JPH0833836B2 - Shared resource management processing method - Google Patents

Description

DETAILED DESCRIPTION [Outline] In advancing processing while occupying shared resources independently of each other, a common resource area is divided and subdivided into a tree structure, and the range in which exclusive occupation occurs is divided. It is disclosed that it can be made as small as possible and the main memory requirement can be made relatively small.

[Industrial applications]

The present invention relates to a shared resource management processing method, in particular, when a plurality of data processing devices access a shared resource independently of each other to proceed with processing, the shared resource is divided into a plurality of shared resource areas, or the shared resource is divided into stages. Subdivided into
Shared resource management systematized in a tree form, variable occupancy range to keep exclusive occupancy to a limited minimum, efficient occupancy, and a small amount of main memory required for occupancy processing It relates to a processing method.

The present invention is an improvement of the shared resource management processing system of Japanese Patent Laid-Open No. 55-6666, which has already been filed, and reduces the main memory area required there.

[Conventional technology]

When a plurality of data processing devices proceed with processing while using shared resources, if one data processing device performs processing while another data processing device undesirably changes the resource, a logical contradiction may occur. . Therefore, in the above case, generally, exclusive occupation is performed and then the process is executed. However, it is desirable that the range in which exclusive occupation is performed be as small as possible in order that a plurality of processing devices can operate in parallel as much as possible, and the range in which exclusive occupation is performed is arbitrary according to the mode of processing. It is desired to be variable.

[Problems to be solved by the invention]

For this reason, the inventor of the present application previously applied for Japanese Patent Application Laid-Open No. 55-6666. However, in the case of this application, the amount of main memory required for shared resource management increases in proportion to the number of resource divisions, and it may not be possible to handle all resources depending on the architecture of the CPU that constitutes the information processing system. When using virtual memory (VS), the amount of storage can be expanded, but there is a case where a page fault occurs and overhead is involved, which is not preferable. Therefore, the main memory should be used as much as possible, and for resources that do not occupy or release resources very often, information that is shared by multiple resource areas is not provided in advance without providing management information corresponding to each resource area. It is desirable to set an area on the main memory and use the area as needed.

[Means for solving problems]

The present invention has solved the above-mentioned point, and divides the common resource into a plurality of resource areas and systematizes each of the resource areas into a tree structure, and the tree structure is fixedly connected including the root of the tree. The tree portion α and the other tree portion β are divided, and the tree portion α is managed by providing a fixed management table corresponding to the tree portion α on the main memory, and the tree portion β is stored in the main memory. A temporary management table is provided above the tree portion β for management.

FIG. 1 is a block diagram showing the principle of the present invention.
Are access units, 3-0, 3-1, ... are processing units, 4-0, 4-1, ... are unique memories, 5 are fixed management tables, 6 are temporary management tables, 7-0 , 7-1, ... represent tree-structured copy information. Also A, B, C ...
Each K represents a common resource section and is organized in a tree structure. The range surrounded by the chain line corresponds to the tree part α, and the remaining part corresponds to the tree part β.

In FIG. 1, the shared resources in the information processing system are systematized in a tree structure composed of nodes A, B, ..., K. It is to be prohibited that another processing device occupies the node A while the processing device occupying at least one of the nodes B, C, ..., K exists. Also, when occupying A, it should be prohibited for another processing device to occupy any one of the clauses B, C, ..., K.

Now, in FIG. 1, it is assumed that the shaded sections C, G, and H are occupied. For the tree part α in FIG. 1, each node A in the tree part is recorded on the fixed management table 5.
The first management information is associated with and exists in correspondence with ~ E. The tree portion β is a portion other than the tree portion α, and the management information of the nodes therein is created in the temporary management table 6 as necessary.

Copy information 7-i indicating the systematized state of the shared resources in the system is stored in the unique memory 4-i corresponding to each processing device 3-i.
It is located in the main memory 1 that can be accessed commonly by -i. In order to occupy the resource area, each processing device 3-i should occupy which node on the tree should be occupied and which address on the main memory 1 the management information related to the occupancy of the node should be. , Copy information on my own memory 4-i 7
See -i. The access unit 2 processes an access request to the main memory 1 from each processing device 3-i, and executes a test and set (TS) instruction, a compare
The present invention provides a means for executing an And Swap (CS) instruction in response to an execution request from one processing device 3-i while excluding access from another processing device 3-j.

When dealing with multiple resource regions that are separated from each other on the tree, do this by occupying only one clause.
That is, for example, when trying to process the sections G and H, the sections G and H
Occupy the lowest of the common upper clauses.
In this example, the common upper nodes are B and A, as is clear from the figure. The lowest node in this is B. Therefore, in this case, the section B is the occupied section.

