JP3732648B2 - Process allocation method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a process allocation for dynamically allocating a process including a plurality of subprocesses as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series.On the wayRelated.
[0002]
In recent years, in fields where real-time processing is required, such as digital communication systems and multimedia systems, high-speed, low-cost, digital signal processing processors (DSPs) that can flexibly handle complex algorithms, Processors that can execute processing at high speed are widely used.
[0003]
However, when trying to process a plurality of processes such as image processing and sound processing CG at the same time, it is difficult to perform processing in real time with a single processor. Further, even in processing realized by a single processor, if the amount of data to be processed increases or the required speed increases, it may not be realized by a single processor. Therefore, it is necessary to increase the processing speed by using a plurality of processors.
[0004]
[Prior art]
A conventional example will be described below with reference to the drawings.
§1: Description of Conventional Example 1
Conventionally, in a multiprocessor system, when assigning a process to each processor, a method of assigning a process in order from an available processor is used. This is because, in the linear array type multiprocessor system, when the communication between processors is performed, there is only one transfer path (linear) or two (ring), so it is unnecessary to consider the communication time.
[0005]
In addition, such a method of finding free processors and sequentially assigning them has an advantage that time required for important assignments is small in dynamic process assignment. However, depending on the application, there is a case where the amount of data transferred through the link used for data transfer between processors is considerably large. In such a case, the amount of data transferred on a link between certain processors becomes large, and other processors The data transfer flow of the link between them may be reduced.
[0006]
In such a case, although the processor capacity is still sufficiently free, data transfer cannot be performed, and the entire processing may not be realized in real time.
[0007]
§2: Description of Conventional Example 2
As an allocation method considering the data transfer amount, for example, there is an invention described in JP-A-9-34847. Hereinafter, this example will be described as Conventional Example 2. In Conventional Example 2, a switch is placed on a communication path, the data transfer amount is counted with the switch, and the communication path is changed accordingly, thereby assigning a process in consideration of the data transfer amount.
[0008]
However, this process allocation method is effective if there are a plurality of communication paths, but in the linear array type processor system, there is only one communication path and only two link configurations, and there is almost no effect.
[0009]
§3: Description of Conventional Example 3
There has already been proposed a process that can realize processing utilizing the capability of a processor by allocating a process that minimizes the maximum data transfer amount between processors. However, since each process allocated here generally requires a plurality of processor resources, a plurality of sub-processes to be processed by a single processor are connected.
[0010]
Therefore, the resource required by the process varies depending on the execution schedule of the sub-process. As for the execution schedule of such a sub-process, a method is known in which a control group is set in advance and the inside of the group is scheduled, as disclosed in JP-A-8-152903.
[0011]
However, in this method, the control group is statically determined first, so in order to distribute the load of the data transfer amount between the groups, it is necessary to change the allocation of the control group each time. Not done.
[0012]
§4: Description of Conventional Example 4
Among systems composed of a plurality of processors, a linear array type multiprocessor system obtained by connecting a plurality of processors in series is Japanese Patent Application No. 9-221617 (hereinafter referred to as “conventional example 4”) previously filed. It is described in. The linear array type multiprocessor system shown in the conventional example 4 has a simple configuration and can constitute the cheapest system.
[0013]
In the linear array type multiprocessor system, the processor can directly transfer data to one of the left and right processors, but data transfer to the other processors can be performed only via other processors. Therefore, when a plurality of processors are dynamically processed or stopped according to a user's request, the ability of the plurality of processors cannot be fully utilized depending on how the processes are allocated.
[0014]
Hereinafter, the outline of Conventional Example 4 will be described with reference to the drawings. FIG. 8 is an explanatory diagram (part 1) of a conventional example, and is an explanatory diagram of a signal processing accelerator (linear array type processor system). The configuration of the signal processing accelerator (linear array type processor system) will be described below with reference to FIG.
[0015]
The signal processing accelerator shown in FIG. 8 includes a plurality of identical information processing units 10. The information processing units 10 are connected to each other and to the host memory bus 30. Each information processing unit 10 includes a signal processor 11, an instruction cache 12, a data RAM 13, link control units 14 and 15, a main cache 16, a link cache 17, and a DRAM controller 19. The signal processor 11 and the data RAM 13 constitute a signal processing unit 25. Further, the link control units 14 and 15, the main cache 16, and the link cache 17 constitute a communication control unit 26. A communication link 20 is connected to the link controllers 14 and 15. Each information processing unit 10 is arranged in series via the communication link 20 and communicates with the adjacent information processing unit 10. By communicating the communication contents from one information processing unit 10 to the next information processing unit 10, communication can be performed between any information processing units. In this case, although three information processing units 10 are shown in FIG. 8, the number is not limited to three and is arbitrary.
[0016]
Each information processing unit 10 is connected to the host memory bus 30 via the DRAM controller 19. Further, a host processor 31 is connected to the host memory bus 30. The signal processor 11 is a processor for realizing a signal processing function, and the instruction cache 12 is a cache memory for storing instructions frequently used by the signal processor 11.
