and storing.

The transfer termination unit 23 outputs a transfer termination signal CL upon receiving the timeout detection signal TD from the timeout monitoring unit 22, and causes the processor 30 to terminate the transfer process (message transmission process). . As a result, it is possible to prevent communication from stagnating due to endless repetition of transfer confirmation as in step S3.

In the case of the present embodiment, it is assumed that the functions of the processor 31 on the receiving side are the same as those of the processor 30 on the transmitting side shown in FIG. Therefore, the internal configuration of the processor 31 is the same as the configuration shown in FIG.

The only difference is that the contents of the transfer processing performed by the transfer processing unit 21 are transmission processing in the transmission processor 30 and reception processing in the reception processor 31.

Hereinafter, the operation of the present embodiment having the above configuration will be described.

(A-2) Operation of the First Embodiment FIG. 1 shows the processing of the communication part of the processor 30 for transmitting a message. FIG. 1 also shows the processing of the communication part of the processor 31 that receives the message. The flowchart of FIG. 1 includes steps S20 to S27.

The content of the processing in step S20 may be the same as that in step S1.
21 may be the same as step S2, step S22 may be step S3, step S23 may be step S4, and normal termination S25 may be the same as step S5.

The network 9 and the processors 30 and 31
In the normal case where there is no problem in the operation such as the operation and no failure occurs, in an ideal case, this flowchart is usually composed of steps S20, S21, S22, S23 and S2.
4. The process proceeds in the order of S25 and ends normally.

At this time, in step S22, the transfer confirmation processing is performed only once (the transmission confirmation processing
The reception-side processor 31 performs a reception confirmation process), and confirms the completion of the transfer in that process.
It will branch to the (Yes) side.

However, in step S24 arranged between steps S23 and S25, transfer confirmation (S22)
If transfer completion cannot be confirmed at the time of
The process branches to o) and the process proceeds to step S26.

In the process of step S 26, the timeout monitoring unit 22 sends the time data 25 of the data structure 26
To determine the transfer start time, take the difference between the transfer start time and the current time, and determine whether the difference is equal to or greater than a preset timeout time.

If it is determined that the difference is equal to or longer than the timeout period, step S26 branches to the Y side and the process proceeds to step S27. If it is determined that the difference is less than the timeout period, the process proceeds again. The process proceeds to step S22. This is a procedure in anticipation of the possibility that transfer completion may be seen within a normal range within the timeout period even if transfer completion cannot be confirmed in the first transfer confirmation (S22).

In step S27, the transfer termination unit 23
Cancels the transfer request, that is, aborts the transfer process.

Although it depends on how long the timeout time is set, usually, even if the transfer completion cannot be confirmed by one transfer confirmation (S22), the transfer is not immediately terminated (S27). , Steps S22, S23, S2
4. The loop composed of S26 is repeatedly processed,
If transfer completion is confirmed (S22) within the timeout period, the process ends with normal termination (S25), and if not, it ends with transfer termination (S27).

By terminating the transfer, the processor 3
The processing of the communication part by 0 and 31 may start the transfer processing of the next new message (the destination of this message may be a processor other than the processor 31) irrespective of the transfer completion of the message forever. it can.

Note that M shown in step S27 of FIG.
PI_Cancel is a function name indicating cancellation of a transfer request.

In the receiving processor 31, similarly to the transmitting processor 30, transfer confirmation (reception confirmation, reception completion wait) is repeated until the time is less than the timeout time, and transfer (reception) is terminated when the time exceeds the timeout time. Can receive a message other than the message (the source of the message may be a processor other than the processor 30).

(A-3) Effects of the First Embodiment According to the present embodiment, a non-blocking transfer instruction of an MPI function, which conventionally could not perform the next transfer processing until a certain transfer processing was completed, is executed. Even if it is used, the next transfer process can be performed after the timeout period, so that communication stagnation is prevented, smooth data (message) transfer is performed, overall communication efficiency is improved, and usability is improved.
The reliability of communication can be improved.

Further, in this embodiment, by performing timeout monitoring, it is possible to determine the normality of a transfer path such as a network or a processor of a communication partner.

(B) Second Embodiment (B-2) Configuration and Operation of Second Embodiment FIG. 10 shows the internal configuration of the transmitting processor 40 of the present embodiment, and FIG. Receiving processor 50
2 shows the internal configuration of FIG. The processors 40 and 50 are included in the network 60 of the present embodiment shown in FIG.

