JPH05334269A - Multi-cpu device - Google Patents

Abstract

PURPOSE:To unnecessitate the change of the application program of a single CPU device, to automatically attain a load distribution, and to attain an update by a latest message packet even when the plural message packets which update the shared information are generated. CONSTITUTION:A shared space 3 commonly enabling an access from plural CPU 1 and 2, and disabling a task arrangement, is equipped with a load managing means 6 which searches the load ratio of each CPU 1 and 2, and an updating means 7 which stamps the generating time on the message packet indicating the update content at the time of the generation of the update of the shared information, and updates the latest message packet by referring to the stamp even when the plural message packets which update the same shared information are generated. Local spaces 41 and 42 of each CPU 1 and 2 are equipped with task managing means 10 which accept a request for the activation of entire tasks under the control of the CPU, allows the tasks to be executed under the control of the CPU whose load is reduced by referring to the load ratio of each CPU 1 and 2, and manage the flow of the task.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement of a multi-CPU device used as a control device in a small-scale real-time system such as a building management system.

In a small-scale real-time system such as a building management system, a single CPU device has been mainly used as a control device, but recently, a multi-CPU device has also been used.

Therefore, even if the application program of the single CPU device is not changed, the load can be automatically distributed, and even if a plurality of message packets for updating the same shared information are generated, the update by the latest message packet is performed. There is a desire to provide a multi-CPU device that can do this.

[0004]

2. Description of the Related Art FIG. 5 is a diagram showing a task flow and a shared information update flow of a conventional multi-CPU device.

A single CPU device has an operating system (hereinafter referred to as an OS) capable of dynamically performing task management, memory area management, interrupt management, and task generation / disappearance. In a multi-CPU device, a plurality of CPUs are commonly used. There is a shared space where you can access tasks but cannot place tasks, and a local space where you can access each CPU.
Task management and memory area management are performed for each local space, and interrupt management is performed by defining the corresponding CPU and handler name for each interrupt point to activate the corresponding handler and creating a primitive and task for communication between CPUs. It has an OS with primitives that can be dynamically deleted.

In the single CPU device, it is not necessary to consider the arrangement of programs and data because they are all in the same logical space, but in the multi-CPU device, since the awareness of the arrangement of tasks is required, the application The program developer considers the load distribution to each CPU in the algorithm of each task of the application program prepared by the single CPU device, and describes the arrangement of which CPU it is under. The application program can be used in a multi-CPU device.

For example, in the event of editing the data and outputting it to the printer, as an application program, a task 20 'for editing the data so that the data can be output to the printer is assigned to the printer under the CPU2. The output task 21 'is directly activated and output to the printer.

Further, for example, a message packet 27 for updating the record A of the magnetic disk 30 which is commonly used by the CPUs 1 and 2 is generated at time T1.
A task for updating the magnetic disk 30 under the CPU2 when a message packet 28 for updating the record A of the magnetic disk 30 is input to the task 25 for updating the magnetic disk 30 at a time T2 later than the time T1. 26, the load on CPU2 is light, and task 26
When the update by is faster than the update by task 25,
The record A on the magnetic disk 30 is updated with the new contents of the latest message packet 28 and then with the old contents of the oldest message packet 27.

[0009]

However, since the application programs are different in each system, the load distribution to each CPU is considered in the algorithm of each task of the application program to determine which CPU it is under. The arrangement must be described for each application program in each system, and the more systems there are, the more troublesome it is to modify the application program.

Since the load distribution is statically determined,
There is also a problem that even if one CPU has a high load, the load cannot be dynamically distributed to the remaining CPUs. Further, when a plurality of shared information updates for updating shared information occur, there is a problem in that new content is updated and then old update content is updated, which may cause a discrepancy.

According to the present invention, it is not necessary to modify the application program used in the single CPU device, load balancing can be automatically executed, and a plurality of shared information updates for updating the same shared information occur. The purpose is to provide a multi-CPU device that can be updated with the latest update contents.