[0017]
A program executed by the signal processor 11 is stored in the DRAM 18 in addition to the instruction cache 12. The data RAM 13 is used as a work area for storing intermediate results when the signal processor 11 processes data. The main cache 16 and the link cache 17 are cache memories for storing data to be processed by the signal processor 11.
[0018]
The main cache 16 stores data fetched from the DRAM 18 of the information processing unit 10 itself, and the link cache 17 stores data fetched from other information processing units 10 via the link control units 14 and 15.
[0019]
Even if the data in the main cache 16 is swapped out, if the data becomes necessary, the data can be read again from its own DRAM 18. On the other hand, when the data in the link cache 17 is swapped out, it is necessary to bring the data from another information processing unit 10 through the communication link 20 again.
[0020]
If the main cache 16 and the link cache 17 are the same cache memory, for example, the data acquired from the other information processing unit 10 by storing the data from its own DRAM 18 in the cache memory even when the communication load is heavy. Problems such as swapping out. In this example, the main cache 16 and the link cache 17 are divided by function.
[0021]
When performing data processing, the plurality of information processing units 10 perform parallel processing or pipeline processing while communicating with each other. For example, while some information processing units 10 execute image data processing in parallel, some other information processing units 10 can process audio data in parallel.
[0022]
Communication between the plurality of information processing units 10 is performed by the communication link 20. Therefore, the host memory bus 30 is not involved in communication between the plurality of information processing units 10 and can provide a data transfer path to other processes such as an OS process executed by the host processor 31.
[0023]
Each information processing unit 10 writes the processed data in the DRAM 18. The host processor 31 can write data after data processing by accessing the DRAM 18 via the host memory bus 30. The signal processing accelerator of FIG. 8 includes a plurality of information processing units 10 that can communicate without interposing the host memory bus 30 and causes the information processing units 10 to execute parallel processing. Therefore, data processing speed is reduced due to bus contention. Thus, high-speed signal processing can be realized.
[0024]
Also, since a different information processing unit 10 can be assigned to each of a plurality of processes such as an image processing process and a sound processing process, it is suitable for multimedia signal processing that requires processing of a plurality of different signals. . In addition, the signal processor 25 (signal processor 11, instruction cache 12, data RAM 13), communication controller 26 (cache memories 16 and 17, and link controllers 14 and 15), and memory (DRAM 18 and DRAM controller 19). Can be integrated on a single chip and mounted on a personal computer in the same form as a conventional memory.
[0025]
Therefore, there is an advantage that the cost can be overlapped with the conventional memory bus and the signal processing accelerator embedded in the memory bus can be utilized by software. Accordingly, it is possible to reduce the cost for adding and expanding hardware and to construct a system excellent in function expansion.
[0026]
§5: Explanation of process allocation method ... See FIG.
FIG. 9 is an explanatory diagram of the conventional example (part 2), and shows two different process allocation methods (A in FIG. 1 and Example 2 in FIG. B). Hereinafter, two different process allocation methods will be described with reference to FIG. A case where process 1 is assigned to processor elements PE1 and PE3 and process 2 is assigned to processor elements PE2 and PE4 as shown in FIG. 9A will be described.
[0027]
If the data transfer amount between two PEs in one processor is M, the data transfer amount between PE1 and PE3 via PE2 is M, and data transfer between PE2 and PE4 via PE3 The amount is also M. Therefore, the data transfer amount is M between PE1 and PE2, 2M between PE2 and PE3, and M between PE3 and PE4.
[0028]
Next, as shown in FIG. 9B, the case where the process 1 is assigned to the processor elements PE1 and PE2 and the process 2 is assigned to the processor elements PE3 and PE4 will be described. In this case, the data transfer amount between PE1 and PE2 is M, the data transfer amount between PE3 and PE4 is M, and no data transfer is performed between PE2 and PE3.
[0029]
If the data transfer capability between PEs is, for example, 1.5M, in the case of FIG. 9A (Example 1), the processing of one process cannot be realized, whereas FIG. In the case of FIG. 9B (example 2), both processes can be realized simultaneously. As described above, the data transfer amount of each link varies depending on how the processes are allocated, and there are cases where complete simultaneous processing of a plurality of processes is possible and impossible.
[0030]
If simultaneous processing is not possible, the overall data processing speed naturally deteriorates. In addition, since the number and timing required from other processor elements are completely unknown before an actual request is issued, it is necessary to dynamically allocate processes. Therefore, an efficient dynamic process allocation algorithm is required.
[0031]
§6: Description of dynamic process allocation algorithm ... See FIGS.
10 to 13 are explanatory diagrams (No. 3) to (No. 6) of the conventional example, showing a dynamic process allocation algorithm. Hereinafter, a dynamic process allocation algorithm will be described with reference to FIGS.
[0032]
(1): Description of the main routine ... See FIG.
FIG. 10 is an explanatory diagram (part 3) of the conventional example and is a flowchart showing a main routine of the dynamic process allocation algorithm. The main routine of the dynamic process allocation algorithm will be described below with reference to FIG. S1 to S5 indicate each processing step.
