JPH02280266A - Parallel electronic computer system - Google Patents

Abstract

PURPOSE:To improve the performance of a parallel system by checking the presence/absence of another processor optimum for the distribution of a load when a new load occurs in its own processor based on load quantity information held by a control means, and making the optimum processor charge the new load when the optimum processor exists. CONSTITUTION:When the load occurs newly in a certain processor unit 10, the processor unit 10 judges whether or not the load is already heavy at a load movement judging part 2. The load movement judging part 2, when judging that the heavy load is already applied on its own processor 1, judges to which processor the optimum allocation can be performed based on information controlled by a load quantity control part 3. As a result, the load is allocated to the processor at a position where a load quantity is minimum. Thereby, since no time required for the allocation of the load is required, the processing performance of a parallel computer system can be improved.

Description

[Detailed Description of the Invention] [Objective of the Invention] (Industrial Application Field) The present invention relates to a parallel computer system that has a plurality of processors and processes a load by allocating it to each processor. The present invention relates to a parallel computer system in which a load distribution method is improved so that the load (work amount) can be allocated to each processor in a short time so that the amount of work is evenly distributed, and processing efficiency can be improved.

(Prior Art) A parallel computer system is a system in which multiple processors (processing units; CPUs) are provided and one application is processed by these processors under the control of one operating system (O8). By sequentially allocating the load to these processors and distributing the processing, the system can increase the processing capacity as a whole.

Conventionally, in such parallel computer systems, when a processor with a heavy load tries to transfer its load to another processor, it first searches for the processor with the lightest load, and then uses the processor with the light load that it finds. Either the load can be shifted to one processor, or the heavily loaded processor can randomly choose another processor, and then determine whether the chosen processor is less loaded than itself. If the other processor's load was light, it would transfer part of its own load to that processor. However, when a processor with a heavy load transfers its own load to another processor with a light load, it has to check the other processor's load one by one, and it takes time and effort to find the processor with the lightest load. In the method of selecting at random, the load of the selected processor is heavier than its own load, so it may have to be selected again and again many times, and it takes 11.7 hours. , Even if the load on a randomly selected processor is lighter than your own, there may actually be other processors with lighter loads, and the load may be higher on multiple processors that make up the system. It was difficult to efficiently allocate it evenly to the processors.

(Issue 8 to be solved by the invention) In a parallel computer system, the load that occurs sequentially is sequentially assigned to one of multiple processors to process it, but conventionally, such a parallel computer system has a heavy load. When a processor transfers its own load to another processor, it may find the processor with the lightest load and then transfer the load to that processor, or the heavily loaded processor may randomly select another processor and reduce the other processor's load. I decided whether the processor was lighter than me or not, and if it was lighter, I would transfer part of my load to that processor.

However, in the conventional method, when a heavily loaded processor transfers its own load to another lightly loaded processor, it must sequentially check the other party's load, and therefore it is difficult to find the processor with the lightest load. It takes a lot of time and effort, and if you select the processor at random, you will find that the load on the selected processor is heavier than your own.
There may be cases where the selection has to be made again and again, or there may be other processors with an even lighter load than the finally selected processor, so it is necessary to efficiently allocate the load to each processor in a short time and evenly. It was difficult.

Therefore, an object of the present invention is to provide a parallel computer system that can allocate loads to each processor evenly and efficiently in a short period of time.

[Configuration of the Invention] (Means for Solving the Problems) In order to achieve the above object, the present invention is configured as follows. In other words, in a parallel computer system that has multiple Bromo-Sosas and processes the generated loads by assigning them to each processor, each processor receives each Bromo-Sosa load amount information from other processors and processes these loads. a management means for updating and retaining information on the amount of load associated with the execution of the load on the self-processor, and updating and storing the information on the amount of load associated with the execution of the load on the self-processor; A transmission means for sending the load amount to the Bromo Tusa and receiving the load amount sent from the other processor side, and when there is a newly generated load for the own Bromo Tusa, based on the load amount information held by the management means. The present invention is configured to include a determining means for checking whether there is another processor that is optimal for distributing the load, and, if there is an optimal processor, controlling the newly generated load to be distributed to the optimal processor. Further, when the judgment means judges that there is an optimal Bromo Tsusa and the negative 61 quantity of the self-processor and the optimum processor are similar, the efficiency is calculated, and when no improvement in efficiency is expected, a new generation The configuration includes a situation determining means for instructing the determining means to process the load by the O processor.