The fixed management table 5 is provided corresponding to the tree portion α. Also, the contents correspond to the occupied state of the tree. A positive value means that the in-process instruction information is 0 and the lower occupancy count value is positive, and a value of -1 means that the in-process instruction information is 1 and the lower-order in-process count value is 0. Is shown.

The value of A is +3 because three nodes C, H, and G located under A on the tree are occupied. The value of C is -1 because node C is occupied.

The numerical value in parentheses associated with each section in the tree portion β is the element number in the temporary management table. That is, it shows the place where the second management information is set on the temporary management table 6. In this example, clauses I and J are "2", G and H
Is assigned an element number of "4". The reason why the element numbers are duplicated is as follows. In other words, for resource areas that are rarely occupied or released, it is possible to reduce the amount of memory required for management by allocating as needed, rather than setting a fixed management information area corresponding to each resource area. is there. However, if this duplication occurs in a resource area that is frequently occupied or released, as will be apparent from the following description, the overhead of management processing increases, which is an unfavorable result. Therefore, the tree part α,
How to divide into β, the number of elements of the temporary management table, how to allocate the element number on the temporary management table to each section of the tree portion β, etc. should be decided according to the information processing system.

[Action]

For example, the nodes A, B, C, D, and E in the range shown as the illustrated tree portion α including the root A in the tree structure are accompanied by the fact that they are occupied by themselves or one of the subordinate nodes is occupied. It is managed by the fixed management table 5 as having a high possibility of being used. That is, the management information is always held in a one-to-one form corresponding to each node. On the other hand, the nodes F, G, H in the range shown as the illustrated tree portion β
For I, J, and K, the frequency of occupation is not so great, and if the management information is held in a one-to-one form corresponding to each clause, the capacity of the main memory 1 may become undesirably large. Absent. Therefore, when an occupied state occurs for the time being, it is registered and managed in the temporary management table 6.

The temporary management table 6 is divided into a numbered element area 6-1 and an extension chain area 6-2. And the element number is given corresponding to the element position in the numbered element area,
In the illustrated nodes F, G, ..., Element numbers are associated with each other, such as "3", "4", etc., depending on which element position should be registered. Further, in the extended chain area, when overlap occurs in the divided areas in the numbered element area, the pointed area is used from the numbered element area for use. When two nodes G and H having the element number "4" are to be occupied together as in the example shown in FIG. 1, the above overlap occurs and the extension chain region is used.

〔Example〕

FIG. 2 shows a configuration example of a temporary management table for the tree portion β shown in FIG. As described above, the element numbers “4” are associated with the sections G and H. Since the nodes G and H are both occupied, duplication occurs in the second management information, and the second management information includes two third management information corresponding to G and H.
It points to a chain of management information.

The temporary management table has two parts. One is a numbered element area 6-1 that exists corresponding to the element number given to each section of the tree portion β, and the second management information is stored in that element. The other one is the extension chain area 6-2,
This is because when the second management information is duplicated, that is, when a plurality of resource areas involved in occupancy correspond to the same element number, the second management information cannot be used as information indicating occupancy, and therefore This is an area for providing third management information for each section and managing each section individually.
In this example, the unused chain in the extension chain area 6-2 is pointed to by the unused chain entrance pointer 12 shown in FIG. Further, each element in the unused chain points to the next unused element by the chain pointer except the last element (stores the address on the main memory).

In FIG. 2, the duplicated display information 21 exists corresponding to each element in the numbered element area 6-1 in the temporary management table 6, and 0 is not duplicated when storing the management information of the tree portion β,
1 indicates a state with overlap. When there is an overlap, the occupation information of each section is stored in the extension chain area 6-2 pointed by the chain pointer 27 from there.

The processing instruction information 22 is information that is set to 1 when the corresponding resource area is being processed (that is, occupied).

Free / in-use display bit 23 is also a numbered element area
It is in each element in 6-1 and 0 indicates that it is empty and 1 indicates that it is in use. Free means that the element does not store occupancy information of any resource area, and there is no chain to the third management information. Thus, if the free / busy indicator bit 23 is 0, then none of the nodes of the tree portion β associated with this element are occupied. Further, the in-process instruction information 22 and the low-order in-process count value 24 of all the same sections can be regarded as 0.

The meaning of the low-order processing count value / duplication count value 24 differs depending on the case.

(I) In the numbered element area (first area) 6-1, the free / busy display bit 23 is 1 (in use), and the duplicate display information 21 is 0 (no duplication). Case: Hold the lower processing count value of one node of the tree portion β.

(Ii) In the numbered element area (first area) 6-1, the free / busy display bit 23 is 1 (in use), and the duplicate display information 21 is 1 (duplicate). Case: Multiplicity,
That is, the chain length (count value 2 in the above example) in the expansion chain area pointed from here is held.