In FIG. 10, the processor 40 includes a transfer processing unit 41, a timeout monitoring unit 42, a transfer termination unit 43, and a retransmission processing unit 44.

The transfer processing section 41 corresponds to the transfer processing section 21 of the first embodiment. Similarly, the timeout monitoring unit 42 corresponds to the timeout monitoring unit 22, and the transfer termination unit 43 corresponds to the transfer termination unit 23.

Therefore, the substantial difference in operation between the processor 40 of the present embodiment and the processor 30 of the first embodiment is limited to the portion related to the retransmission processing unit 44.

The operation of the processor 40 is shown in the flowchart of FIG. The flowchart of FIG.
38, each step S30 indicating transfer start (transmission start) corresponds to step S20 in the transmitting processor 30 of the first embodiment.

Similarly, step S31 is the same as step S31.
Step S32 corresponds to Step S22, Step S33 corresponds to Step S23, Step S34 corresponds to Step S24, Step S35 corresponds to Step S25, and Step S36.
Corresponds to step S26, and step S37 corresponds to step S27.

Therefore, in FIG. 8, only step S38 is a step unique to this embodiment.

After all, the processor 40 of the present embodiment and the first
In the configuration of the processor 30 of this embodiment, the substantial difference in operation is limited to the portion related to the retransmission processing section 44 and the retransmission process S38.

On the other hand, in FIG.
Has a transfer processing unit 41, a timeout monitoring unit 42, a transfer termination unit 43, a synchronous data check unit 51, and a discard processing unit 52.

The functions of the portions denoted by the same reference numerals as those in FIG. 10 are the same as those in FIG. This is part 4
1, 42, and 43.

The operation of the processor 50 is shown in the flowchart of FIG. The flowchart of FIG.
The step S40 indicating the transfer start (reception start) corresponds to step S20 in the reception-side processor 31 of the first embodiment.

Similarly, step S41 is performed in step S41.
21, step S42 corresponds to step S22, step S43 corresponds to step S23, step S44 corresponds to step S24, step S45 corresponds to step S25, and step S46 corresponds to step S46. Step S47 corresponds to S26.
Corresponds to the step S27.

Therefore, in FIG. 9, only step S48 is a step unique to this embodiment.

After all, the processor 50 of the present embodiment and the first
In the configuration of the processor 31 of the third embodiment, the substantial difference in operation is limited to the synchronous data check unit 51, the discard processing unit 52, or the part related to the synchronous data check processing S48.

In the following, only differences between the present embodiment and the first embodiment will be described.

In the transmission-side processor 40, the retransmission processing section 44 operates to perform the retransmission processing in step S38 after the transfer termination processing in step S37 is performed.

That is, the difference between the transfer start (transmission start start) time and the current time is calculated, and the difference is equal to or longer than a preset timeout time, and it is determined that a transfer timeout has occurred. This is the case.

When occurrence of a transfer timeout is recognized in the transmitting processor 40, it is highly likely that a normal message has not been received by the receiving processor 50 in the previous transfer, so that a message having the same content is re-transmitted by the retransmission processing S38. Transmit and ensure communication reliability.

On the other hand, in the receiving processor 50, the synchronous data check unit 51 operates to perform the synchronous data check processing in step S48.
This is to prepare for the reception of the message transmitted by the transmission-side processor 40 by 37.

In the present embodiment, a message (transfer text) ME transmitted from the transmitting processor 40 and received by the receiving processor 50 is, as shown in FIG. The value of the synchronous data SD having the structure with the added SD is changed based on a predetermined regularity every time the transfer processing unit 41 of the transmitting processor 40 transmits the message ME having the new transmission data DC. . In the present embodiment, this change operation is assumed to be an increment (+1).

Therefore, when normal transmission and reception are performed, the synchronization data SD of the time-series message ME transmitted and received is sequentially incremented. However, when retransmission and reception thereof are performed, the synchronization data SD having the same value is obtained. The message ME having the SD is subsequently transmitted and received.

In the receiving processor 50 that receives these messages ME, the synchronization data check unit 51 stores at least the value of the synchronization data SD relating to the message ME received last time.
The value of the synchronous data SD of the message ME received this time is compared, and if they are the same, a coincidence signal indicating the coincidence of the synchronous data SD is transmitted to the discard processing unit 52.

