JP3818172B2 - Multiprocessor system, process control method, and process control program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention , Ma Multi-processor system , Process control method, and process control program In particular, Establish a busy lock dedicated processor Multiprocessor system , Process control method, and process control program .
[0002]
Here, the “busy lock target resource use process” is a process that is executed as a part of a program from a plurality of programs, and is a process that uses a resource to be locked (proprietary lock), and , A process executed in a busy manner (frequently). Note that a resource to be locked in the busy lock target resource use process as described above is referred to as a “busy lock target resource”.
[0003]
[Prior art]
An example of a conventional multiprocessor system is described in Japanese Patent Application Laid-Open No. 07-325791.
[0004]
FIG. 9 is a block diagram showing a configuration of a multiprocessor system (CPU (Central Processing Unit) fixed system) described in the publication.
[0005]
As shown in FIG. 9, this multiprocessor system includes a CPU fixed flag setting means for setting a CPU fixed flag at the start of a job (program execution unit, corresponding to “process” in the present invention), and fixing of the job. Fixed CPU number setting means for storing a user-specified CPU number in the CPU number storage area, CPU assigned job selection means for selecting a job to which a CPU is assigned from an execution waiting job queue when a free CPU occurs, and a fixed CPU for the job CPU fixing determining means and CPU assigning means for assigning a CPU to the job when the number and a free CPU number match, and a fixed for changing the value of the fixed CPU number storage area of the job to the CPU number designated by the operator command From the CPU number resetting means and the CPU executing the job Is configured to include a CPU reallocation means (CPU fixing flag and the fixed CPU number storage area is n (a positive integer) is present).
[0006]
In the conventional multiprocessor system including the multiprocessor system having such a configuration, at the time of executing a program, one of the processors for each process (program execution unit; expressed as “job” in the above publication) Is assigned and control is performed so that the processor executes all the processes of the process (control to execute the process by a specific processor for each process).
[0007]
Therefore, even for a program including a busy lock target resource utilization process, all instructions of all the processes of the program (the busy lock target resource utilization process and other processes) are executed by a specific processor.
[0008]
[Problems to be solved by the invention]
In the above-described conventional multiprocessor system, control is performed so that a specific processor executes each process (program execution unit), so that the busy lock target resource utilization processing can be executed equally by all processors. Therefore, when a busy lock target resource use process that is executed as a “part of program” is executed in a plurality of programs, cache miss hits frequently in each processor, and the busy lock target resource use process However, there is a problem in that the processing performance of the system may be reduced, and the throughput of the system may be reduced.
[0009]
For example, consider a case in which a buffer control function shared by a system is used for database access in a multiprocessor system in which a database is frequently accessed online from many processes.
[0010]
In this case, the buffer control function is frequently called from each process. Here, since the buffer control function is shared by the system, it is necessary to serialize between processes by lock processing. Therefore, when control is performed such that a specific processor is executed for each process as in a conventional multiprocessor system, a plurality of processes that execute processing for realizing a buffer control function are alternately performed by a plurality of processors. Every time the processor that executes the process that implements the buffer control function is changed, a cache memory miss occurs when referring to shared information on the memory (such as management information and control information related to the buffer control function). The performance of the frequently operated buffer control function is degraded, resulting in a decrease in system throughput.
[0011]
In view of the above-described points, an object of the present invention is not to perform control by a specific processor for each process in a multiprocessor system including a plurality of processors each having a cache memory for each processor, but to perform busy lock target resources. By executing the use process only with a specific processor, the information on the memory area referred to in the busy lock target resource use process is continuously placed in the cache memory of the processor, thereby increasing the cache hit rate and improving the system throughput. It is to provide a multiprocessor system capable of performing the above.
[0012]
In addition to the above publication, as a patent publication relating to the prior art for the multiprocessor system of the present invention, a system comprising a plurality of processors (in the following publications, expressed as “processor module” or “computer”) There are "JP-A 04-340648" and "JP-A 11-65863" which are similar to the present invention in that they are considered.
[0013]
The technique described in the above Japanese Patent Application Laid-Open No. 04-340648 (“selection method of representative processor module”) is a computer system including a plurality of processor modules having several central processing units and memories. This is a technique for appropriately selecting a representative processor module for monitoring and the like.
[0014]
Further, the technique described in the above Japanese Patent Application Laid-Open No. 11-65863 (“shared resource management method”) provides a system that performs exclusive control of shared resources centrally in a parallel computer system using a coupling mechanism. It is a technique to do.
[0015]
However, these technologies do not consider at all what is characteristic of the present invention, “to execute a busy lock target resource use process by a specific processor”, and are essentially different from the present invention. .
