JPH0575141B2 - - Google Patents

Description

[Detailed Description of the Invention] (Industrial Application Field) The present invention relates to a computer system, in particular, controls a plurality of computers connected in a ring shape in relation to each other, and performs function distribution processing, load distribution processing, etc. at will. A computer that can be selected and executed as appropriate.
Regarding the system.

(Prior art) In recent years, multiple processor units (hereinafter referred to as
Attempts have been made to speed up arithmetic processing by distributing one program or processing multiple programs in parallel.

Conventional methods of this type connect multiple PUs via bus lines or communication channels, and
Distributed multi-processor systems configured so that PUs cooperate to process information are common, and these are further classified into function distribution systems and load distribution systems, depending on the processing content assigned to each PU.

The device configuration in the function distribution method is, for example, the fourth
One main processor MPU as shown in Figure a
and a plurality of processors PU ₁ to PU _o connected to this in a star shape, each of the processors PU ₁ to PU _o having a predetermined function. For example, PU ₁ is a printer, PU ₂ is a CRT display device, PU _o
contains a program with functions to control each floppy disk device.

In this system, a program desired by the user is loaded into the main processor MPU and executed, but during the execution, processing is requested to each PU for the functions that each PU has.

On the other hand, the general configuration of the load balancing method is
As _shown in _FIG . n of the program group
This method involves assigning it to a PU, loading it, executing it, and collecting the results in the main processor to obtain the final result.

However, in such conventional distributed processing systems, the system is configured in advance to support either function distribution or load distribution, and it is often difficult to satisfy both at the same time.

In other words, in the function distribution method, most of the functions that can be distributed in advance are limited to input/output processing of peripheral devices of the computer system, as described above.
Although this is effective when distributing a program whose main role is such processing (hereinafter referred to as an IO limit program) to each PU, it is of little use for other distributed processing.

In addition, in the load balancing method, contrary to the function distribution method, programs that are mostly calculations (hereinafter referred to as
For computationally limited programs), this program can be distributed to multiple PUs and executed in parallel, which is effective in improving processing power.
Parallel processing capabilities can hardly be demonstrated for IO-limited programs.

As described above, conventional distributed processing methods using computer systems cannot satisfy both the function distribution function and the load distribution function, and have the disadvantage that depending on the program to be executed, the computer system cannot be used effectively. Ta.

In view of these circumstances, the applicant has already proposed a computer system that can perform both function distribution and load distribution at will (Japanese Patent Application No. 1987-87782). 2 channel processors that independently manage
A plurality of microprocessors (hereinafter referred to as PEs) each having a channel processor are connected in a ring shape through the channel processor.

However, in the system of the above-mentioned application, multiple ring-shaped communication channels are used to send required data to microprocessors assigned distributed processing, or to send and receive signals between multiple required ring-shaped processors. Because this process is performed through a computer, it is difficult to perform high-speed processing beyond a certain limit. That is, according to the computer system of the above-mentioned application, it is possible to cut a single program into several blocks and have each block individually distributed and executed by several processors. Frequently, signals are transferred to one processor and other processors perform further processing based on the results, but all signal exchange between these processors is via adjacent processors connected sequentially via a ring-shaped communication channel. Since the data is transferred to the required one, there is a natural limit to the traffic on the communication channel, and while other data is being transferred, a waiting time is required until the communication channel is freed up. Furthermore, the processing load increases for each PE that passes through sequentially to determine whether or not the received signal is related to the process frame group that it manages, which also has the drawback of hindering faster processing. It was hot.

(Object of the Invention) The present invention relates to an improvement of a computer system already proposed by the same applicant, and enables high-speed processing by directly connecting required PEs among PEs connected in a ring. The system provides a computer system with

(Summary of the invention) In order to achieve this object, the present invention includes:
First, as a basic computer system, a plurality of network processing elements (PE) each having at least four channel processors that independently manage communication channels are connected in a ring shape through the channel processors,
Furthermore, the necessary NPEs are processed and
A manager is designated as a manager and is assigned to manage a certain number of processes/frames attached to them, as well as to manage other PEs.
At least one of the process managers is designated as a master manager that manages the process managers, and in response to a request from the master manager, the process managers allocate processes as appropriate to required ones of the process frames that they manage themselves. The system is characterized in that the system is configured to connect to other PEs via channels other than those required for ring-shaped connections among the channel processors.