[0033]
This dynamic process allocation algorithm is a process executed by a resource management program (not shown). This dynamic process allocation algorithm allocates a resource (in this case, the resource indicates a PE) based on the following two indexes. The first indicator is that the data transfer of the assigning process does not overlap with other data transfers as much as possible. The second index is that, when the next process is assigned as a result of assigning the process, it does not overlap with other data transfers as much as possible.
[0034]
First, the first index is set so that the maximum value of the data transfer amount on the communication link 20 caused by the process to be allocated is minimized. Next, among the allocation methods having the same maximum value, an allocation method is obtained by using the second index so that the next allocation process is not inhibited as much as possible.
[0035]
As shown in FIG. 10, in this algorithm, the allocation method is separately obtained when one PE is requested and when a plurality of PEs are requested. This is because when only one PE is requested, data transfer on the communication link 20 by the process does not occur, and therefore the influence on the next process allocation may be considered.
[0036]
On the other hand, when a plurality of PEs are allocated, since data transfer via the communication link 20 is necessary, the efficiency of the process itself is affected by how the processes are allocated.
[0037]
In step S1 of FIG. 10, the number of PEs that can be used as free resources is checked. If there is no PE that can be used, the algorithm is terminated, and if it exists, the process proceeds to step S2. In step S2, it is determined whether the required number of PEs is one. If there is one, the process proceeds to step S3, and if there is more than one, the process proceeds to step S4.
[0038]
In step S3, an assignment is made for one PE request. If the allocation is unsuccessful, the algorithm is terminated. If the allocation is successful, the process proceeds to step S5. In step S4, allocation is performed for a plurality of PE requests. If the allocation fails, the algorithm is terminated. If the allocation is successful, the process proceeds to step S5. In step S5, the process ID is updated. That is, a new process ID is assigned to a new process. This ends the algorithm.
[0039]
(2): Explanation of allocation for one PE request-see Fig. 11
FIG. 11 is an explanatory diagram (part 4) of the conventional example, and is a flowchart showing an allocation process for one PE request in step S3 of FIG. In addition, S11-S17 show each process step.
[0040]
In step S11, an available PE is searched. Next, in step S12, a loop is formed for all available PEs. That is, the following steps are repeated for all available PEs. In step S13, one PE is provisionally allocated. In step S14, the next allocation efficiency is calculated. The result of calculating the allocation efficiency is referred to as “Result”.
[0041]
In step S15, Result holds the minimum value. In step S16, the loop is terminated. In step S17, the PE whose Result is the minimum value is actually assigned to the process. The process ends here.
[0042]
(3): Explanation of allocation for multiple requests of PEs ... see Fig. 12
FIG. 12 is an explanatory diagram of the conventional example (No. 5), and is a flowchart showing allocation of a plurality of PE requests. In the following, allocation of a plurality of PE requests in step S4 in FIG. 10 will be described. In addition, S21-S31 show each process step.
[0043]
In step S21, an available PE is searched. In step S22, the first loop is configured for all combinations of the required number of available PEs. That is, the following steps are repeated for all combinations of the required number of available PEs.
[0044]
In step S23, the data transfer amount of the communication link 20 generated by the allocation of the process is calculated. In step S24, the allocation with the minimum value of the data transfer amount of the communication link 20 is held. In step S25, the first loop is terminated. In step S26, the allocation that minimizes the increase in the data transfer amount of the communication link 20 is set as the allocation selected in the first loop, and the second loop is configured for all the selected allocations.
[0045]
In step S27, a plurality of PEs are provisionally allocated with the selected allocation method. In step S28, the next allocation efficiency is calculated. The result of calculating the next allocation efficiency is denoted as Result. In step S29, the result Result holds the minimum assignment. In step S30, the second loop is terminated. In step S31, the assignment is actually assigned to the process by the assignment method with the smallest Result. The above process ends.
[0046]
 (4): Explanation of the next allocation efficiency calculation process ... See FIG.
FIG. 13 is an explanatory diagram of the conventional example (No. 6), and is a flowchart showing the next allocation efficiency calculation process in step S14 of FIG. 11 and step S28 of FIG. Hereinafter, the above-described allocation efficiency calculation process will be described with reference to FIG. S41 to S43 indicate each processing step.
[0047]
In step S41, the leftmost PE among the available PEs is selected as PE-L. In step S42, the rightmost PE among the available PEs is selected as PE-R. In step S43, the number of communication links from PE-L to PE-R is obtained, and this number of communication links is defined as Result. The process ends here.
[0048]
In the flowchart of FIG. 13, the leftmost PE and the rightmost PE are selected, and the number of communication links between the two PEs is obtained. That is, the allocation efficiency in allocating the next process is determined by the number of communication links. This can be considered as follows. That is, the fact that the number of communication links between both PEs is small means that there are a group of available PEs. Conversely, if the number of communication links between both PEs is large, the available PEs are spread over a wide range.
[0049]
If a process is allocated to PEs grouped in a narrow range, the number of PEs interposed between the allocated PEs is naturally small, and the maximum value of the data transfer amount after allocation is likely to be small. If a process is assigned to PEs that are separated from each other over a wide range, the number of intervening PEs is large, and there is a high possibility that it overlaps with the data transfer of another process. High nature.