[0016]
[Means for Solving the Problems]
A multiprocessor system according to the present invention includes a plurality of processors including a cache, an initial setting unit, a process start unit, an execution process selection unit, a general-purpose process execution unit, and a busy lock execution unit, and a memory connected to the processor. A multiprocessor system comprising:
Stores the “busy lock target resource address”, which is the instruction address of the processing that uses the busy lock target resource, which is a frequently locked resource, in the “busy lock address storage area” of the memory, and uses the busy lock target resource The initial setting means for storing a “busy lock dedicated processor number” indicating a busy lock dedicated processor that executes only the processing to be performed in the “busy lock dedicated processor number storage area” of the memory;
The process start means for assigning a process number to a program and registering a set of a process number and a start instruction address of the program as general process information in a “general process queue” of the memory;
If the own processor number is equal to the “busy lock dedicated processor number” in the “busy lock dedicated processor number storage area”, control is passed to the busy lock execution means, and if not equal, control is passed to the general-purpose process execution means. The execution process selection means;
If the “general process queue” is empty, control is passed to the execution process selection means. If the “general process queue” is not empty, the general process information is extracted from the “general process queue”, the instruction is a lock instruction, and “ If it is equal to the “busy lock target resource address” in the “busy lock address storage area”, a set of process number and instruction address is registered in the “busy lock dedicated process queue” of the memory as dedicated process information, and the execution processing selection means If the instruction is not a lock instruction or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the instructions are sequentially executed up to the last instruction of the program, and the execution processing selecting means Control or a lock command, and the “busy lock address storage area” Said generic process executing means for sequentially executing until it equals a click target resource address ",
If the “busy lock dedicated process queue” is empty, control is passed to the execution process selection means. If the “busy lock dedicated process queue” is not empty, the dedicated process information is extracted from the “busy lock dedicated process queue” and the instruction is unloaded. If it is a lock instruction and is equal to the “busy lock target resource address” in the “busy lock address storage area”, the combination of the process number and instruction address is registered in the “general process queue” of the memory as general process information. Control is passed to the execution processing selection means, and if the instruction is not an unlock instruction or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the instructions are sequentially executed up to the last instruction of the program. Then, control is passed to the execution process selection means or an unlock command is issued, and “busy lock Said sequentially performed until equal to "busy lock target resource address" in the address storage area "in Busy lock Execution means;
Comprising the processor.
[0017]
A process control method of the present invention includes a plurality of processors each including a cache, an initial setting unit, a process start unit, an execution process selection unit, a general-purpose process execution unit, and a busy lock execution unit, and a memory connected to the processor. A process control method in a multiprocessor system including:
The initial setting means stores a “busy lock target resource address”, which is a command address for processing using a busy lock target resource, which is a resource frequently locked, in the “busy lock address storage area” of the memory, A procedure of storing a “busy lock dedicated processor number” indicating a busy lock dedicated processor that executes only a process using the busy lock target resource in the “busy lock dedicated processor number storage area” of the memory;
The process starting means assigns a process number to a program and registers a set of a process number and a program start instruction address as general process information in the “general process queue” of the memory;
The execution process selection means passes control to the busy lock execution means when its own processor number is equal to the “busy lock dedicated processor number” in the “busy lock dedicated processor number storage area”. A procedure for passing control to the process execution means;
If the “general process queue” is empty, the general process execution means transfers control to the execution process selection means. If the “general process queue” is not empty, the general process execution means takes out the general process information from the “general process queue”. If it is a lock command and is equal to the “busy lock target resource address” in the “busy lock address storage area”, a set of process number and command address is stored in the “busy lock dedicated process queue” of the memory as dedicated process information. If the command is not a lock command or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the commands are sequentially transferred to the last command of the program. Execute and pass control to the execution process selection means, or is a lock command and “busy lock A step of sequentially executed until it equals the "busy lock target resource address" Les storage area "within
Said Busy lock The execution means gives control to the execution processing selection means when the “busy lock dedicated process queue” is empty, and extracts the dedicated process information from the “busy lock dedicated process queue” when the “busy lock dedicated process queue” is not empty. If the instruction is an unlock instruction and is equal to the “busy lock target resource address” in the “busy lock address storage area”, the combination of the process number and the instruction address is used as the general process information in the “general process queue” of the memory. If the instruction is not an unlock instruction or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the instruction is transferred to the last of the program. Are executed sequentially until the instruction and control is passed to the execution process selection means, or is the unlock instruction, A step of sequentially executed until it equals the "busy lock target resource address" in the "busy lock address storage area",
including.
