JPH0799517B2 - Shared resource management method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a shared resource management method in a system in which a plurality of processors share resources on a shared memory.

[0002]

2. Description of the Related Art In a so-called multiprocessor system, when a plurality of processors share a resource on a shared memory, they can be roughly classified into collective centralized management and divided local management. FIG. 2 shows a block diagram of an apparatus adopting the conventional collective centralized management method. In the figure, a plurality of processors (here, three processors 1A, 1B, 1C)
Are connected to the shared memory 3 via the bus lines 2, respectively. The shared memory 3 stores resources that are required when the three processors 1A, 1B, 1C cooperate with each other to execute work. This resource refers to a program or data written in the shared memory 3, or an area for writing data. In the collective centralized management method, all the processors make all the resources on the shared memory 3 accessible. A shared memory management unit 4 is provided for this management. The shared memory management unit 4 includes a free resource pointer 5 for designating a free resource when each processor requests the allocation of the resource on the shared memory 3.

FIG. 3 shows a conceptual diagram of such free resource management on the shared memory. On the shared memory 3, there are a large number of resources 6 each having a unit amount selected to have an appropriate size. These resources 6 are not necessarily arranged physically consecutively, and for management thereof, they are connected in a chain using pointers to form a so-called chain. The free resource pointer 5 is a free resource 6 existing in the shared memory 3.
Among them, it holds the address of the free resource at the head of the chain. When a free resource allocation request is made, the free resource at the head of this chain is allocated to the requested processor, and the free resource pointer 5 is rewritten to point to the second free resource in the chain. In this way, free resources are allocated one after another, and after the last free resource is allocated, there are no free resources already on the shared memory 3. In this case, even if there is an allocation request, the processor waits until other free resources are released. On the other hand, when the resource allocated to any of the processors is released, it becomes a free resource and is added to the head of the chain of free resources shown in FIG. Then, the content of the free resource pointer 5 is rewritten to point to the address of the leading free resource. In the collective centralized management, the shared memory is managed by the above method.

FIG. 4 shows a block diagram of an apparatus adopting a conventional divided local management method. In the case of the apparatus shown in the figure, a plurality of processors, here three processors 1A, 1B, 1C
Are connected to the shared memory 3 via the bus line 2, and each processor has a memory management unit 7 individually.
A, 7B and 7C are connected. Then, in the shared memory 3, resources 3A, 3B, and 3C that are exclusively used by the processors 1A, 1B, and 1C are set in advance. For example, this resource 3A is the free resource 6 previously described in FIG.
Etc. are included in a certain amount. Memory management unit 7
The resources A, 7B, and 7C individually manage the resources occupied by the corresponding processors. The management method is the same as that described in FIG. In the divided local management, the resources considered to be necessary are allocated in advance to each processor in this way.

[0005]

By the way, the conventional management methods as described above have the following problems. First, according to the collective centralized management method, since the shared memory management unit 4 collectively manages the resources on the shared memory 3, the management itself can be reduced. However, if allocation requests to the shared memory management unit 4 are concentrated, there is a risk that many access waits will occur. Such concentration of access results in access based on exclusive control, which may significantly increase the traffic of the bus line 2 shared by the processors. Therefore, there is a problem that the number of processors that use the shared memory is greatly limited in the system that performs centralized management in a lump.

On the other hand, in the case of divided local management, since the memory management unit is provided for each processor as described above, it is possible to eliminate the bottleneck of the shared bus due to access concentration. However, since the memory management unit separate for each processor manages the resources, the resource allocation is unbalanced as a whole, and the resources are wasted. That is, if resources are fixedly allocated to each processor in advance, some processors may be unnecessarily allocated with free resources, and some processors may have insufficient free resources. This is due to the non-uniformity of the load, which causes a fragmentation problem in the memory resource. The present invention has been made in view of the above points, and an object of the present invention is to provide a shared resource management method that effectively utilizes resources and does not cause bottlenecks and other failures due to access concentration.

[0007]

A shared resource management method of the present invention is a system in which a plurality of processors share resources on a shared memory, and a shared memory management unit for managing free resources on the shared memory; In order to transfer and access a predetermined resource of the shared memory, a cache memory individually provided for each processor and a cache memory management unit that manages free resources on the cache memory are provided, and each processor is provided. Issues a free resource allocation request to the cache memory management unit, and when there is no free resource in the cache memory, makes a free resource allocation request to the shared memory management unit. The resource is transferred to the shared memory management unit when the resource is expelled from the cache memory. Than it is.