[0050]
That is, the flowchart of FIG. 13 gives an indication that the available PEs remaining after the process allocation are in a narrow range as much as possible. In other words, after the process allocation, when the remaining available PE is allocated the next time, an index is given to improve the data transfer efficiency as much as possible.
[0051]
[Problems to be solved by the invention]
The conventional apparatus as described above has the following problems.
(1): In Conventional Example 1, although the processor capacity is still sufficiently open, data transfer cannot be performed, and thus the entire processing may not be realized in real time.
[0052]
(2): In the conventional example 2, the process allocation method is effective if there are a plurality of communication paths, but in the linear array type processor system, there is only one communication path and only two link configurations, and there is almost no effect. .
[0053]
(3): In Conventional Example 3, since the control group is statically determined first, in order to distribute the load of the data transfer amount between the groups, it is necessary to change the allocation of the control group each time. Distribution is not done dynamically.
[0054]
(4): In the dynamic process allocation in the linear array processor system shown in the conventional example 4, the process to be allocated requires a plurality of processor resources. Therefore, in general, each process has a configuration in which a plurality of sub-processes processed by a single processor are connected. Depending on the execution schedule of the sub-process, the number of processors required by the process and the data transfer amount are variable.
[0055]
Therefore, there is a problem that the execution schedule of the sub-process affects the increase / decrease in the data transfer amount between the processors due to the dynamic process allocation, and this makes it impossible to fully demonstrate the capacity of the entire system.
[0056]
In addition, by using the execution schedule of a sub-process that requires a small number of processors and the amount of data transfer required by the process, even if the execution speed of other processes can be satisfied, the processing speed of the single process decreases, There is a problem that processing cannot be realized.
[0057]
The present invention solves such a conventional problem and obtains the execution schedule of a sub-process in advance by a static scheduling means in order to maximize the capacity of the entire system by dynamic process allocation in a linear array type processor system. The purpose is to enable efficient processing.
[0058]
[Means for Solving the Problems]
In order to achieve the above object, the present invention is configured as follows.
  (1): In a process allocation method for dynamically allocating a process including a plurality of sub-processes as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series, A procedure for statically obtaining in advance the execution schedule information including the number of processors required by the process and the amount of data transferred between the processors of the sub-process constituting the process for performing by the static scheduling means of the sub-process, A dynamic process allocating unit that receives execution schedule information of the obtained sub-process and performs dynamic process allocation to each of the processors based on the execution schedule information;In addition, the static scheduling means of the sub-process obtains a sub-process schedule that minimizes the number of processors required for the process to finish within a specified time when performing the static scheduling, and The allocation is attempted with the determined number of processors, and it is determined whether the allocation is possible. If the allocation is impossible, the number of processors is increased to determine whether the allocation is possible again. When it becomes possible, a procedure for obtaining a schedule that minimizes the number of processors required by the process;It is characterized by having.
[0059]
  (2):A process allocation method for dynamically allocating a process including a plurality of sub-processes as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series. A process of statically obtaining execution schedule information including the number of processors required by the process and the amount of data transferred between processors in advance by a static scheduling means of the subprocess, and dynamic process allocation Means for receiving execution schedule information of the obtained sub-process and performing dynamic process allocation to each processor based on the execution schedule information, and static scheduling means for the sub-process The static schedule When performing a ring, a procedure for obtaining a sub-process schedule that minimizes the processing time required to complete the process with the specified number of processors and an operation that divides the total processing time of the sub-processes by the number of target processors Attempts to allocate under the procedure for obtaining the processing time and the condition of processing time T equal to or greater than the obtained processing time, and determines whether or not the allocation is possible. , Repeatedly determining whether allocation is possible again, and when allocation becomes possible, a procedure for obtaining a schedule that minimizes the processing time of the process aloneIt is characterized by having.
[0060]
  (3):A process allocation method for dynamically allocating a process including a plurality of sub-processes as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series. A process of statically obtaining execution schedule information including the number of processors required by the process and the amount of data transferred between processors in advance by a static scheduling means of the subprocess, and dynamic process allocation Means for receiving execution schedule information of the obtained sub-process and performing dynamic process allocation to each processor based on the execution schedule information, and static scheduling means for the sub-process The static schedule When performing ringing, the procedure for dividing the process up to the number of sub-processes, the procedure for scheduling the divided processes and obtaining the execution time of each sub-process, and the time constraint conditions at this time are as follows: If the above condition is not satisfied, another division is attempted again, and it is repeatedly determined whether or not the above condition is satisfied. If the above condition is satisfied, between each division at that time It is determined whether or not the maximum value of the data transfer amount is minimum, and if not, the process is repeated. A procedure for performing the process for all the division patterns and setting the stored maximum data transfer amount to the minimum as a sub-process schedule of the process;It is characterized by having.
[0061]
  (Four) :A process allocation method for dynamically allocating a process including a plurality of sub-processes as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series. A process of statically obtaining execution schedule information including the number of processors required by the process and the amount of data transferred between processors in advance by a static scheduling means of the subprocess, and dynamic process allocation Means for receiving execution schedule information of the obtained sub-process and performing dynamic process allocation to each processor based on the execution schedule information, and static scheduling means for the sub-process Is composed of multiple A plurality of subprocess static scheduling means to select and use a subprocess schedule result optimum for dynamic process allocation from a plurality of subprocess schedules obtained in advance; and When a plurality of pieces of information including a process allocation status to the current processor, a data transfer amount between the current processors, and future process deallocation information are updated for each process and the next process is allocated Selecting and assigning optimal information from the plurality of information;It is characterized by having.