[0018]
A process control program according to the present invention includes a plurality of processors including a cache, an initial setting unit, a process start unit, an execution process selection unit, a general-purpose process execution unit, and a busy lock execution unit, and a memory connected to the processor. A process control program in a multiprocessor system including:
In the initial setting means, a “busy lock target resource address” that is a command address of processing that uses a busy lock target resource that is a resource that is frequently locked is stored in the “busy lock address storage area” of the memory, Executing a procedure of storing a “busy lock dedicated processor number” indicating a busy lock dedicated processor that executes only a process using the busy lock target resource in the “busy lock dedicated processor number storage area” of the memory;
The process start means assigns a process number to a program, and executes a procedure for registering a set of a process number and a program start instruction address as general process information in the “general process queue” of the memory,
If the own processor number is equal to the “busy lock dedicated processor number” in the “busy lock dedicated processor number storage area”, the control is passed to the busy lock executing means. Execute the procedure to pass control to the process execution means,
If the “general process queue” is empty, control is passed to the execution process selection means, and if the “general process queue” is not empty, the general process information is extracted from the “general process queue” If it is a lock command and is equal to the “busy lock target resource address” in the “busy lock address storage area”, a set of process number and command address is stored in the “busy lock dedicated process queue” of the memory as dedicated process information. If the command is not a lock command or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the commands are sequentially transferred to the last command of the program. Execute and pass control to the execution process selection means, or is a lock command and “busy lock To execute the steps of sequentially executed until it equals the "busy lock target resource address" Les storage area "within
Said Busy lock If the “busy lock dedicated process queue” is empty, control is passed to the execution processing selection means. If the “busy lock dedicated process queue” is not empty, the dedicated process information is extracted from the “busy lock dedicated process queue”. If the instruction is an unlock instruction and is equal to the “busy lock target resource address” in the “busy lock address storage area”, the combination of the process number and the instruction address is used as the general process information in the “general process queue” of the memory. If the instruction is not an unlock instruction or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the instruction is transferred to the last of the program. Are executed sequentially until the instruction and control is passed to the execution process selection means, or is the unlock instruction, A step of sequentially executed until it equals the "busy lock target resource address" in the "busy lock address storage area", it is executed.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
Next, the present invention will be described in detail with reference to the drawings.
[0022]
(1) First embodiment
[0023]
FIG. 1 is a block diagram showing a configuration of a multiprocessor system according to the first embodiment of the present invention.
[0024]
Referring to FIG. 1, the multiprocessor system according to the present embodiment includes a plurality of processors 100 and a memory 200.
[0025]
Each processor 100 includes a means group 1 including means for realizing the present invention, and a cache memory 101. In the multiprocessor system shown in FIG. 1, except that any one of the processors 100 becomes a processor that exclusively executes a busy lock target resource use process (a process that uses a busy lock target resource) (a busy lock dedicated processor). The processors 100 are in an equal relationship.
[0026]
The memory 200 includes a busy lock address storage area 10 for storing a busy lock address, which is information indicating the location (address) of the busy lock target resource, and a busy lock dedicated processor number storage area for storing a processor number of the busy lock dedicated processor. 11 and information on a process waiting for execution of processing other than the busy lock target resource use processing (the process number of the process and the instruction address of the instruction to be executed next in the program to which the process is assigned) Information on the process waiting for execution of the busy lock target resource use processing (the process number and the process are assigned) Instruction add of the next instruction to be executed Busy lock only process information is a set) of the scan is and a busy lock dedicated process queue 13 which is registered in the queue format.
[0027]
The means group 1 in each processor 100 includes an initial setting means 2, a process start means 3, an execution process selection means 4, a general process execution means 5, and a busy lock execution means 6.
[0028]
FIG. 2 is a flowchart (flow chart) showing processing of the initial setting means 2. This processing includes a busy lock address storing step 201, a busy lock dedicated processor number storing step 202, a general-purpose process queue empty setting step 203, and a busy lock dedicated process queue empty setting step 204.
[0029]
FIG. 3 is a flowchart showing the processing of the process starting unit 3. This process includes a process number assignment step 301 and a general process information registration step 302.
[0030]
FIG. 4 is a flowchart showing the process of the execution process selection unit 4. This process includes a self processor number determination step 401, a general process execution means control shift step 402, and a busy lock execution means control shift step 403.
[0031]
FIG. 5 is a flowchart showing the processing of the general-purpose process execution means 5. This process includes a general process queue empty determination step 501, a general process information extraction step 502, a busy lock target resource lock command success / failure determination step 503, an instruction execution step 504, a busy lock dedicated process information registration step 505, a program It consists of a final instruction adequacy determination step 506, an instruction address progression step 507, and an execution process selection means control shift step 508.
[0032]
FIG. 6 is a flowchart showing the processing of the busy lock execution means 6. This processing includes a busy lock dedicated process queue empty determination step 601, a busy lock dedicated process information extraction step 602, a busy lock target resource unlock command success / failure determination step 603, an instruction execution step 604, and a general process information registration step 605. And a program last instruction appropriateness determination step 606, an instruction address advance step 607, and an execution process selection means control shift step 608.
[0033]
FIG. 7 is a diagram for explaining a specific operation of the multiprocessor system according to the present embodiment.
[0034]
Next, the overall operation of the multiprocessor system according to the present embodiment configured as described above will be described in detail with reference to FIGS.
[0035]
First, general and schematic operations will be described with reference to FIGS.
[0036]
First, the operation at system startup will be described.
[0037]
When the system is started up, the initial setting means 2 in one of the processors 100 performs an initial setting process of information related to the present invention in the memory 200 as shown below (see FIG. 2). It should be noted that which processor 100 is in charge of the initial setting process is determined by a selection algorithm (selection algorithm similar to the prior art) set in advance in the multiprocessor system.
[0038]
First, an address (busy lock address) for specifying the position of the busy lock target resource is stored in the busy lock address storage area 10 in the memory 200 (step 201).
[0039]
Further, the processor number (busy lock dedicated processor number) of a dedicated processor (busy lock dedicated processor) for executing the busy lock target resource use processing is stored in the busy lock dedicated processor number storage area 11 in the memory 200 ( Step 202).