[0012]

FIG. 1 is a block diagram showing the principle of the present invention. As shown in FIG. 1, a shared space 3 that has a plurality of CPUs 1 and 2 and that can be commonly accessed by the plurality of CPUs 1 and 2 but does not allow task allocation, and local spaces 41 and 42 that can be accessed by each CPU 1 and 2 are provided. The local space 41 and 42 have task management, memory area management, and communication between the CPUs 1 and 2, and a primitive that can dynamically create and delete tasks O
In the multi-CPU device having S5, the shared space 3
Is a load management means 6 for obtaining the load factor of each CPU 1, 2.
When the shared information update occurs, the generation time is stamped on the message packet indicating the update content, and when the update is performed, this stamp is seen and the latest message packet is generated even if multiple same shared information update message packets occur. And an update means 7 for updating the
Each local space 41, 42 for each 2 receives the activation request of all the tasks under the CPU, sees the load factor of each of the CPUs 1, 2, and executes the tasks under the CPU with the light load, and The task management means 10 for managing is provided.

[0013]

According to the present invention, the load management means 6 of the common space 3
Then, the load factors of the CPUs 1 and 2 are obtained.

When the task management means 10 in the local space 41 receives the activation of the task B from the task A, for example, the task management means 10 looks at the load factors of the CPUs 1 and 2, and the one with the lower load factor, for example, The task B activation information is sent to the task management means 10 of the CPU 2, and the task management means 10 activates the task B under the CPU 2.

That is, the task does not directly activate the task, but activates the task via the task management means 10 in each of the local spaces 41 and 42.
It is possible to use the multi-CPU device without modifying the application program of the single-CPU device because the subordinate arrangement is not made conscious.

Further, since the tasks under the CPU having a light load are activated and executed by observing the load factors of the CPUs 1 and 2, the load can be automatically distributed. Further, when the updating means 7 updates the common information commonly used by the CPUs 1 and 2, the event occurrence time is stamped in the message packet, and this stamp is viewed when updating. Even if a plurality of message packets for updating the same shared information are generated, the latest message packet can be used for updating.

[0017]

FIG. 2 is a diagram showing a flow of load management of a multi-CPU device of an embodiment of the present invention, FIG. 3 is a diagram showing a flow of task management of a multi-CPU device of an embodiment of the present invention, and FIG. It is a figure which shows the flow of shared information update of the Example of invention.

OS of the multi-CPU device according to the present invention
However, it is the same as the OS in the case of the conventional example. Next, the load management means 6 will be described with reference to FIG.

For each of the local spaces 41 and 42 shown in FIG. 2, a processing loop that takes 100 ms, for example, once is created as the idle processing 24, and is executed permanently at the lowest level, and the counter 23 is executed each time it is executed. Try to count up.

Since the processing of the application program is higher than this level, the idle processing 24 does not operate and the counter 23 does not count up while the processing of the application program is being performed. The count value of the counter 23 is referred to every second at a high level, the vacancy rate is calculated by the following formula, the load rates of the CPUs 1 and 2 are calculated, and the shared space 3
The write counter 23 is reset to the load factor area 25 of the
The load factors of the CPUs 1 and 2 can be known. Vacant rate = [(100ms x count value) / 10000m
s] × 100% Next, task management will be described with reference to FIG.

Local space 41 of CPUs 1 and 2 in FIG.
For each 42, task management units 10-1 and 10-2 are provided, and as for task activation, as shown in FIG. 3A, the event name and the notification source program code are written in the header part, and the information area is written with information. The message to write is used.

For example, in the event of editing the data and outputting it to the printer, the task 20 for editing the data to output the data to the printer, for example, under the control of the CPU 1, is shown in FIG.
Output is written to the printer in the event name of the message (A), the code of the task 20 is written in the notification source program code, and the data to be edited is written in as information, and sent to the task management unit 10-2.

When the processing of the task 20 is completed, the task management unit 10-2 sends a notification that the next message can be dispatched and the received message to the task management unit 10-1.
Send to.

The task management unit 10-1 refers to the load factors of the CPUs 1 and 2 in the load factor region 25 of the shared space 3, and when the CPU 2 has a low load factor, a message is sent to the task management unit 10-1 under the CPU 2. To transfer.

The task management unit 10-1 under the CPU 2
By referring to the task management information area 31 indicating the generated task, it is checked whether the task to be output to the printer has been generated. If the task 21 to be output to the printer has not been generated, the task 21 is generated and The fact that the output task 21 has been generated is written in the task management information area 31.