(Function) In such a configuration, each of the multiple processors constituting the parallel computer system that has multiple processors and processes the generated load by assigning it to each processor receives processor load amount information from other processors. There is also a management means that edits and updates these in the order of load amount, and also obtains information on the load amount accompanying the execution of the load on the own processor and updates and stores it.When a new load is assigned, this management means updates and stores the load amount of its own processor, provides this new load amount information to a transmission means, and transmits its own new load amount information to other processors. When the management means on each processor side receives the transmitted load amount information, it edits the information in the order of the load amount and updates and holds the information. On the other hand, when there is a new load on the self-processor, the judgment means on the self-processor side selects another processor that is most suitable for distributing the load based on the load information held by the management means on the self-processor side. The presence or absence of the load is checked, and if there is an optimal processor, the newly generated load is controlled to be shared by that optimal processor.

In this way, the load amount information of all processors is updated and stored on each processor side, and the processor side where a new load has occurred refers to the load amount information of all processors held by itself and selects the optimal processor to allocate the new load. By selecting a processor and having that processor execute the processing of the new load, when a new load occurs, it is possible to immediately select the most suitable processor to allocate the new load and have it execute the process.

Further, when a situation determining means is provided, the situation determining means determines that the determining means has an optimal processor, and
When the loads of the self-processor and the optimum processor are similar, the efficiency is calculated, and when no improvement in efficiency can be expected, the determination means is instructed to process the newly generated load by the self-processor. Therefore, even if there is an optimal processor to allocate the newly generated load, even if the newly generated load is transferred from itself to the optimal processor, the overall processing time reduction effect is small when considering the time required to control the transfer. Since the processor with the newly generated load processes the newly generated load itself, meaningless changes in processing allocation can be eliminated.

Therefore, according to the present invention, in a parallel computer system that has a plurality of processors and processes a load by assigning it to each processor, the load (work amount) on each processor can be reduced in a short time so that the load on each processor is equalized. It is possible to provide a parallel computer system that can perform allocation and significantly improve processing efficiency.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing one embodiment of the present invention. In the figure, 10 is a processor unit, and this processor unit 10 constitutes a system by connecting multiple units via a bus, and under the control of the operation system, these units are assigned processing for the generated load. It is designed to execute processing for that load. Each processor unit 10 includes a processor 1 which is the execution center for control and processing, a load shift determination section 2,
It consists of a load amount management section 3 and a transmitting/receiving circuit section 4.

Among these, when a load occurs, the load shift determining unit 2: determines the lightness or weight of the current load on its own processor 1, and if it determines that the load is light, allocates the generated load to its own processor 1; Furthermore, when it is determined that the load is heavy, it is determined to which processor unit among the plurality of processor units the generated load is to be assigned.

In addition, the load amount management unit 3 updates each processor with respect to the load amount that has changed due to load generation, allocation, and load termination, etc., and edits the load amount in the order of the load amount, and also updates the load amount of the own processor. It notifies all other processor units when a change occurs in the load status information as shown in FIG. It has a load state information part 6 and a own processor load holding part 5 as shown in FIG. 2(a) which holds the load amount information of the own processor. The load shift determination section 2 refers to these to determine whether the load of the current processor is light or heavy and to which processor unit the load is to be assigned.

The transmitting/receiving circuit unit 3 transmits and receives load (work) and load amount, etc., which are input/output information of the processor 1, and also transmits and receives allocation information of the load shift determining unit 2, the load amount, etc. It is used to send and receive information required by the management section 3.

A system is constructed by connecting a plurality of processors with such a configuration. In this system with such a configuration, for example, when a new load occurs in a certain processor unit 10, that processor unit IO The load shift determining unit 2 determines whether or not the load No. 1 has already become heavy. If the load shift determination unit 2 determines that the load on the own processor has not become heavy yet, the generated load is processed by the own processor 1. Then, the own processor 1 changed as a result of this
The own processor load amount holding unit 5 updates and stores the load amount. At the same time, information on the newly changed load amount of the own processor is passed to the transmitting/receiving circuit section 4 in order to notify all other processor units IO. The transmitting/receiving circuit section 4 sends this load amount information to the bus.

In addition, if the load shift determining unit 2 determines that the load on its own processor is already heavy, it determines which processor is best to allocate the load to, based on information managed by the load amount managing unit 3. to decide.

