JPH0573507A - Reliefing device in message communication between electronic computers - Google Patents

Abstract

PURPOSE:To enable relief in such a manner as software at the time of interruption extinction in message transmission by providing an interruption handier which process a message through software with signals from a reliefing task. CONSTITUTION:A message block 1 which makes a message 3a, stored on a shared memory device 7 that plural electronic computers Q and R share and sent from the message transmission part 3 of the electronic computer Q at a transmission source, into blocks and adds a unique index 1c to a header 1b, an entry 2 which tunes the message block 1 to the electronic computer R at a reception destination, and a hardware mechanism which interrupts the electronic computer R at the reception destination are shared. Further, the device is equipped with the remedy task 6 which scans the index 1c in a message block 3a at every constant period and outputs it unless the message block 1 is processed within a predetermined time and the interruption handler 5 which processes the message 3a through the software with the signal from the remedy task 6.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a rescue device for message communication between electronic computers.

[0002]

2. Description of the Related Art Conventionally, in contrast to a centralized processing system in which one computer concentrates and processes all jobs,
Alternatively, the distributed processing method divides one work into smaller units and distributes them to the mutually connected calculation elements. In this case, the calculation elements are the degree of division of work, that is, the distributed processing. Depending on the level, it may be a small computer, a processor having a central processing unit (CPU) and memory, or an arithmetic circuit. According to the distributed processing method, a high-performance computer system can be realized at a relatively low cost.

Regarding the complex computer system,
About several to a dozen or so computers that can operate as a single unit,
It is a computer system that is loosely coupled in a relatively short distance and is also called a multi-computer or a computer complex. It is used to realize communication and information transfer between the computers that make up the computer system. From the standpoint of whether or not there is a method, there is a method that only has a communication means between computers and a method that has a storage medium (file) that can be commonly accessed from a plurality of computers.

Further, with recent advances in hardware technology, inexpensive small-sized computers such as microprocessors and dedicated processors equipped with special arithmetic functions have been easily developed, and processing conventionally performed by large-sized computers has been performed. It is becoming possible to execute the above with a function-distributed complex computer system in which a plurality of processors are combined.

In such a function distribution system,
It is very important to select the configuration method of the complex computer system considering the processing efficiency of the whole system and to allocate the optimal function to each computer. However, a general configuration method and function allocation algorithm have not been established, and a comprehensive data flow is analyzed for each application field and target, and functional division, communication method between computers, interface, software It is necessary to optimize the method of division and scheduling.

Regarding the function of the shared memory, in a tightly coupled multiprocessor, information is generally exchanged between the processors via the shared memory.
The cycle and communication between the processors are realized by reading and writing in a predetermined area of the shared memory. This is because it has the characteristics that it is faster and has less overhead than the communication method by other means.

When the memory is shared by a large number of processors, the access is concentrated on one memory unit, which delays the access time and often becomes a bottleneck of the entire system. To avoid this, by dividing the memory into banks and using cache memory,
This is dealt with by increasing the effective memory access speed.

Further, in order to perform synchronization and communication between processors, a function that supports an essential mutual exclusion function is required.

In a multiprocessor system,
The calculation is carried out by the parallel execution of the tasks assigned to each of the plurality of processors.
In order for parallel tasks to cooperate with each other to proceed with one calculation, it is essential that they can synchronize with each other and communicate (message exchange).

In this respect, the token system is a synchronous system used in the case of a ring bus. Normally, data called a token is circulated, and a processor desiring to send a message first sends a message when the token arrives. Send tokens. In the interrupt method,
The processor that wants to send a message interrupts the other processor, and the interrupted processor must receive it.

Further, the central processing unit has a CP
U is composed of a functional unit generally called an instruction control unit, a storage control unit, and an arithmetic control unit. The instruction control unit reads and decodes an instruction, reads a base address and an index from a general-purpose register, and performs instruction displacement. In addition, the operand address is generated. Further, it is a central functional unit that checks whether or not there is an interrupt factor each time the processing of one instruction is completed, and if there is a factor, performs an interrupt process and controls the progress of the entire CPU.

Next, regarding the activation queuing function,
When encountering during I / O startup, channel use or IOC use, it is necessary to reschedule the I / O path and the trigger of activation.

When executing an input / output instruction in the start-up queuing, the IOP selects a channel path which is a physical route to a target peripheral device only by designating and executing a logical peripheral device number. Start the device. When the selected channel path is in use for other I / O processing,
Select another channel path. Further, when all are in use, the I / O request from the CPU is stuck in the temporary IOP, and the rescheduling of the activation is carried out.

The activation queuing is conventionally executed by the operating system without interrupting the CPU even if the IOP accepts the input / output request from the CPU program even when the channel or the IOC is used. The hardware automatically performs the re-scheduling process for booting, which is effective in reducing the overhead of the operating system.

Conventionally, when transmitting a message to another computer, an interrupt is generated in the destination computer to notify that there is message transmission. At this time, the information was added and the message was notified when the interrupt occurred. However, when the amount of the information becomes large, a method of collecting the message body into blocks and queuing the entries to separate from the interrupt information comes to be adopted. The computer notified of the message transmission by the interrupt cuts out the message block from the entry and receives the message. in this case,
The computer on the receiving side replied back to confirm whether or not the message was transmitted, and the transmitting side confirmed it by retrying due to a timeout.

[0016]

As described above, although a redundant process is required to confirm the completion of message transmission, adding a redundant process may reduce reliability. there were.

The present invention is to provide a rescue device for message communication between electronic computers, which performs a software-like rescue in the computer system collectively when an interrupt disappears when a message is transmitted by an interrupt between computers having a shared memory. To aim.