[0065]
(Function)
The operation of the present invention based on the above configuration will be described.
(a): Action of (1) above
The sub-process static scheduling means statically obtains the sub-process schedule information for performing dynamic process allocation in advance. As described above, when dynamic process allocation is performed, schedule information is statically obtained in advance before performing dynamic process allocation for sub-processes constituting the allocation process.
[0066]
Then, based on the execution schedule of the subprocess obtained there, dynamic process allocation is performed based on the number of processors required by the process and the data transfer amount between processors, which contributes to efficient realization of the entire processing. It ’s big.
[0067]
(b): Action of (2) above
When obtaining the sub process schedule information, the sub process static scheduling means obtains the schedule information that minimizes the number of processors required by the process. In this way, when the static scheduling means of a subprocess performs static scheduling, a dynamic process allocation suitable for a system with a small number of processors is obtained by obtaining a subprocess schedule that minimizes the number of processors required by the process. It can be performed.
[0068]
(c): Action of (3) above
When sub-process static scheduling means obtains sub-process schedule information, it obtains schedule information that minimizes the processing time of the single process. In this way, when the static scheduling means of a subprocess performs static scheduling, a dynamic process allocation with a short processing time for all processes is performed by obtaining a subprocess schedule that minimizes the processing time of the process itself. be able to.
[0069]
(d): Action of (4) above
When the sub-process static scheduling means obtains the sub-process schedule information, the sub-process static scheduling means obtains the schedule information that minimizes the inter-processor data transfer amount requested by the process. In this way, when the static scheduling means of the sub-process performs the static scheduling, the sub-process schedule that minimizes the maximum value of the inter-processor data transfer amount required by the process is obtained, so that the inter-processor data transfer is achieved. The dynamic process allocation suitable for a system having a small capacity can be performed.
[0070]
(e): Action of (5) above
The sub-process static scheduling means obtains the sub-process schedule information based on a plurality of conditions, and selects and assigns appropriate schedule information according to the process requirements. .
[0071]
In this way, when the static scheduling means of the sub-process performs static scheduling, a plurality of sub-process schedules are obtained, and the optimum sub-process schedule is selected from among them by the dynamic process allocation. Dynamic process allocation adapted to the allocation status at that time can be performed.
[0072]
(f): Action of (6) above
The static scheduling means of the subprocess of the static scheduling apparatus statically obtains the schedule information of the subprocess for performing dynamic process allocation in advance. As described above, when dynamic process allocation is performed, schedule information is statically obtained in advance before performing dynamic process allocation for sub-processes constituting the allocation process.
[0073]
Then, based on the execution schedule of the subprocess obtained there, dynamic process allocation is performed based on the number of processors required by the process and the data transfer amount between processors, which contributes to efficient realization of the entire processing. It ’s big.
[0074]
(g): Action of (7) above
By reading and executing the program on the recording medium, the computer statically obtains in advance schedule information of sub-processes for performing dynamic process allocation. As described above, when dynamic process allocation is performed, schedule information is statically obtained in advance before performing dynamic process allocation for sub-processes constituting the allocation process.
[0075]
Then, based on the execution schedule of the subprocess obtained there, dynamic process allocation is performed based on the number of processors required by the process and the data transfer amount between processors, which contributes to efficient realization of the entire processing. It ’s big.
[0076]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described in detail below with reference to the drawings.
§1: System description ... See Fig. 1
FIG. 1 is an explanatory diagram of the system. Hereinafter, a system used in the present embodiment will be described with reference to FIG. As shown in FIG. 1, this system is a system for performing dynamic process allocation to a linear array type multiprocessor system.
[0077]
The linear array type multiprocessor system is a system in which a plurality of processors 32 (PE1, PE2, PE3, PE4...) Are connected in series, and is a system having the same configuration as that of the conventional example 4. In this case, each of the processors 32 corresponds to the information processing unit 10 of the conventional example 4.
[0078]
A host processor 31 that performs dynamic process allocation to each processor 32 is connected to each processor 32 by a bus. In addition, a static scheduling device 33 for performing static allocation of sub-processes is connected to the host processor 31 before performing dynamic process allocation to the processors 32 by the host processor 31.
[0079]
The host processor 31 is provided with dynamic process allocation means 34 for performing dynamic process allocation to each processor 32. The static scheduling device 33 includes a sub process for performing static scheduling of sub processes. A process static scheduling means 35 is provided.
[0080]
In this case, the static scheduling device 33 may be directly connected to the linear array processor system by bus coupling or the like, but such a connection is not necessary. For example, the static scheduling device 33 and the host processor 31 need only be in a state where data communication is possible by some means, and are connected via a communication line such as a LAN and used in a state where data can be transmitted to each other. It is also possible to do.