That is, the load shift determination unit 2 uses the information in the load status information unit 6, which has already been edited by the load management unit 3, in which the ranking and load information of each processor is edited in the order of load, and saved as a table. , compares the load amount of the processor at the position with the minimum load amount (always compared with the one at this position) and the load amount of the own processor.

As a result, if the load shift determining unit 2 determines that "the load amount of its own processor is larger," it decides to allocate the generated load to the processor with the minimum load amount and issues the command. It is passed to the transmitter/receiver circuit section 4. As a result, the transmitting/receiving circuit section 4 of the processor unit corresponding to the command information receives this information, passes the commanded load to its own processor, and causes the processor to execute it.

Furthermore, if the load shift determining unit 2 determines that "the load amount of the own processor is smaller" or "the same", it means that the load amount of the own processor is the smallest among all the processors. Therefore, the load shift determining unit 2 notifies its own processor 1 of this, and as a result, its own processor 1 starts executing this load. In addition, the load amount management unit 3 generates current load amount information accompanying this new load execution on its own processor 1 and provides it to its own processor load amount storage unit 5, updates and stores the load amount information, and The information on the load amount of its own processor is passed to the transmitting/receiving circuit unit 5 in order to notify all other processor units.

The transmitter/receiver circuit unit 5 sends this load amount information to all other processors.

When the transmitting/receiving circuit unit 5 receives information about a change in load amount from another processor, the load amount status information unit is updated to store the information on the load amount for the processor unit notified by the load amount management unit 3. 6. Edit and update.

Further, when a load is received or taken from another processor unit, the load shift determining section 2 determines whether the load on the own processor has already become heavy. If the load shift determining unit 2 determines that the load on the own processor 1 has not become heavy, the receiver instructs the own processor 1 to process the load taken by the own processor 1, and
Information to that effect is also given to the load amount management section 3. Thereby, the load amount management unit 3 updates and stores information on the changed load amount of the own processor in the own processor load amount holding unit 5. At the same time, the information on the changed load amount of the own processor is passed to the transmitting/receiving circuit 5 in order to notify all other processor units.

Furthermore, if the load shift determining unit 2 determines that the load on the own processor has already become heavy when receiving or taking a load from another processor unit, the load shift determining unit 2 retains the load amount management unit 3. Based on the load amount information registered in the load amount state information section 6 and the load amount information in the own processor load amount holding section 5, the load amount of the processor with the minimum load amount and the load amount of the own processor are determined. Compare with. Then, when determining that the load amount of its own processor is larger, the receiver decides to allocate the taken load to the processor unit at the position where the load amount is the smallest, and passes the command information to the transmitting/receiving circuit section 5. .

In addition, if the load shift determination unit 2 determines that "the load amount of the own processor is smaller" or "the load amount of the own processor is the same", the load amount of the own processor is lower than the load amount of the own processor among all processor units. The load is said to be the minimum, and therefore, the receiver instructs its own processor 1 to process the load taken by its own processor 1, and also provides information to that effect to the load amount management section 3. As a result, the load amount management section 3
updates and stores this changed information on the load amount of its own processor in the 0 processor load amount holding unit 5.

At the same time, the information on the changed load amount of the own processor is passed to the transmitting/receiving circuit 5 in order to notify all other processor units.

In this way, in this embodiment, each processor unit is provided with a communication means, and the load amount information of all processor units can be collected for each processor unit, so that the latest information on the load amount of all processor units is always updated and maintained. In addition, each processor unit has a function to know which processor currently has the lightest load based on the load amount information of all processor units it owns, so that it can select the processor with the lightest load. Each processor unit can easily know which unit it is, and when a load occurs, it uses the load information of its own processor and the load information of other processor units to
It is possible to select which processor unit is optimal for executing this generated load in consideration of the overall balance, and to allocate the load to the optimal processor unit. This makes it possible to allocate the load to each processor unit in a well-balanced and efficient manner in a short time, and because it eliminates the need for load allocation, the processing performance of the parallel computer system can be greatly improved. can.

FIG. 3 is a block diagram showing another embodiment of the present invention.

Here, a new load situation determination section 7 is added to the configuration shown in FIG.
This is provided. This load situation judgment unit 7 is the self-processor 1.
If the load is transferred to the processor of the processor unit that currently has the lightest load, it is determined whether or not the expected increase in efficiency can be expected, and if it is determined that the efficiency cannot be improved much even if the load is transferred. The processor has a function that instructs the processor to perform processing on its own processor without shifting the load to the lighter-load processor unit.