(Iii) When it is in the extended chain area (second area) 6-2, and it is in the chain starting from any element of the numbered element area (first area) 6-1 being used. : Tree part β
Holds the in-process count value of one clause of When the count value is -1, it is considered that the corresponding resource area is being processed.

To retrieve one free element from the extension chain area 6-2, as is well known, for example, an unused chain entry pointer 12
The leading element indicated by is taken out, and as the processing, the value of the next element pointer written in the leading element is set again in the unused chain entry pointer 12.

On the contrary, when returning one element to the unused chain, the value of the unused chain entry pointer 12 at that time is set in the next element pointer area in the returned element, and the unused chain entry pointer In 12, set the address of the element to be returned.

Since it is necessary for one processing device to perform the processing of taking out and returning the empty elements in the extension chain area, an unused chain access semaphore 13 is provided as shown in FIG. To obtain this, when the value is 1, it is rewritten to 0 while excluding access from other processing devices. This is possible, for example, with test and set instructions. When the unused chain element is retrieved or returned, the value of the unused chain access semaphore 13 is changed from 0 to 1.

In the above setting example, a pointer to the third management information corresponding to the node G is stored in the second management information corresponding to the nodes G and H by the chain pointer 27. 2 corresponding to sections G and H
The value "-1" (the processing instruction information is 1) is stored in one of the third management information. In addition, in the third management information,
Resource area names "FILE.G" and "FILE.H" for sections G and H, respectively
Stores a pointer to.

A processing procedure for reading the management information of the resource area H having the name "HILE.H" in the state of FIG. 2 will be described.

First, it can be seen that the resource area H is in the tree portion β on the tree. Therefore, it is determined at least by section B (or may be common to the entire tree, or the tree part β
, Which may be determined by the divided area of the tree portion β including the section B) of the H name "FILE.
Convert H "to element number (algorithm is available to do such conversion).

As a result, it is assumed that the value 4 is obtained as illustrated. Then, the fourth element (one piece of second management information) corresponding to the value “4” on the temporary management table 6 is read. In the second management information, the free / busy display bit is 1, the duplicate display information is 1, and the duplicate count value is 2. Therefore, in addition to H, this element is used as the second management information. There is a section (in this case, section G). Therefore, the semaphore 25 for extended chain access in the fourth element is acquired,
Main memory 1 using the value of the chain pointer 27 as an address
Read an element from. Then, in this example, first, the third management information corresponding to the section G is obtained. The pointer to the resource area unique name of the element points to "FILE.G", but as a result of the comparison, it is found that the resource area does not belong to the desired resource area.
Therefore, the chain pointer 27 is further traced to read the next element. This time, the name of "FILE.H" matches the name of the required resource area, so the management information for the resource area H has been obtained.

The duplicate count value in the second management information used when the duplicate display information is 1 is provided for the following purpose.

When releasing one resource area, if the duplicate count value becomes 1 from 2, it can be seen that the use of the extension chain area 6-2 is no longer necessary, so the remaining extension chain area
The management information in 6-2 is transferred to the second management information in the numbered element area 6-1. Then, the duplicate display information is set to 0. Also,
The elements of the extension chain area 6-2 that are no longer needed are returned to the unused chain by the method described above.

The reason why the management information of the section is stored in the second management information although there is a possibility of duplication as described above is that if the probability of duplication of the management information is expected to be less than a certain value, duplication will occur. This is because it is possible to reduce the average overhead in the case of occurrence and the case of not occurring. If they do not overlap, only the second management information needs to be accessed, and the processing time is short.

In addition, the semaphore for extended chained access shown in FIG.
Part 25 is for ensuring that the process of changing the extension chain is performed one by one as described above, and is used in the following cases.

In the occupancy process, it was revealed from the second management information that duplication has occurred. Therefore, when a third management information area is secured in the extension chain area 6-2 to create a new chain, When adding to the chain of 3 management information, or when changing the 3rd management information.

In the resource releasing process, when the chain of the third management information is reduced, or when the third management information is deleted and the management information is transferred to the second management information, or when the third management information is changed.

When reading the contents of the third management information.

In the present embodiment, since the processing of and is performed in pairs prior to the processing, the same semaphore is acquired in and the same semaphore is returned at the end of or. However, when the extension chain is newly set after reading the contents of the second management information, the semaphore in the second management information is acquired again. When the semaphore is acquired, it is confirmed that the semaphore is 1, and then it is changed from 1 to 0 while excluding access from other processing devices. Conversely, when it is returned, it is changed from 0 to 1.