[0081]
That is, the processor 33 may be placed near the host processor 31 (for example, in the same personal computer) or may be placed at a remote location (in a different apparatus). This can be implemented at least if data can be transmitted to a device provided with the dynamic process allocation means 34. In this case, the dynamic process allocation means 34 and the sub-process static scheduling means 35 are realized by executing a program.
[0082]
§2: Explanation of process allocation method ... See Figure 2
FIG. 2 is an explanatory diagram of sub-process processing. Hereinafter, processing of the sub-process will be described with reference to FIG. In the system shown in FIG. 1, the host processor 31 performs dynamic process allocation processing on each processor 32 to perform processing by the linear array processor system.
[0083]
In this case, since each process to be allocated generally requires a plurality of processor resources, a plurality of subprocesses to be processed by a single processor are connected. Execution of this sub-process changes the resources required by that process. In the following description, a case where the subprocess constitutes an element of one process and a plurality of subprocesses exist in one process will be described.
[0084]
As described above, when the host processor 31 performs dynamic process allocation, a plurality of sub-processes constituting the dynamic allocation process are provided in the static scheduling apparatus 33 before performing dynamic process allocation. The sub-process static scheduling means 35 performs static scheduling.
[0085]
The number of processors requested by the process and the inter-processor data transfer amount are transferred to the host processor 31 from the sub-process execution schedule information determined there, and the dynamic process allocation means 34 receives it. Thereafter, the dynamic process allocation unit 34 of the host processor 31 performs dynamic process allocation to each processor 32 based on the number of processors requested by the received process and the data transfer amount between processors. In this way, efficient realization of the entire processing is possible.
[0086]
In this case, as a process allocation method for allocating a process including a sub-process as a component to each processor 32 of the linear array type multiprocessor system, there are the following methods.
[0087]
{Circle around (1)} The process allocation method 1 is a method (time-constrained scheduling) in which the static scheduling means 35 of the sub-process obtains a schedule that minimizes the number of processors required by the process.
[0088]
{Circle around (2)} Process allocation method 2 is a method in which the sub-process static scheduling means 35 obtains a schedule that minimizes the processing time of a single process (scheduling with a limited number of processors).
[0089]
{Circle around (3)} The process allocation method 3 is a method in which the static scheduling means 35 of the sub-process obtains a schedule that minimizes the inter-processor data transfer amount requested by the process.
[0090]
{Circle over (4)} Process allocation method 4 is a method in which schedule information of the sub-process is obtained based on a plurality of conditions, and an appropriate schedule is selected and allocated according to the process requirement conditions.
[0091]
§3: Explanation of process allocation method 1 ... See FIG.
FIG. 3 is a flowchart (part 1) of process allocation processing. The process allocation method 1 is a method in which the sub-process static scheduling means 35 obtains a schedule that minimizes the number of processors required by the process (scheduling with time constraints), and will be described below with reference to FIG. The following processing is processing performed by the static scheduling means 35 of the subprocess, and S51 to S54 indicate each processing step.
[0092]
When performing static scheduling, the sub-process static scheduling means 35 obtains a sub-process schedule that minimizes the number Np of processors required for the process to be completed within a specified time. Therefore, first, the number of processors Np is obtained by the equation Np = (total processing time of sub-processes) / (target processing time) (S51).
[0093]
Next, allocation is attempted by the Np processor (S52), and it is determined whether the allocation is possible (S53). As a result, if the allocation is possible, the process is terminated. If the allocation is impossible, Np is updated (Np = Np + 1) (S54), and the process of S52 is repeated. In this way, it is possible to obtain a schedule that minimizes the number of processors required by the process at the time when assignment becomes possible in the processing of S53.
[0094]
§4: Explanation of process allocation method 2 ... See FIG.
FIG. 4 is a process allocation process flowchart (part 2). The process allocation method 2 is a method (static processor-constrained scheduling) in which the sub-process static scheduling means 35 obtains a schedule that minimizes the processing time of the single process, and will be described below with reference to FIG. The following processing is processing performed by the static scheduling means 35 of the subprocess, and S61 to S64 indicate each processing step.
[0095]
When performing static scheduling, the sub-process static scheduling means 35 obtains a sub-process schedule that minimizes the processing time T for completing the process with the specified number of processors. Therefore, first, the processing time T is obtained by the equation T = (total processing time of subprocesses) / (target number of processors) (S61).
[0096]
Next, allocation is attempted under the condition of processing time ≦ T (S62), and it is determined whether the allocation is possible (S63). As a result, if the allocation is possible, the process is terminated. If the allocation is impossible, T is updated {T = T + (unit time)} (S64), and the process is repeated from S62. In this way, a schedule that minimizes the processing time of a single process can be obtained at the time when assignment is possible in the processing of S63.
[0097]
§5: Explanation of process allocation method 3 ... See FIG.
FIG. 5 is a process allocation process flowchart (part 3). FIG. 6 is an explanatory diagram of process allocation processing (part 1). The process allocation method 3 is a method in which the static scheduling means 35 of the sub-process obtains a schedule that minimizes the inter-processor data transfer amount required by the process, and will be described below with reference to FIGS. The following processing is processing performed by the static scheduling means 35 of the sub-process, and S71 to S79 indicate each processing step. N is the number of sub-processes, and Mi is the number of division patterns for i division.