(Example) Hereinafter, the present invention will be explained in detail based on an example, but in order to help the understanding of the present invention, the above-mentioned computer system proposed by the same applicant (Japanese Patent Application No. 62
-87782) will be briefly explained.

FIG. 5 is a block diagram showing the principle of a computer system proposed by the same applicant, in which a network processing element (PE) is equipped with at least two channel processors that independently manage communication channels. the channel
One of these PEs is connected in a ring through processors, and one of them is a master manager MM that monitors and controls all the PEs, some of the other PEs are process managers SM ₁ to SM _o , and some of the other PEs are are defined as servers S ₁ to S _r for driving peripheral devices etc. added to the system, and a required number of other PEs are defined as communication managers CM for communicating with other systems, and the above-mentioned process·
Managers CM ₁ to SM _o each have an environment for executing processes (hereinafter referred to as processes and frames)
A process with a certain number of PEs (PF ₁ to PF _k )
It has a frame group and is configured to be monitored, executed, and controlled by each of the process managers SM ₁ to SM _o .

In this configuration, execution of any process is executed by the master manager MM.
The process name and execution command are passed as a message to the managers SM ₁ to SM _o , and the process manager that receives this message sends a command to read the executable program corresponding to each process name to the servers S ₁ to S _r as a message. (in this case, the server is called a file server), and the executable program is sent back from the corresponding file server to each process manager via the network.
The executable program thus returned is allocated to an appropriate process frame PF in the process frame group managed by each process manager and executed.

Also, processes can be stopped by the master manager.
Sends a message from the MM to the process manager with the name of the process to be stopped and a stop command;
Upon receiving this, the process manager stops execution of the process.

Communication between processes running in a given process frame group involves sending and receiving message packets from the sender's process frame through the network, including the sender process name, designated receiver process name, and communication content. It is done by.

If the process corresponding to the specified recipient process name in the message packet is being executed in a process frame group that it manages, the process manager located on the network will Notify the content of the communication.

In response, the receiving process frame returns a response message indicating that the message packet has been received, and communication is then started.

Additionally, if the specified recipient process name in the received message packet does not match any of the process frames it manages, or if the process frame is not being executed, the process manager ignores the message packet and sends the message packet to the adjacent process frame. The message is sent as is to the manager, and each manager sequentially performs such operations until the specified recipient process name contained in the message packet reaches the manager who manages it.

The main program in this system is loaded via the master manager MM.
Also, the communication manager CM in the ring is used when communicating with other computers or other computer systems, and converts the messages of the system into a signal form specific to the other computer system and sends and receives them. It is something that fulfills a function.

According to a computer system configured in this way, it is possible to carry out so-called load distribution processing in which a part of a program is distributed and executed by all PEs (processors) connected in a ring. The server (server) is equipped with an IO interface for peripheral devices such as disks, CRTs, displays, and printers, so it can drive peripheral devices without placing a load on the main processor, and this function is useful for distributed processing. Hit. In other words, all the processors that make up the system can support both function distribution and load distribution, so they can be executed as appropriate without being particularly aware of the nature of the program to be executed, that is, whether it is an IO limit program or a computation limit program. can be partially assigned to multiple processors for execution.

In this way, multiple programs can be used for process processing in cooperation with each other as if they were a single computer system because each processor independently manages the communication channel.
This is because it is equipped with a processor.

However, in this computer system, all signals are connected in a ring.
As mentioned above, since the data is transferred via the processor, there is a limit to the improvement in processing speed due to the traffic limitations of the channel.

Therefore, the present invention is configured as follows.

FIG. 1 is a block diagram showing an embodiment of a PE used in the present invention, in which, for example, four channel processors ch and P are added to a CPU, and these channel processors communicate by themselves. Control and manage the communication channel and read and read message packets or other message contents sent through the communication channel.
Contact CPU.

2 out of 4 channel processors ch.P
One is used to connect to an adjacent PE, and the remaining two channel processors ch.P are used to connect to other PEs (PEs connected to other PEs may also be included).

FIG. 2 is a block diagram showing an embodiment of the computer system of the present invention configured using this PE. Only some of the SMs are extracted.

In the same figure, PE1 to PEn correspond to the CPU, respectively.
A microprocessor consisting of four channel processors CH and P independently managed by the CPU, and these PEs are configured to control two of the channels connected to the channel processor, e.g.
They are connected to adjacent PEs via ch1 and ch2, and are connected in a ring with other PEs (not shown), and the remaining two channels of each PE, ch3 and ch4, are both connected to the crossbar switch CBS. It is connected to the required end of.