[0040]
Further, the general-purpose process queue 12 in the memory 200 is set to be empty (step 203), and the busy lock dedicated process queue 13 in the memory 200 is set to be empty (step 204).
[0041]
Secondly, the operation when the user gives an instruction to start a program (a program including a busy lock target resource utilization process) will be described.
[0042]
When the start of the program is instructed by the user, the process start means 3 in one of the processors 100 performs a process start process as shown below (see FIG. 3). Note that which processor 100 is in charge of the process start processing is determined by a selection algorithm (selection algorithm similar to the prior art) set in advance in the multiprocessor system.
[0043]
First, a process number (a process number of a process for executing the program) is assigned to the program (step 301).
[0044]
Further, information indicating a set of the process number and the start instruction address of the program is registered in the general-purpose process queue 12 in the memory 200 as general-purpose process information (step 302).
[0045]
Third, the operation of the execution process selection unit 4 in each processor 100 will be described.
[0046]
The execution process selection means 4 performs the following process when the own processor (the processor 100 in which the self is present. The expression “own processor” is also used for other means) executes the process (see FIG. 4).
[0047]
First, referring to the busy lock dedicated processor number storage area 11, it is determined whether or not the processor number of the own processor and the processor number (busy lock dedicated processor number) of the busy lock dedicated processor number storage area 11 are equal (step 401). ), It is determined whether or not the own processor is a busy lock dedicated processor.
[0048]
When it is determined in step 401 that “the two processor numbers are not equal” (when it is determined that the own processor is not a busy lock dedicated processor), control is passed to the general-purpose process execution means 5 in the own processor (step 402). .
[0049]
On the other hand, if it is determined in step 401 that “both processor numbers are equal” (when it is determined that the own processor is a busy lock dedicated processor), control is passed to the busy lock execution means 6 in the own processor (step 403). ).
[0050]
Fourthly, the operation of the general-purpose process execution means 5 in each processor 100 will be described.
[0051]
When the general process execution means 5 receives control from the execution process selection means 4 in its own processor, it performs the following processing (see FIG. 5).
[0052]
First, it is determined whether the general-purpose process queue 12 is empty (step 501).
[0053]
If it is determined in step 501 that “the general-purpose process queue 12 is empty”, control is passed to the execution processing selection means 4 in its own processor (step 508).
[0054]
On the other hand, if it is determined in step 501 that “the general-purpose process queue 12 is not empty”, a set of process number and instruction address (general-purpose process information) is taken out from the general-purpose process queue 12 (step 502), and the instruction address Attempts to execute each instruction in the program to which the process is assigned in order from the instruction specified in.
[0055]
When each instruction is executed, the busy lock address storage area 10 is referred to before the instruction is executed, and the instruction is a lock instruction (the busy lock target resource lock instruction) indicating the start of the busy lock target resource use process. It is determined whether or not. That is, it is determined whether or not the instruction is a lock instruction and the lock address of the lock instruction (information indicating the location of the lock target resource) is equal to the busy lock address in the busy lock address storage area 10. (Step 503).
[0056]
If it is determined in step 503 that “the command is a lock command indicating the start of busy lock target resource use processing”, the command is not executed and the set of the process number and the command address of the command ( The busy lock dedicated process information) is registered in the busy lock dedicated process queue 13 (step 505), and control is passed to the execution process selection means 4 (step 508).
[0057]
On the other hand, if it is determined in step 503 that “the command is not a lock command indicating the start of busy lock target resource use processing”, the command is executed (step 504), and the command is the last command ( It is determined whether it is a final instruction) (step 506).
[0058]
If it is determined in step 506 that “the instruction is the last instruction of the program”, control is passed to the execution process selection means 4 (step 508).
[0059]
On the other hand, if it is determined in step 506 that "the instruction is not the last instruction of the program", the instruction address is advanced to the next instruction (step 507), and the "next instruction" is targeted for step 503 and subsequent steps. Repeat the process and determination.
[0060]
Fifth, the operation of the busy lock execution means 6 in each processor 100 will be described.
[0061]
When the busy lock execution means 6 receives control from the execution process selection means 4 in its own processor, it performs the following process (see FIG. 6).
[0062]
First, it is determined whether or not the busy lock dedicated process queue 13 is empty (step 601).
[0063]
If it is determined in step 601 that “the busy lock dedicated process queue 13 is empty”, control is passed to the execution process selection means 4 in the own processor (step 608).
[0064]
On the other hand, when it is determined in step 601 that “the busy lock dedicated process queue 13 is not empty”, a set of the process number and instruction address (busy lock dedicated process information) is extracted from the busy lock dedicated process queue 13 (step 51). 602) Attempts to execute each instruction in the program to which the process is assigned in order from the instruction specified by the instruction address.
[0065]
When each instruction is executed, the busy lock address storage area 10 is referred to before the instruction is executed, and the instruction indicates that the busy lock target resource use process has ended (unlock instruction for the busy lock target resource). It is determined whether or not. That is, whether the instruction is an unlock instruction and whether the unlock address of the unlock instruction (information indicating the location of the resource to be unlocked) is equal to the busy lock address in the busy lock address storage area 10 It is determined whether or not (step 603).