[0008]

In this method, each processor separately prepares a cache memory in order to access the resources on the shared memory. When an empty resource occurs in the resource transferred to the cache memory, it is managed by the cache memory management unit without immediately returning it to the shared memory side. Therefore, each processor issues a free resource allocation request to the cache memory management unit. Only when there is no free resource in the cache memory, the shared memory management unit is requested to allocate the free resource. On the other hand, when the resource is expelled from the cache memory, if the resource is a free resource, the management is transferred to the shared memory management unit. As a result, the free resource once transferred to the cache memory is preferentially allocated to the processor accessing the cache memory. This allocation is done locally and is split-locally managed. Further, the amount of resources transferred to the cache memory is limited, and most of them are managed on the shared memory 3, so that efficient collective centralized management can be performed.

[0009]

The present invention will be described in detail below with reference to the drawings. FIG. 1 is a flow chart showing an embodiment of the method of the present invention. Before describing the illustrated method, hardware for implementing the present invention will be described. FIG. 5 is an apparatus block diagram for carrying out the method of the present invention. In FIG. 5, a plurality of processors, that is, three processors 1A, 1 in this case.
B and 1C are configured to share the resources on the shared memory 3 via the bus line 2, respectively. In this case,
In the method of the present invention, each processor has a shared memory 3
The cache memory 12 is accessed without directly accessing the cache memory 12. In order to manage the free resources of the cache memory 12, cache memory management units 11A, 11B and 11C are provided respectively. Although the shared memory management unit 10 is provided to manage the free resources on the shared memory 3, this configuration is as described above with reference to FIG. 3 is as described above.

FIG. 6 is a diagram for explaining the operation of the cache memory management unit for implementing the present invention. As shown in the figure, the cache memory management unit 11 is configured to store the starting addresses of the free resources generated in the cache memory 12 as pointers in a stack. Therefore, for example, when there are two free resources in the cache memory 12,
The cache memory management unit 11 stores two pointers in a stack. Then, when the free resource among them is allocated to the corresponding processor, the pointer in the cache memory management unit 11 disappears. Moreover, when a new free resource is generated in the cache memory 12, a new pointer is added to the cache memory management unit 11. In the embodiment of the present invention, it is preferable that the cache memory management unit always has the pointers indicating the head addresses of all free resources in this manner to facilitate free resource management.

The method of the present invention is basically realized by expanding the parallel cache mechanism in a multiprocessor having a shared memory. The parallel cache mechanism has a structure as described with reference to FIG. 5, and each processor has its own cache memory, transfers necessary resources from the shared memory, and individually allocates the cache memory. Refers to the method of access.

FIG. 7 is a diagram for explaining a normal access operation in such a system. In FIG. 7A, when a fixed resource is transferred from the shared memory 3 to the cache memory 12, the processor 1 using the cache memory accesses the transferred resource. Cache memory 12
Necessary resources are transferred one after another, and the cache memory 12
When is full, the oldest accessed resource is evicted and returned to the shared memory 3. This is a well-known method of using a cache memory. Further, as shown in FIG. 7B, when the resource on the cache memory 12 is updated by the processor 1, if the resource is left as it is, when another processor tries to access the resource, the transfer from the shared memory 3 is performed. Doing so causes inconsistencies in the content. Therefore, processing for matching the contents of the resources is performed at a predetermined timing.

Returning to FIG. 1, the method of the present invention will be described step by step. As described above, in the method of the present invention, the access of each processor is divided local management by the cache memory. Therefore, as shown in step S1 of FIG. 1, when a resource allocation request is first issued, the processor executes the cache memory management unit 11 as shown in step S2.
Request resource allocation to. The cache memory management unit 11 previously manages the free resources as shown in FIG. 6, and first determines whether or not there are free resources (FIG. 1).
Step S3). If there is a free resource, the free resource in the cache memory 12 is assigned to the processor (step S4). On the other hand, if there are no free resources, then a request for free resource allocation is issued to the shared memory management unit 10 (step S5). That is, in this case, the procedure is the same as the conventional centralized management. The shared memory management unit 10 determines whether or not there are free resources in the shared memory (step S6). If there is a free resource on the shared memory, the free resource is allocated to the requested processor (step S7). If there are no free resources, the allocation is rejected (step S8). Such free resource management is as shown in FIG. The allocated free resources are transferred to the cache memory.