Considering the case where the crossbar switch CBS is connected as shown in FIG. 3 in a device configured as described above, the third channel ch13 of PE1 is connected to the fourth channel ch24 of PE2 via the crossbar switch CBS. connected, and the same applies below.
ch23 of PE2 is chn4 of PEn, chn3 of the PEn
are connected to ch14 of PE1 again and PE1,
This means that the three processors PE2 and PEn are connected in a ring.

Therefore, for these three PEs, all PEs
In addition to the basic channel route that is ring-connected to the network, it has the above-mentioned bypass channel route, which enables distributed processing as described below.

That is, first, one program is divided into required blocks via the basic channel route, and each block is divided into
Assign some PEs to execute.

Among PEs to which distributed processing is assigned, PEs that require the processing results of other PEs or those that require commands from the master manager during processing can be connected to each other through the bypass channel route. When transferring data between PEs related to PEs, if the bypass channel route is used instead of the basic channel route, communication etc. will be performed independently regardless of the usage status of the basic channel route. As a result, processing speed can be significantly improved.

It should be noted that the configuration and operation of the crossbar switch are known techniques, so a detailed explanation will be omitted.

The embodiment described above is based on PEs connected in a ring shape.
Although a case is shown in which only some of them are equipped with four channel processors, the present invention is not limited to this, and various modifications are possible.

For example, each PE connected in a ring is equipped with four channel processors, and
All channel processors other than those required for ring connection may be connected via the above-mentioned crossbar switch, or in this case, only the necessary ones may be selectively connected via a bypass ring channel route. It is possible. It will also be clear that more than four PE channel processors may be provided.

Furthermore, in carrying out the present invention, the technology described in Japanese Patent Application No. 87782/1987 filed by the same applicant may be used in combination.

(Effects of the Invention) As explained above, the present invention connects processors each having four or more channel processors that can be independently managed in a ring shape through the channel processors, and further connects the processors that are necessary among them. are selected and interconnected by a bypass channel route different from the ring-shaped route, so regardless of the usage status of the basic channel route when executing the related distributed program. First, communication between the required processors can be performed, which is effective in further improving the speed of distributed processing.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the processor used in the present invention, FIG. 2 is a partial configuration diagram showing one embodiment of the computer system of the present invention, and FIG. 3 is a block diagram showing one embodiment of the computer system of the present invention. Computer/
A configuration diagram to explain the wiring state of the system,
4a and 4b are conceptual diagrams showing the configuration of a conventional computer system, and FIG. 5 is a configuration diagram showing an embodiment of a computer system proposed by the same applicant. 1...CPU, ch/P...channel processor, PE to PEn...channel independently managed
A processor equipped with a processor, CBS...Crossbar switch.

Claims (1)

[Claims] 1. A channel that independently manages communication channels.
A plurality of microprocessors each having at least four or more processors are connected in a ring shape by a communication channel or a bus line through the channel processor, and a required number of microprocessors among the plurality of microprocessors is a fixed number. A process manager that manages a group of process frames, and at least one other is a master manager that manages the process managers, and a process frame that the process manager manages himself/herself in response to a request from the master manager. In a computer system configured to appropriately allocate and execute a process to one of the groups, the required microprocessors can communicate with each other through a channel processor other than the one used to connect the adjacent microprocessor among the channel processors. A computer system characterized in that a bypass route is provided by connecting between. 2. The invention is characterized in that a bypass route can be set between arbitrary processors by inserting a crossbar switch between each channel processor used for the bypass route. A computer system according to scope 1. 3. At least one of the microprocessors connected in a ring shape is a communication manager having a function of communicating with other computer systems, and the microprocessors are connected to other computer systems via the communication manager. A computer system according to claim 1 or 2. 4. The computer described in claim 3.
A computer system characterized in that a plurality of systems are connected in a ring shape with the communication manager as a node. 5. In the computer system according to claim 4, if there is no process frame that can perform process allocation within one computer system, the process frame is transferred to another computer system via the communication manager. 1. A computer system characterized in that, by requesting execution, related program processing can be executed in a distributed manner by all or as required of the plurality of computer systems connected in the ring. 6. Adding the master manager and communication manager to a computer system other than the computer system according to claims 1 to 5, thereby connecting them to each other;
Furthermore, the computer system is characterized in that it is configured such that a field bypass route can be formed between these computer systems.