[0066]
If it is determined in step 603 that “the command is an unlock command indicating the end of busy lock target resource use processing”, the set of the process number and the command address of the command is executed without executing the command. (General process information) is registered in the general process queue 12 (step 605), and control is passed to the execution process selection means 4 (step 608).
[0067]
On the other hand, if it is determined in step 603 that “the command is not an unlock command indicating the end of busy lock target resource use processing”, the command is executed (step 604), and the command is the last command in the program. It is determined whether it is (final instruction) (step 606).
[0068]
If it is determined in step 606 that “the instruction is the last instruction of the program”, control is passed to the execution process selection means 4 (step 608).
[0069]
On the other hand, if it is determined in step 606 that “the instruction is not the last instruction of the program”, the instruction address is advanced to the next instruction (step 607), and the “next instruction” is targeted as the step 603 and subsequent steps. Repeat the process and determination.
[0070]
Next, a specific operation of the multiprocessor system according to the present embodiment will be described with reference to FIG. 7 and the above-described flowcharts (FIGS. 2 to 6).
[0071]
In the example illustrated in FIG. 7, it is assumed that there are two processors 100 (processor 701 and processor 702), and their processor numbers are “01” and “02”, respectively.
[0072]
Further, the processor 702 is a busy lock dedicated processor, and the address (busy lock address) of the busy lock target resource 703 is set to L.
[0073]
Further, in the program 704 that the user instructs to start, the start instruction address is set to PA1, the address of the lock instruction (the lock instruction for the busy lock target resource 703) for starting the busy lock target resource use process is set to PA2, and the busy lock target resource Let PA3 be the address of an unlock command for ending the use process (unlock command for the busy lock target resource 703).
[0074]
First, the operation at system startup will be described.
[0075]
When the system is started up, the initial setting means 2 in the processor 701/702 performs the following processing.
[0076]
In step 201 of FIG. 2, L of the address (busy lock address) of the busy lock target resource 703 is stored in the busy lock address storage area 10, and in step 202, the busy lock dedicated processor number storage area 11 stores the busy lock dedicated processor number. Stores the processor number “02”.
[0077]
In step 203, the general-purpose process queue 12 is emptied, and in step 204, the busy lock dedicated process queue 13 is emptied.
[0078]
Secondly, an operation when the user gives an instruction to start the program 704 will be described.
[0079]
When the user gives an instruction to start the program 704, the process start means 3 in the processor 701/702 assigns a process number to the program 704 in step 301 of FIG. 3 (in this example, this process number is assigned to “P1 In step 302, general process information indicating a set of the process number P1 and the start instruction address PA1 of the program 704 is registered in the general process queue 12.
[0080]
Third, when the processor 701 (the processor 100 whose processor number is “01”) executes the process P1 assigned to the program 704 (from the start instruction (the instruction at the address PA1) in the program 704), the lock instruction at the address PA2 Will be described.
[0081]
At this time, the execution process selection means 4 in the processor 701 determines that the processor number “01” of its own processor and the processor number “02” (busy lock dedicated processor in the busy lock dedicated processor number storage area 11) in step 401 of FIG. In step 402, control is passed to the general-purpose process execution means 5 in its own processor in order to determine that the two are not equal.
[0082]
The general-purpose process execution means 5 determines whether or not the general-purpose process queue 12 is empty in step 501 of FIG.
[0083]
In this case, since the general process information consisting of the combination of the process number P1 of the process P1 and the instruction address PA1 is registered in the general process queue 12 by the process start unit 3, the general process execution unit 5 In 501, it is determined that “the general-purpose process queue 12 is not empty”.
[0084]
Therefore, the general-purpose process execution means 5 takes out the general-purpose process information consisting of the set of the process number P1 and the instruction address PA1 from the general-purpose process queue 12 in Step 502, and in Step 503, the instruction at the address PA1 is a lock instruction. Further, it checks (determines) whether or not the lock address is equal to the busy lock address L in the busy lock address storage area 10.
[0085]
In this case, since it can be determined that “the instruction at the address PA1 (the start instruction of the program 704) is not the lock instruction with the lock address L”, the general-purpose process execution means 5 executes the instruction at the address PA1 in step 504. In step 506, it is checked (determined) whether the instruction is the last instruction (final instruction) of the program.
[0086]
In this case, since it can be determined that “the instruction is not the final instruction of the program (address PA1 is not the end of the program)”, the general process execution means 5 advances the instruction address to the next instruction in step 507. Then, the process returns to the determination in step 503 for the “next instruction”.
[0087]
Thereafter, the general-purpose process execution means 5 performs the processes and determinations of Step 503, Step 504, Step 506, and Step 507 until the instruction address advances to the address PA2 of the lock instruction that starts the process using the busy lock target resource 703. Repeat each instruction in the program 704 sequentially.