The discard processing unit 52 that has received the coincidence signal discards the contents (data DC) of the previously received message ME stored in the transfer processing unit 41. Performing such a discarding process is likely to have a problem in the transfer route of the message ME when a timeout occurs,
This is because the possibility that a transmission error has occurred in the data DC is considered to be higher than when no timeout occurs.

In order to explain the operation of the present embodiment, a case where the transfer processing as shown in FIG. 7 is performed between the processors 40 and 50 will be described as an example.

In FIG. 7, in the transfer of the sequence SE1, the value of the synchronous data SD is set to “1” and the message ME is set.
No timeout is detected by either the processor 40 on the transmission side or the processor 50 on the reception side.

In this case, the flowchart of FIG. 8 ends with the normal end S35, and the flowchart of FIG. 9 ends with the normal end S35.
After 45, the synchronous data check processing S48 is performed and the processing ends. In the current synchronization data check process S48, a value different from the “1”, for example, “0” is detected as the value of the synchronization data SD of the message ME received last time. The unit 52 does not operate.

Therefore, the transmission-side processor 40 starts transmission activation processing S30 for transmitting the next message ME.
Is performed, and the next message M
A reception start process S40 for receiving E is performed.

Next, in the message transfer of the sequence SE2-1, a timeout is detected on the transmission side for the transfer of the message ME in which the value of the synchronous data SD is "2", and the transfer termination processing S37 is performed.

At this time, since no timeout has been detected in the receiving processor 50, the synchronous data check processing S48 is performed following the normal end processing S45 as in the previous case, and the message ME with the value of the synchronous data SD of "2" is executed. Is received, and reception start processing S40 for receiving the next message ME is performed.

In the sequence SE2-2 following the sequence SE2-1, the retransmission processing unit 44 of the transmitting processor 40 functions to perform the retransmission processing S38 of the message ME whose synchronous data SD is “2”.

The retransmission processing S38 itself has the same logical structure as the entire flow chart of FIG. 8, and if necessary, the retransmission processing may be repeatedly performed two or more times.

Assuming that the message ME retransmitted in step S38 is normally received on both the transmitting side and the receiving side without detecting a timeout, the receiving processor 50 checks the synchronous data after the normal end processing S45. In process S48, the message ME received this time
Of the synchronous data SD is "2", which is the same as the previously received message ME, and the transmitting side performs retransmission processing S3.
Recognizing that step 8 has been performed, the discard processing unit 52 is operated.

The discard processing unit 52 discards the message ME whose previously received synchronous data SD is “2”, and treats the message ME whose received synchronous data SD is “2” as a genuine received message.

In the next sequence SE3-1, the timeout of the transfer of the message ME whose synchronous data SD is "3" has been detected by the receiving processor 50, so that the transmitting processor 40 has also successfully completed the transmission in step S33. Cannot be confirmed, and a timeout is detected on both the transmitting side and the receiving side.

At this time, the processing of the transmitting processor 40 ends with the transfer abort processing S37, and the processing of the receiving processor 50 also ends with the transfer abort processing S47.

In the sequence S3-2, the transmission-side processor 40 that has completed the previous message transmission in the transfer termination processing S37 executes the same retransmission processing S47 of the message ME whose synchronous data SD is “3” as the previous time.

This message transfer is normally completed on both the transmitting side and the receiving side without detecting a timeout (S3
5, S45), the synchronous data check processing S48 is executed on the receiving side, but the timeout has been detected on the receiving side the previous time, so that the synchronous data SD of “3” is received by the receiving side processor 50. Message M
This is the first reception of E.

Thereafter, the same operation as each sequence in FIG. 7 can be repeated.

(B-2) Effects of the Second Embodiment According to the present embodiment, the same effects as those of the first embodiment can be obtained.

In addition, in the present embodiment, by performing retransmission at the time of transfer timeout, it is possible to increase the reliability of data transfer in the sense that the frequency of occurrence of data (message) that cannot be transferred can be reduced. It is.

(C) Other Embodiments In the second embodiment, the retransmission processing S38 itself has the same logical structure as the entire flow chart of FIG. 8, and if necessary, repeats the retransmission processing at least twice. Is executed, but it is not necessary to do so.