Then, the task 21 is activated in the dispatch state to output the edited data to the printer. When the processing of the task 21 is completed, an unnecessary message is sent to the task management unit 10-2 under the CPU 2.
Upon receiving this, the task management unit 10-2 rejects the unnecessary message and makes the next message dispatchable.

By doing so, the application program does not need to be aware of the placement of the task to be activated in the task and the configuration of the single CPU device or the multi CPU device. Therefore, the application program of the single CPU device is not required. Can be used as a multi-CPU device without modification.

Also, a CPU with a low load factor is automatically selected,
A task under the CPU with a low load factor can be activated and executed. Next, the shared information update will be described with reference to FIG.

When a message packet 27 for updating the record A of the magnetic disk 30 commonly used by the CPUs 1 and 2 in FIG. 4 is generated at time T1, the updating means 7 stamps this packet 27 at time T1. , CPU
It is input to the task 25 for updating the record A of the magnetic disk 30 under the control.

Similarly, when a message packet 28 for updating the record A on the magnetic disk 30 is generated at time T2 after time T1, the updating means 7 uses this packet 2
The time T2 is stamped at 8 and input to the task 26 for updating the record A of the magnetic disk 30 under the control of the CPU 2.

When the load factor of the CPU 2 is low and the update by the task 26 is to be performed at time T3 earlier than the update time T4 by the task 25, the task A updates the record A on the magnetic disk 30.

At time T4, when the task 25 tries to update the record A on the magnetic disk 30, the time T1 that is stamped in the message packet 27 input to the task 25 is input to the task 26 that was previously updated. Since the time stamp T2 is stamped on the message packet 28 and the content of the latest message packet 28 is updated, the updating unit 7 does not update the task 25.

Even if the tasks 25 and 26 are operated under the same CPU due to the load factor, the operation is the same. Although the above description has been made in the case of using two CPUs, the present invention can of course be applied to a multi-CPU device using more CPUs.

[0034]

As described in detail above, according to the present invention, there is no need to modify the application program used in a single CPU device, load distribution can be automatically executed, and shared information can be updated. Even if the shared information update occurs, the multi-CPU device that can be updated with the latest update content is obtained.

[Brief description of drawings]

FIG. 1 is a block diagram of the principle of the present invention,

FIG. 2 is a diagram showing a flow of load management of a multi-CPU device according to an embodiment of the present invention,

FIG. 3 is a diagram showing a flow of task management of a multi-CPU device according to an embodiment of the present invention,

FIG. 4 is a diagram showing a flow of updating shared information according to the embodiment of the present invention,

FIG. 5 is a diagram showing a task flow and a shared information update flow of a conventional multi-CPU device.

[Explanation of symbols]

 1, 2 is a CPU, 3 is a common space, 5 is an operating system, 6 is a load management means, 7 is an updating means, 10 is a task management means, 10-1 and 10-2 are task management sections, 20, 21, 20 ', 21', 25 and 26 are tasks, 22 is a sampling process, 23 is a counter, 24 is an idle process, 25 is a load factor area, 27 and 28 are message packets, 30 is a magnetic disk, and 41 and 42 are local spaces. Show.

Claims (1)

[Claims] 1. A shared space (3) having a plurality of CPUs (1, 2), which can be commonly accessed by the plurality of CPUs (1, 2), but task placement is not possible, and each CPU (1, 2) Have local space (41, 42) that can be accessed for each
An operating system (5) that has task management, memory area management for each local space (41, 42) and communication between the CPUs (1, 2), and a primitive system capable of dynamically creating and deleting tasks ) In the multi-CPU device, each shared space (3) has a C
Load management means (6) for obtaining the load factor of PU (1, 2)
When the shared information update occurs, the generation time is stamped on the message packet indicating the update content, and when the update is performed, this stamp is viewed and the latest message packet is generated even if multiple same shared information update message packets occur. And updating means (7) for updating
In each local space (41, 42) of (1, 2),
Accepts the activation requests of all tasks under the CPU,
The multi-C is characterized in that a task management means (10) is provided for observing the load factor of the PU (1, 2) and executing the task under the control of a CPU having a light load and managing the flow of the task.
PU device.