[0098]
The sub-process static scheduling means 35 first attempts to divide a process from i = 1 to N, where N is the number of sub-processes. That is, i = 1 to N loops (S71), and the process is divided (S72). In this case, if the number of division patterns when i is divided is Mi, division patterns exist from j = 1 to Mi loop (S73).
[0099]
For example, when N = 3 and {1,2,3} is divided into two, there are M2 = 2 patterns of {{1,2}, {3}}, {{1}, {2,3}} . Therefore, j = 1 to Mi loop (S73), the divided processes are scheduled, and the execution start time of each sub-process is obtained (S74). At this time, it is determined whether or not the condition (time constraint) can be satisfied (S75). If the condition is satisfied, the process proceeds to S76. If not satisfied, the process proceeds to S78, and another division is tried.
[0100]
When the above condition is satisfied (when the time constraint is satisfied), it is determined whether or not the maximum value of the data transfer amount between each division at that time is minimum (S76). If not, the process of S78 is performed. If it is minimum, the data transfer amount with the minimum maximum value is stored in a memory or the like (S77).
[0101]
This is repeatedly executed for all division patterns. When the double loop (S78, S79) ends, the schedule stored in the memory or the like is set as the sub-process schedule of the process.
[0102]
Hereinafter, the process of the process allocation method 3 will be described in detail with reference to FIG. The process shown in FIG. 6 includes eight subprocesses {1, 2, 3, 4, 5, 6, 7, 8}. When i = 3, that is, in the case of three divisions, for example, {1, 5, 7}, {2, 4}, {3, 6, 8} are divided.
[0103]
Scheduling is performed on the result, and the start time of each sub-process is determined for the three processors. If there is a schedule that satisfies the condition (time constraint), the transfer amount is calculated, and if the maximum value is smaller than the previously obtained allocation, it is stored.
[0104]
§6: Explanation of process allocation method 4 ... See FIG.
FIG. 7 is an explanatory diagram of process allocation processing (part 2). The process allocation method 4 is a method in which schedule information of sub-processes is obtained based on a plurality of conditions, and an appropriate schedule is selected and allocated according to the process requirements, and will be described below with reference to FIG.
[0105]
In this processing, the sub-process static scheduling means 35 is composed of N pieces, and the sub-processes are statically scheduled by these N means. When performing the static scheduling, an optimum subprocess schedule result at the time of dynamic process allocation is obtained from a plurality of subprocess schedules obtained in advance by the static scheduling means (1 to N) of the N subprocesses. Select and use.
[0106]
1. Process allocation status to the current processor (which processor is free)
2. Current data transfer between processors
3. Future process deallocation information (when, from which processor, which process is deallocated)
In the process, the three pieces of information 1 to 3 are updated every time the process is assigned. When allocating the next process, for example, if there are few processors available, the sub-processor allocation that minimizes the number of requested processors is selected from these three pieces of information, and the process allocation is performed. Also, when the data transfer amount is small, sub-processor allocation that minimizes the data transfer amount is selected to perform process allocation.
[0107]
Furthermore, the sub-process allocation that minimizes the processing time of the next process is used when the sum of this processing time and the time for deallocating is within the time constraints of this process.
[0108]
§7: Description of recording medium and program
The static scheduling processing of the subprocess performed by the static scheduling means 35 of the subprocess of the static scheduling device 33 is realized as follows by the CPU in the static scheduling device 33 executing the program.
[0109]
The static scheduling device 33 is provided with a hard disk device, and a program for realizing the processing, various other data, and the like are stored in a recording medium (hard disk) of the hard disk device. When the processing is performed, the program and data stored in the recording medium of the hard disk device are read out and taken into a memory provided in the static scheduling device 33 under the control of the CPU.
[0110]
Thereafter, the CPU performs the processing by sequentially reading and executing the necessary programs from among the programs stored in the memory. The program to be recorded on the recording medium of the hard disk device is recorded (stored) as follows.
[0111]
(1): A program (program data created by another device) stored in a flexible disk (floppy disk) is read by a flexible disk drive device provided in the static scheduling device 33, and a recording medium (hard disk) of the hard disk device is read. ).
[0112]
{Circle over (2)}: Data stored in a storage medium such as a magneto-optical disk or CD-ROM is read by a drive device provided in the static scheduling device 33 and stored in a recording medium (hard disk) of the hard disk device.
[0113]
(3): Data transmitted from another device via a communication line such as a LAN is received by the static scheduling device 33, and the data is stored in a recording medium (hard disk) of the hard disk device.
[0114]
【The invention's effect】
As described above, the present invention has the following effects.
  (1) :subThe process static scheduling means statically obtains in advance schedule information of sub-processes for performing dynamic process allocation. As described above, when dynamic process allocation is performed, schedule information is statically obtained in advance before performing dynamic process allocation for sub-processes constituting the allocation process.
[0115]
Then, based on the execution schedule of the subprocess obtained there, dynamic process allocation is performed based on the number of processors required by the process and the data transfer amount between processors, which contributes to efficient realization of the entire processing. It ’s big.