In the case of this system, for example, when a new load occurs on the own processor, the load status judgment section can shift the load of the own processor to the processor with the lightest load at present, thereby increasing the efficiency as much as expected. Judge whether or not. If it is determined that even if the load is transferred, the efficiency cannot be improved much, that is, the processor with the lightest load is about the same load as the own processor, the load will not be transferred and the processing will be performed by the own processor. The load movement determination section 2 is instructed to perform the following. Based on this command, the load shift determining unit 2 instructs its own processor to implement the newly generated load.

By using this method, if transferring processing to another processor has little effect, the processor can immediately process the new load, and the selected processor unit can be given a command to allocate the new load. Since the processing of the new load is not performed, the amount of notifications between processors can be reduced, and there is no time loss as there is no time for handing over execution, and therefore, the parallel computer This makes it possible to further improve the performance of the system.

In this way, this system has a function on each processor side to edit and update the negative 61 amount information of other processors in order of size, and also update and hold the current load amount information of its own processor. In addition to providing this function, when the load amount on its own processor side changes, it has a function that notifies all other processors of the changed load amount, and each processor side that receives the notification The load amount information is updated, and at the same time, the load amount of the processor whose load amount has changed is edited to indicate the number among all managed processors. This allows each processor to easily find out which processor has the lightest load among the processors. When a load occurs on the own processor, it determines which processor to allocate the newly generated load to by comparing the load on the own processor with the load on the processor with the lightest load among all the managed processors. As a result, it is possible to easily allocate the generated load to the processor with the lightest load, and also to be able to perform this task in a short time, thus making it possible to process the load efficiently. .

Note that the present invention is not limited to the embodiments described above and shown in the drawings, and can be implemented with appropriate modifications within the scope of the gist. Although the load amount status information section and the load amount state information section are provided in the load amount management section, it is needless to say that the present invention is not limited to this.

[Effects of the Invention] As detailed above, the present invention notifies all other processors of this information when the load amount of its own processor changes, and the processor that receives the notification updates the information for each processor that has received the notification. Update and edit in order of load 2, so that when a load occurs, you can easily find another processor with the lightest load, and check the load amount of the own processor and the processor with the lightest load. By comparing the amount of load, it is possible to allocate the generated load to the processor with the current lightest load, so it is possible to find the optimal processor for load allocation in a short time. , it is possible to provide a parallel computer system that can efficiently distribute loads and improve the performance of the parallel system.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention, FIG. 2 is a diagram showing the table structure of the own processor load amount holding section and the load amount state information section, and FIG. 3 is a block diagram showing another embodiment. It is. DESCRIPTION OF SYMBOLS 1... Processor, 2... Load movement judgment part, 3...
- Load amount management section, 4... Transmission/reception circuit section, 5... Own processor load amount holding section, 6... Load amount state information section, 7
...Load status judgment unit.

Claims (2)

[Claims] (1) In a parallel computer system that has multiple processors and processes the generated load by assigning it to each processor, each processor receives processor load amount information from other processors and edits the information in order of load amount. a management means for updating and retaining information on the amount of load associated with the execution of the load on the self-processor, and updating and storing the information; and sending the information on the amount of load on the self-processor updated and stored by the management means to other processors. However, there is a transmission means for receiving the load amount sent from other processors, and when there is a new load on the own processor, it is optimal for distributing the load based on the load amount information held by the management means. 1. A parallel electronic computer system comprising: determining means for checking the presence or absence of other processors and, if there is an optimal processor, controlling the newly generated load to be shared with the optimal processor. (2) In a parallel computer system that has multiple processors and processes the generated load by assigning it to each processor, each processor receives processor load amount information from other processors and edits the information in order of load amount. a management means for updating and retaining information on the amount of load associated with the execution of the load on the self-processor, and updating and storing the information; and sending the information on the amount of load on the self-processor updated and stored by the management means to other processors. However, there is a transmission means for receiving the load amount sent from other processors, and when there is a new load on the own processor, it is optimal for distributing the load based on the load amount information held by the management means. determining means for checking the presence or absence of other processors, and controlling the newly generated load to be shared by the optimal processor if there is an optimal processor; A situation determining means is provided which calculates the efficiency when the load amount of the optimum processor is approximate, and instructs the determining means to process the newly generated load by the own processor when no improvement in efficiency is expected. A parallel computer system featuring features.