[0088]
When the instruction address advances to PA2, the general-purpose process execution means 5 determines in step 503 that "the instruction at the address PA2 is a lock instruction and the lock address is equal to L in the busy lock address storage area 10". Therefore, in step 505, the set of process number P1 and instruction address PA2 is registered in the busy lock dedicated process queue 13 as busy lock dedicated process information, and in step 508, control is passed to the execution process selection means 4.
[0089]
Fourth, when the processor 702 (the processor 100 whose processor number is “02”) executes the process P1 assigned to the program 704 (from the lock instruction at the address PA2 to the immediately before the unlock instruction at the address PA3 in the program 704). The operation up to the time of the instruction will be described.
[0090]
At this time, the execution process selection means 4 in the processor 702 compares the processor number “02” of the own processor with the processor number “02” in the busy lock dedicated processor number storage area 11 in step 401 of FIG. In order to determine that “both are equal”, in step 403, control is passed to the busy lock execution means 6 in its own processor.
[0091]
In step 601 of FIG. 6, the busy lock execution means 6 determines whether or not the busy lock dedicated process queue 13 is empty.
[0092]
In this case, since the general-purpose process execution means 5 has just registered the busy lock dedicated process information consisting of the process number P1 of the process P1 and the instruction address PA2 in the busy lock dedicated process queue 13, the busy lock execution is executed. The means 6 determines in step 601 that “the busy lock dedicated process queue 13 is not empty”.
[0093]
Accordingly, the busy lock execution means 6 takes out the busy lock dedicated process information consisting of the combination of the process number P1 and the instruction address PA2 from the busy lock dedicated process queue 13 in step 602, and in step 603, the instruction of the address PA2 is unloaded. It is checked (determined) whether it is a lock command and its unlock address is equal to the busy lock address L in the busy lock address storage area 10.
[0094]
In this case, since it can be determined that “the instruction at the address PA2 is not an unlock instruction with the unlock address L”, the busy lock execution means 6 executes the instruction at the address PA2 in step 604, and in step 606. Then, it is checked (determined) whether or not the instruction is the last instruction (final instruction) of the program.
[0095]
In this case, since it can be determined that “the instruction is not the final instruction of the program”, the busy lock execution means 6 advances the instruction address to the next instruction in step 607 and targets the “next instruction”. Return to the determination of step 603.
[0096]
Thereafter, the busy lock execution means 6 performs the processes and determinations of steps 603, 604, 606, and 607 until the instruction address proceeds to the address PA3 of the unlock instruction that ends the process of using the busy lock target resource 703. Are repeated, and each instruction in the program 704 is sequentially executed.
[0097]
When the instruction address advances to PA3, the busy lock execution means 6 determines in step 603 that "the instruction at the address PA3 is an unlock instruction and the unlock address is equal to L in the busy lock address storage area 10". In order to make a determination, in step 605, the set of the process number P1 and instruction address PA3 is registered in the general process queue 12 as general process information, and in step 608, control is passed to the execution process selection means 4.
[0098]
Fifth, the operation when the processor 701 executes the process P1 assigned to the program 704 (when executing from the unlock instruction to the final instruction of the address PA3 in the program 704) will be described.
[0099]
At this time, the execution processing selection means 4 in the processor 701 compares the processor number “01” of its own processor with the processor number “02” in the busy lock dedicated processor number storage area 11 in step 401 of FIG. In order to determine that “they are not equal”, in step 402, control is passed to the general-purpose process execution means 5 in the processor.
[0100]
The general-purpose process execution means 5 determines whether or not the general-purpose process queue 12 is empty in step 501 of FIG.
[0101]
In this case, since the general-purpose process information consisting of the set of the process number P1 of the process P1 and the instruction address PA3 is registered in the general-purpose process queue 12 by the busy lock execution means 6, the general-purpose process execution means 5 In step 501, it is determined that “the general-purpose process queue 12 is not empty”.
[0102]
Therefore, the general-purpose process execution means 5 takes out the general-purpose process information comprising the set of the process number P1 and the instruction address PA3 from the general-purpose process queue 12 in step 502, and in step 503, the instruction at the address PA3 is a lock instruction. Further, it checks (determines) whether or not the lock address is equal to the busy lock address L in the busy lock address storage area 10.
[0103]
In this case, since it can be determined that “the instruction at the address PA3 is not a lock instruction with the lock address L”, the general-purpose process execution means 5 executes the instruction at the address PA3 in step 504, and in step 506, It is checked (determined) whether the instruction is the last instruction (final instruction) of the program.
[0104]
In this case, since it can be determined that “the instruction is not the final instruction of the program”, the general-purpose process execution unit 5 advances the instruction address to the next instruction in step 507 and targets the “next instruction”. Return to the determination of step 503.
[0105]
Thereafter, the general-purpose process execution means 5 repeats the processing and determination of step 503, step 504, step 506, and step 507 until the instruction address advances to the final instruction of the program 704, and sequentially executes each instruction in the program 704. To go.
[0106]
When the instruction address advances to the final instruction of the program 704, the general-purpose process execution unit 5 determines in step 506 that "the instruction to be determined is the final instruction of the program". Pass control to.