[0018]

According to the present invention, a message stored in a shared memory device shared by a plurality of computers and transmitted from a message transmission unit of a transmission source computer is divided into blocks, and a unique index is stored in a header. A message block to be added, an entry that tunes this message block to the receiving computer, and a hardware mechanism that causes an interrupt to the receiving computer,
A rescue task that scans the index in the message block at regular intervals and outputs when the message block has not been processed within a predetermined time, and an interrupt handler that processes the message in software by a signal from this rescue task. It is a relief device for exchanging messages between electronic computers.

[0019]

In the rescue apparatus for message communication between electronic computers according to the present invention, the messages sent from the message sending unit of the sender computer are stored in a shared memory device shared by a plurality of computers and are uniquely divided into blocks. Index is added to the header, the message block is tuned to the receiving computer, an interrupt is generated in the receiving computer, the index in the message block is scanned at regular intervals, and the message block is scanned within a predetermined time. Is processed when the message has not been processed, and the message is processed in software by the signal from the rescue task.

[0020]

EXAMPLE An example of the present invention will be described below. Figure 1
The message transmission unit 3 of the transmission source computer Q is stored in the shared memory device 7 shared by the plurality of computers Q and R.
The message 3a sent from is blocked, and the unique index 1c is added to the header 1b, the message block 1 is entered, the entry 2 that tunes the message block 1 to the receiving computer R, and the receiving computer R is interrupted. A hardware mechanism (not shown) that generates a message, a rescue task 6 that scans the index 1c in the message block 3a at regular intervals, and outputs if the message block 1 is not processed within a predetermined time, and a rescue task Message 3a by signal from 6
2 shows a rescue device for message communication between electronic computers, which comprises an interrupt handler 5 for processing the above by software.

That is, the present invention relates to a software rescue method when an interrupt to another computer disappears during data transmission between computers having a shared memory device 7 shared by the computers Q and R. When sending a message between multiple computers that have a memory that is shared by the computers (shared memory), store the message in a block, queue the block in the entry of the destination computer, and assign a unique index to the block. A program to add, and hardware to notify the destination computer that there is a message transmission by interrupt,
It has a task that operates periodically and monitors whether or not the message block is processed. If the message block is not processed even though the sender computer has issued an interrupt notification, the message block can be processed by software. It is a software relief method at the time of disappearance of an interrupt notification to another computer, which is characterized in that there is a notification that there is.

1 is an unused message block,
The message block 1 is prepared separately for each calculation, and 1'shows its detailed data structure.

Reference numeral 2 is an entry for queuing a message block, 2 is an entry for the computer R,
2a is an entry for the computer Q.

Reference numeral 3 denotes a message transmission unit, which reserves an unused message block, sets a message and an index, and tunes to an entry.

Reference numeral 4 is an interrupt generation, which causes the transmission source computer to generate an interrupt.

Reference numeral 5 denotes an interrupt handler corresponding to the interrupt 4, which takes out the message block 1 from the entry 2 and performs a process corresponding to the message.

Reference numeral 6 is a rescue task which is activated at regular intervals, and monitors the transmission status of the message block 1 and substitutes the interrupt handler 5 for software scheduling when necessary.

The present invention is roughly divided into the following three elements. That is, when sending a message from another computer,
The message is grouped into blocks, a unique index for the block is added to the block header, and a software mechanism that tunes to the entry of the receiving computer, a hardware mechanism that causes an interrupt to the receiving computer, and operates at regular intervals When there is a message block in the entry corresponding to the local computer in the task to perform, if the message block is not processed within a certain period of time by looking at the index of the header of that message block, software corresponds to (b). This is a software mechanism for activating an interrupt handler, which has an effect that software can be uniquely relieved when an interrupt disappears in the computer without adding redundant processing on the message sending side and the message receiving side.

Next, a general message transmission operation will be described.

An unused message block 1 is reserved by the program of the message transmitting unit 3, and the message block 1 is tuned to the entry 2 corresponding to the transmitting computer R. At this time, the index 1c unique in the system is added to the header 1b as data for the rescue task 6. Next, an interrupt is generated in the destination computer R, the interrupt handler 5 is activated by the interrupt generation, the message block 1 is secured from the entry 2, and the corresponding processing is performed.

When the transmission / reception of the above message transmission is not completed due to the disappearance of an interrupt or the like, the rescue task 6
The relief method by the method will be described with reference to the flowchart of FIG.

In step 11, this repair processing is performed by a task, and this task is activated at regular intervals using a timer mechanism. In step 12, when activated, it is determined whether a message block is chewed in the entry, and if it is, the index of that block is stored. In step 13, after waiting for a certain period of time, the index is obtained as in step 12. Then, in step 15, step 12,
The indexes fetched in 13 are compared. If the indexes are the same, it is determined that an interrupt has not been notified, and an interrupt handler is scheduled from this task to substitute the interrupt generation processing on another computer.

The present invention can be applied to the notification processing relief of the processing of chewing blocks and passing data.

[0034]

According to the present invention, software relief can be performed when an interrupt disappears when a message is transmitted by an interrupt.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a rescue device showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing the operation of FIG.

[Explanation of symbols]

 1 ... Message block 2 ... Entry 3 ... Message sending part 5 ... Interrupt handler 6 ... Relief task

Claims (1)

[Claims] 1. A message block, which is stored in a shared memory device shared by a plurality of computers, blocks a message transmitted from a message transmission unit of a transmission source computer, and adds a unique index to a header, and this message. An entry that tunes the block to the receiving computer, a hardware mechanism that causes an interrupt to the receiving computer, and scans the index in the message block at regular intervals, and the message block within a predetermined time. Is provided when a message is not processed, and a rescue device at the time of message communication between electronic computers, which includes a rescue task and an interrupt handler that processes the message in software by a signal from the rescue task.