When reading the second management information, it is not necessary to acquire the same semaphore in advance, but if the same semaphore is 0, the read data is valid, and if it is 1, it is considered as being changed. Then, the reading is repeated until it becomes 0. In addition, when changing the second management information, it is read once with the same semaphore being 0, then the second management information is changed in the processing device, and then the previous command (CS (compare and swap)) is executed. The information after the change is written back while confirming that the value is the same as the read value and excluding access from other processing devices.
If the contents have changed from before, the process is repeated from the rereading of the second management information. Also, when the second management information is empty and the management information is newly set therein, and vice versa, the same CS instruction is used. If the contents are different from those previously read, the second management information is read again and repeated.

Fig. 3 shows the overall processing flow for resource area occupation processing.
FIG. 4 shows a processing flow relating to management information reading processing, FIG. 5 shows a processing flow of a portion for setting resource area in-process instruction information, and FIG. 6 shows an overall processing flow relating to release processing.

〔The invention's effect〕

As described above, according to the present invention, it is possible to effectively manage shared resources while reducing the memory capacity.

[Brief description of drawings]

FIG. 1 is a block diagram of the principle of the present invention, FIG. 2 is a structural example of a temporary management table, and FIGS. 3 to 6 are process flows in the case of the present invention. In the figure, 1 is a main memory, 2 is an access unit, 3 is a processing device, 4 is a specific memory, 5 is a fixed management table, and 6 is a temporary management table.

Claims (4)

[Claims] 1. A data processing system including a plurality of data processing devices, wherein the data processing device advances processing while independently accessing a plurality of resources included in the processing system, and the at least one data processing. In a data processing system in which a device occupies a resource area where its own processing is involved when accessing the common resource, the common resource is divided into a plurality of resource areas and each resource area is organized into a tree structure. At the very least, the tree structure is defined by a certain connected tree part α including the root of the tree,
The tree part β is divided into other parts, and a fixed management table is provided for the tree part α in the main memory in correspondence with the tree part α. In the fixed management table, one pair is provided for each node of the tree part α. 1. The first management information is provided corresponding to 1, and in the first management information, (i) at least 1-bit in-process instruction information indicating that the own resource area is being processed, (ii) from the own resource area Also prepares a lower processing count value indicating that the resource area located in the lower order is being processed. For the tree part β, a temporary management table is provided in the main memory in correspondence with the tree part β, and the temporary management is performed. The table stores the second management information temporarily set only for the resource area that is involved in the occupation by any one of the resource areas corresponding to the nodes in the tree portion β. , The second tube During information, at least the first
When the information of (i) and (ii) above is provided as in the management information, and one of the processing devices newly attempts to process the resource area or some of the resource areas located lower than the resource area, those (A) If the resource area is in the tree part α, which is higher than the entire clause corresponding to the one or more resource areas of the above, and corresponding to the lowest clause among them, If the in-process instruction information in the first management information corresponding to the section is set or the lower processing count value is not "0", (B) if the relevant resource area is in the tree portion β, then the corresponding section When the second management information that exists is present in the temporary management table and the in-process instruction information stands or the lower in-process count value is not "0", the process for the resource area is prohibited. That Shared resource management processing system in which the butterflies. 2. A temporary management table for expanding an array of elements having the same size (this is a numbered element area) (6-1) and a pointer to each other to form a chain. It is divided into a chained area (6-2) and each element of the numbered element area (6-1) has the above information (i) and (ii) and the element is occupied by one of the resource areas. In-use instruction information (23) indicating whether or not it is used, and a pointer area (27) to the extended chain area are provided, and the tree portion β
Each node of is assigned a number of the element in the numbered element area to be used at the time of occupancy, and when the element is involved in occupying any one resource area of the nodes in the tree portion β, The above-mentioned information (i) and (ii) and the in-use instruction information (23) are stored as 1 as the second management information in the element position corresponding to the section, and if plural resource areas are the same on the temporary management table. When the element number is given and the plurality of resource areas are involved in the current occupation, the in-use instruction information is used without using the above information (i) and (ii) in the element. And the second management information is stored with the overlapping display information as 1, and the same (i) and (ii) as in the first management information is stored on the chain of the third management information pointed by the second management information. Information is stored corresponding to the plurality of resource areas, and the third management information is stored. The shared resource management processing method according to claim (1), which is used as a substitute for the second management information described in claim (1). 3. A unique name in the system is given to each resource area corresponding to each section of the tree portion β, and an element on the temporary management table is determined from the unique name at least for each resource area. An algorithm for converting into a number is given to the processing system, and in the second and third management information,
Shared resource management according to claim (2), characterized in that pointer information indicating a resource area unique name is provided in addition to the information (i) and (ii) in the first management information. Processing method. 4. The semaphore to be occupied in the second management information prior to the access processing to the chain of the second management information and the third management information pointed to by the second management information and the processing of changing the chain are displayed. An occupied bit to
When accessing the third management information, the processing device occupies the occupied bit prior to the access, and the shared resource management processing method according to claim (1).