[0107]
As described above, in this example, in the execution of the program 704, only the processing using the busy lock target resource 703 (processing from the instruction at the address PA2 to the instruction immediately before the instruction at the address PA3) is a busy lock dedicated processor (processor The other processing in the program 704 is executed by the processor 701.
[0108]
Similarly, when a “program including processing using the busy lock target resource 703” other than the program 704 is started, only the processing using the busy lock target resource 703 is executed by the busy lock dedicated processor (processor 702). The Therefore, the processing using the busy lock target resource 703 is always executed by the processor 702, and the processing is always performed using the cache memory 101 of the processor 702. Therefore, the processing is performed using the busy lock target resource 703. The cache hit rate of the memory area to be increased increases, and the system throughput can be improved.
[0109]
As a modification of the first embodiment, the initial setting unit 2 and the process start unit 3 are provided only in a specific processor 100 (not limited to a busy lock dedicated processor). The setting process and the process start process can be performed only by the specific processor 100.
[0110]
(2) Second embodiment
[0111]
FIG. 8 is a block diagram showing the configuration of the second exemplary embodiment of the present invention.
[0112]
Referring to FIG. 8, the second embodiment of the present invention is different from the multiprocessor system according to the first embodiment shown in FIG. 1 in that a process execution control program 800 is provided.
[0113]
The process execution control program 800 is read by each processor 100, and the operation of each processor 100 is controlled by means 1 (initial setting means 2, process start means 3, execution process selection means 4, general-purpose process execution means 5, and busy lock. Control as execution means 6). The operation of each processor 100 under the control of the process execution control program 800 is performed by means 1 (table initial setting means 2, process start means 3, execution process selection means 4, general-purpose process) in each processor 100 in the first embodiment. Since the operations are the same as those of the execution means 5 and the busy lock execution means 6), a detailed description thereof will be omitted.
[0114]
【The invention's effect】
As described above, according to the present invention, in a multiprocessor system composed of a plurality of processors each having a cache memory for each processor, the busy lock target resource use processing is executed only by a specific processor (busy lock dedicated processor). The information on the memory area referenced in the busy lock target resource use processing can be kept in the cache memory of the busy lock dedicated processor, thereby increasing the cache hit rate and improving the system throughput. The effect that it can be made arises.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a multiprocessor system according to a first embodiment of the present invention.
FIG. 2 is a flowchart showing processing of initial setting means in FIG. 1;
FIG. 3 is a flowchart showing processing of a process start unit in FIG. 1;
4 is a flowchart showing processing of an execution processing selection unit in FIG. 1. FIG.
FIG. 5 is a flowchart showing processing of a general-purpose process execution unit in FIG.
6 is a flowchart showing processing of busy lock execution means in FIG. 1. FIG.
7 is a diagram for explaining a specific operation of the multiprocessor system shown in FIG. 1; FIG.
FIG. 8 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.
FIG. 9 is a block diagram showing a configuration of an example of a conventional multiprocessor system (the invention described in Japanese Patent Laid-Open No. 07-325791).
[Explanation of symbols]
1 Means
2 Initial setting means
3 Process start means
4 Execution process selection means
5 General-purpose process execution means
6 Busy lock execution means
10 Busy lock address storage area
11 Busy lock dedicated processor number storage area
12 General-purpose process queue
13 Process queue for busy lock
100 processor
101 Cache memory
200 memory
800 Process execution control program

Claims (3)

A multiprocessor system including a plurality of processors each including a cache, an initial setting unit, a process start unit, an execution process selection unit, a general-purpose process execution unit, and a busy lock execution unit, and a memory connected to the processor,
Stores the “busy lock target resource address”, which is the instruction address of the processing that uses the busy lock target resource, which is a frequently locked resource, in the “busy lock address storage area” of the memory, and uses the busy lock target resource The initial setting means for storing a “busy lock dedicated processor number” indicating a busy lock dedicated processor that executes only the processing to be performed in the “busy lock dedicated processor number storage area” of the memory;
The process start means for assigning a process number to a program and registering a set of a process number and a start instruction address of the program as general process information in a “general process queue” of the memory;
If the own processor number is equal to the “busy lock dedicated processor number” in the “busy lock dedicated processor number storage area”, control is passed to the busy lock execution means, and if not equal, control is passed to the general-purpose process execution means. The execution process selection means;
If the “general process queue” is empty, control is passed to the execution process selection means. If the “general process queue” is not empty, the general process information is extracted from the “general process queue”, the instruction is a lock instruction, and “ If it is equal to the “busy lock target resource address” in the “busy lock address storage area”, a set of process number and instruction address is registered in the “busy lock dedicated process queue” of the memory as dedicated process information, and the execution processing selection means If the instruction is not a lock instruction or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the instructions are sequentially executed up to the last instruction of the program, and the execution processing selecting means Control or a lock command, and the “busy lock address storage area” Said generic process executing means for sequentially executing until it equals a click target resource address ",
If the “busy lock dedicated process queue” is empty, control is passed to the execution process selection means. If the “busy lock dedicated process queue” is not empty, the dedicated process information is extracted from the “busy lock dedicated process queue” and the instruction is unloaded. If it is a lock instruction and is equal to the “busy lock target resource address” in the “busy lock address storage area”, the combination of the process number and instruction address is registered in the “general process queue” of the memory as general process information. Control is passed to the execution processing selection means, and if the instruction is not an unlock instruction or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the instructions are sequentially executed up to the last instruction of the program. Then, control is passed to the execution process selection means or an unlock command is issued, and “busy lock Said busy locking execution means for sequentially executed until equal to "busy lock target resource address" in the address storage area "within