For example, the message transmission is performed uniformly or once a predetermined number of times without performing the transfer confirmation S32 or the status check S33.
38.

Depending on the type of failure that occurs in the network, the processor, or the like, there may be a case where a timeout is detected even if retransmission is repeated many times. In such a case, repeating the retransmission itself will solve the invention. This is because there is a possibility that substantially the same problem as the problem such as communication stagnation or system stoppage described in the section of the problem will be caused.

In the first and second embodiments, the number of processors included in the network is two. However, it is natural that the number of processors can be increased to three or more. .

Further, the present invention can be applied to a standard in which the main part is the same as the MPI, but partially different.

That is, the present invention relates to a method for transmitting and / or receiving unit data using a communication function having a non-exclusive transfer instruction capable of executing an instruction other than the transfer instruction during execution of the transfer instruction. The data communication method and apparatus can be widely applied.

[0109]

As described above, according to the present invention, the communication efficiency can be improved, the usability can be improved, and the communication reliability can be improved while enjoying the advantages of the non-exclusive transfer instruction. .

[Brief description of the drawings]

FIG. 1 is a flowchart showing an operation of the first embodiment.

FIG. 2 is a flowchart showing the operation of a conventional processor on the transmitting or receiving side in a network.

FIG. 3 is a schematic diagram showing a configuration of a network according to the first and second embodiments.

FIG. 4 is a schematic diagram showing a configuration of a data structure used in the first and second embodiments.

FIG. 5 is a schematic diagram illustrating an internal configuration of a processor according to the first embodiment.

FIG. 6 is a schematic diagram illustrating a structure of a message in the second embodiment.

FIG. 7 is an operation explanatory diagram illustrating an inter-processor communication operation according to the second embodiment;

FIG. 8 is a flowchart illustrating an operation of a transmitting processor according to the second embodiment.

FIG. 9 is a flowchart illustrating an operation of a receiving processor according to the second embodiment.

FIG. 10 is a schematic diagram illustrating an internal configuration of a transmission-side processor according to the second embodiment.

FIG. 11 is a schematic diagram illustrating an internal configuration of a receiving processor according to the second embodiment.

[Explanation of symbols]

8, 9, 60: Network, 10, 11, 30, 3
1, 40, 50: Processor, 21, 41: Transfer processing unit, 22, 42: Timeout monitoring unit, 23, 43: Transfer termination unit, 44: Retransmission processing unit, 51: Synchronous data check unit, 52: Discard processing unit , SD: Synchronous data, DC
... (transmission) data, ME ... message.

Continuing from the front page (72) Inventor Shigeru Minami 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (72) Inventor Akihiko Sugisawa 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry F term in reference (reference) 5B045 BB31 BB42 BB44 JJ05 JJ14 JJ45

Claims (4)

[Claims] An inter-processor data communication method for transmitting and / or receiving a unit message using a communication function having a non-exclusive transfer instruction capable of executing an instruction other than the transfer instruction during execution of the transfer instruction. The method according to claim 1, wherein a timeout is monitored for transmission and / or reception of the unit message. 2. The data communication method between processors according to claim 1, wherein identification information for identifying the difference between the information contents contained in said unit message at the time of reception is added to each unit message. When a timeout is detected by the timeout monitoring unit, a retransmission process of retransmitting the data,
And / or upon receiving, based on the identification information, check for differences in information contents contained in each of the received unit messages, and when a plurality of unit messages containing the same information contents are received, finally, A data communication method between processors, comprising executing a received data selection process for processing only a received unit message as a genuine unit message. 3. An inter-processor data communication device for transmitting and / or receiving a unit message using a communication function having a non-exclusive transfer instruction capable of executing an instruction other than the transfer instruction during execution of the transfer instruction. 2. The inter-processor data communication device according to claim 1, further comprising a timeout monitoring unit that monitors a timeout with respect to transmission and / or reception of the unit message. 4. The interprocessor data communication apparatus according to claim 3, wherein identification information for identifying a difference in information content contained in said unit message at the time of reception is added to each unit message. Retransmission means for retransmitting the data when a timeout is detected by the timeout monitoring means,
And / or upon receiving, based on the identification information, check for differences in information contents contained in each of the received unit messages, and when a plurality of unit messages containing the same information contents are received, finally, An inter-processor data communication device, comprising: a received data selection unit that processes only a received unit message as a genuine unit message.