FIG. 8 shows an operation explanatory diagram of the cache memory management unit. As shown in the figure, in the cache memory 12,
Resources to be accessed by the processor 1 are transferred from the shared memory. Here, when the processor 1 uses the resource and the resource is no longer needed, the resource becomes a free resource and is released from the processor 1. However, in the present invention, the free resource is not immediately returned to the shared memory, but the cache memory management unit 11 registers and manages the free resource in the manner described above. This is executed as long as there is an area in the cache memory 12 for securing a free resource. When the processor 1 newly requires a free resource, the shared memory management unit has conventionally been requested for a free resource, but in the present invention, the cache memory management unit 11 is requested for a free resource. Then, the cache memory management unit 11 allocates one of the already registered free resources to the processor 1. By executing such processing, the ratio of the processor 1 requesting a free resource from the shared memory management unit via the bus line is reduced. That is, the cache memory management unit 1
As long as free resources are registered in 1, the divided local management is performed and access concentration is prevented.

FIG. 9 shows an operation explanatory diagram of the shared memory management unit. As described above, the cache memory 12 is
Resources to be accessed by the processor 1 are transferred from the shared memory 3 one after another, and a function of preventing concentration of access to the shared memory 3 is fulfilled. Therefore, for example, when a large number of free resources occur in the cache memory 12 and the cache memory 12 becomes full without any of the free resources being accessed, the free resources are expelled from the cache memory 12. In this case, in the present invention, the cache memory management unit 11 deletes the registration of the free resource, and the free resource returned to the shared memory 3 is placed under the management of the shared memory management unit 10 and newly registered. This procedure is as described above with reference to FIG.

Therefore, when the processor 1 requests the cache memory management unit 11 which manages the cache memory 12 to allocate free resources, and when it is found that there are no free resources in the cache memory 12, the shared memory management unit this time. 10 is requested to allocate free resources, and the conventional centralized management process is executed. In this way, the capacity of the cache memory 12 is limited, and the amount of free resources managed by the cache memory management unit 11 is limited, so that the bias or waste of resources is minimized. The present invention is not limited to the above embodiments. The number of processors connected to the bus line, the connection configuration of the memory management unit cache memory thereof, the number of shared memories, the configuration of the shared memory management unit, and the like may be replaced with various configurations having the same function. Further, the configurations of the shared memory management unit and the cache memory management unit may be replaced with those of the well-known configuration suitable for free resource management as described above.

[0017]

According to the shared resource management method of the present invention described above, the cache is divided and locally managed by the so-called parallel cache mechanism in a multiprocessor when the shared resource in the shared memory is on the cache. When it is evicted from the shared memory, it is entrusted to the management of the shared resource in the shared memory, so despite the centralized management of the shared resource as a whole, the cache is effective in the divided local management in accessing the shared resource Can be accessed.
Moreover, since the cache memory management unit allocates the free resources generated on the cache memory,
Appropriate local management is possible to the extent that there is no bias of shared resources. Therefore, it is possible to prevent a bottleneck caused by the concentration of access and to manage the shared resources efficiently and uniformly.

[Brief description of drawings]

1 is a flow chart showing an embodiment of the method of the present invention.

FIG. 2 is a block diagram of an apparatus adopting a conventional collective centralized management method.

FIG. 3 is a conceptual diagram illustrating a free resource management method on a shared memory.

FIG. 4 is a block diagram of an apparatus adopting a conventional divided local management method.

FIG. 5 is a block diagram of an apparatus for carrying out the method of the present invention.

FIG. 6 is an operation explanatory diagram of a cache memory management unit of the present invention.

FIG. 7 is an explanatory diagram of a normal access operation.

FIG. 8 is an explanatory diagram of operations of a cache memory management unit.

FIG. 9 is an explanatory diagram of operations of a shared memory management unit.

[Explanation of symbols]

 S1 to S7 Each step of the method of the present invention

Claims (1)

[Claims] 1. In a system in which a plurality of processors share resources on a shared memory, a shared memory management unit that manages free resources on the shared memory and a predetermined resource of the shared memory are transcribed and accessed. A cache memory individually provided for each processor, and a cache memory management unit for managing free resources on the cache memory. Request for allocating free resources to the shared memory management unit when there is no free resource in the cache memory,
The shared resource management method, wherein the management of the free resources on the cache memory is transferred to the shared memory management unit when the resources are expelled from the cache memory.