A multiprocessor system comprising the processor comprising: Process control method in a multiprocessor system including a plurality of processors each including a cache, an initial setting unit, a process start unit, an execution process selection unit, a general-purpose process execution unit, and a busy lock execution unit, and a memory connected to the processor Because
The initial setting means stores a “busy lock target resource address”, which is a command address for processing using a busy lock target resource, which is a resource frequently locked, in the “busy lock address storage area” of the memory, A procedure of storing a “busy lock dedicated processor number” indicating a busy lock dedicated processor that executes only a process using the busy lock target resource in the “busy lock dedicated processor number storage area” of the memory;
The process starting means assigns a process number to a program and registers a set of a process number and a program start instruction address as general process information in the “general process queue” of the memory;
The execution process selection means passes control to the busy lock execution means when its own processor number is equal to the “busy lock dedicated processor number” in the “busy lock dedicated processor number storage area”. A procedure for passing control to the process execution means;
If the “general process queue” is empty, the general process execution means transfers control to the execution process selection means. If the “general process queue” is not empty, the general process execution means takes out the general process information from the “general process queue”. If it is a lock command and is equal to the “busy lock target resource address” in the “busy lock address storage area”, a set of process number and command address is stored in the “busy lock dedicated process queue” of the memory as dedicated process information. If the command is not a lock command or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the commands are sequentially transferred to the last command of the program. Execute and pass control to the execution process selection means, or is a lock command and “busy lock A step of sequentially executed until it equals the "busy lock target resource address" Les storage area "within
The busy lock execution means gives control to the execution process selection means when the “busy lock dedicated process queue” is empty, and if the “busy lock dedicated process queue” is not empty, the dedicated process starts from the “busy lock dedicated process queue”. If the information is taken out and the command is an unlock command and is equal to the “busy lock target resource address” in the “busy lock address storage area”, the combination of the process number and the command address is used as general process information in the “ Register in the "general-purpose process queue" and transfer control to the execution process selection means. If the instruction is not an unlock instruction or is not equal to the "busy lock target resource address" in the "busy lock address storage area", program the instruction Are executed sequentially until the last instruction of the program is passed to the execution process selection means, or unlocked. A step of sequentially executed until a decree, and equal to "busy lock target resource address" in the "busy lock address storage area" within
A process control method comprising: Process control program in a multiprocessor system including a plurality of processors each including a cache, an initial setting unit, a process start unit, an execution process selection unit, a general-purpose process execution unit, and a busy lock execution unit, and a memory connected to the processor Because
In the initial setting means, a “busy lock target resource address” that is a command address of processing that uses a busy lock target resource that is a resource that is frequently locked is stored in the “busy lock address storage area” of the memory, Executing a procedure of storing a “busy lock dedicated processor number” indicating a busy lock dedicated processor that executes only a process using the busy lock target resource in the “busy lock dedicated processor number storage area” of the memory;
The process start means assigns a process number to a program, and executes a procedure for registering a set of a process number and a program start instruction address as general process information in the “general process queue” of the memory,
If the own processor number is equal to the “busy lock dedicated processor number” in the “busy lock dedicated processor number storage area”, the control is passed to the busy lock executing means. Execute the procedure to pass control to the process execution means,
If the “general process queue” is empty, control is passed to the execution process selection means, and if the “general process queue” is not empty, the general process information is extracted from the “general process queue” If it is a lock command and is equal to the “busy lock target resource address” in the “busy lock address storage area”, a set of process number and command address is stored in the “busy lock dedicated process queue” of the memory as dedicated process information. If the command is not a lock command or is not equal to the “busy lock target resource address” in the “busy lock address storage area”, the commands are sequentially transferred to the last command of the program. Execute and pass control to the execution process selection means, or is a lock command and “busy lock To execute the steps of sequentially executed until it equals the "busy lock target resource address" Les storage area "within
The busy locking execution unit, only from the "busy lock dedicated process Queue" passes the control to the execution processing selection means to be empty, the "busy lock dedicated process queue" is not empty "busy lock dedicated process queue" process If the information is taken out and the command is an unlock command and is equal to the “busy lock target resource address” in the “busy lock address storage area”, the combination of the process number and the command address is used as general process information in the “ Register in the "general-purpose process queue" and transfer control to the execution process selection means. If the instruction is not an unlock instruction or is not equal to the "busy lock target resource address" in the "busy lock address storage area", program the instruction Are executed sequentially until the last instruction of the program is passed to the execution process selection means, or unlocked. A decree, and to execute the steps of sequentially executed until it equals the "busy lock target resource address" in the "busy lock address storage area",
A process control program characterized by that.

Images