[0116]
  (2) :When obtaining the sub process schedule information, the sub process static scheduling means obtains the schedule information that minimizes the number of processors required by the process. In this way, when the static scheduling means of a subprocess performs static scheduling, a dynamic process allocation suitable for a system with a small number of processors is obtained by obtaining a subprocess schedule that minimizes the number of processors required by the process. It can be performed.
[0117]
  (3) :When the sub-process static scheduling means obtains the sub-process schedule information, the sub-process static scheduling means obtains the schedule information that minimizes the processing time of the single process. In this way, when the static scheduling means of a subprocess performs static scheduling, a dynamic process allocation with a short processing time for all processes is performed by obtaining a subprocess schedule that minimizes the processing time of the process itself. be able to.
[0118]
  (Four) :When the sub-process static scheduling means obtains the sub-process schedule information, the sub-process static scheduling means obtains the schedule information that minimizes the inter-processor data transfer amount requested by the process. In this way, when the static scheduling means of the sub-process performs the static scheduling, the sub-process schedule that minimizes the maximum value of the inter-processor data transfer amount required by the process is obtained, so that the inter-processor data transfer is achieved. The dynamic process allocation suitable for a system having a small capacity can be performed.
[0119]
  (Five) :The sub-process static scheduling means obtains the sub-process schedule information based on a plurality of conditions, and selects and assigns appropriate schedule information according to the process requirements. .
[0120]
In this way, when the static scheduling means of the sub-process performs static scheduling, a plurality of sub-process schedules are obtained, and the optimum sub-process schedule is selected from among them by the dynamic process allocation. Dynamic process allocation adapted to the allocation status at that time can be performed.
[0121]
(6): In claim 6, the static scheduling means of the sub-process of the static scheduling device statically obtains the schedule information of the sub-process for performing dynamic process allocation in advance. As described above, when dynamic process allocation is performed, schedule information is statically obtained in advance before performing dynamic process allocation for sub-processes constituting the allocation process.
[0122]
Then, based on the execution schedule of the subprocess obtained there, dynamic process allocation is performed based on the number of processors required by the process and the data transfer amount between processors, which contributes to efficient realization of the entire processing. It ’s big.
[0123]
(7): In claim 7, the computer reads out and executes the program of the recording medium, and statically obtains schedule information of sub-processes for performing dynamic process allocation in advance. As described above, when dynamic process allocation is performed, schedule information is statically obtained in advance before performing dynamic process allocation for sub-processes constituting the allocation process.
[0124]
Then, based on the execution schedule of the subprocess obtained there, dynamic process allocation is performed based on the number of processors required by the process and the data transfer amount between processors, which contributes to efficient realization of the entire processing. It ’s big.
[Brief description of the drawings]
FIG. 1 is an explanatory diagram of a system in an embodiment of the present invention.
FIG. 2 is a process explanatory diagram of a sub process in the embodiment of the present invention.
FIG. 3 is a process allocation process flowchart (part 1) according to the embodiment of the present invention;
FIG. 4 is a process allocation process flowchart (part 2) according to the embodiment of the present invention;
FIG. 5 is a process allocation process flowchart (part 3) according to the embodiment of the present invention;
FIG. 6 is an explanatory diagram (part 1) illustrating process allocation processing according to the embodiment of this invention.
FIG. 7 is a process allocation process explanatory diagram (2) according to the embodiment of the present invention;
FIG. 8 is an explanatory diagram (part 1) of a conventional example.
FIG. 9 is an explanatory diagram (part 2) of a conventional example.
FIG. 10 is an explanatory diagram of a conventional example (part 3);
FIG. 11 is an explanatory diagram of a conventional example (part 4);
FIG. 12 is an explanatory diagram (No. 5) of a conventional example.
FIG. 13 is an explanatory diagram of a conventional example (No. 6).
[Explanation of symbols]
10 Information processing unit
11 Signal processor
12 Instruction cache
13 Data RAM
14, 15 Link control unit
16 Main cache
17 Link cache
18 DRAM
19 DRAM controller
20 Communication link
25 Signal processor
26 Communication control unit
30 Host memory bus
31 Host processor
33 Static scheduling device
34 Dynamic process allocation means
35 Static scheduling means for subprocesses

Claims (4)

In a process allocation method for dynamically allocating a process including a plurality of subprocesses as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series,
Execution schedule information including the number of processors required by the process and the amount of data transferred between processors of the sub-process constituting the process for performing the dynamic process allocation is statically set in advance by the static scheduling means of the sub-process. The steps you're looking for,
A dynamic process allocating unit that receives the execution schedule information of the obtained sub-process and, based on the execution schedule information, performs a dynamic process allocation to each of the processors ;
The static scheduling means of the subprocess is:
When performing static scheduling, a procedure for obtaining a sub-process schedule that minimizes the number of processors required for the process to finish within a specified time; and
Allocation is attempted with the determined number of processors, and it is determined whether the allocation is possible. If the allocation is impossible, the number of processors is increased to determine whether the allocation is possible again. When the process becomes possible, a procedure for obtaining a schedule that minimizes the number of processors required by the process;
A process allocation method characterized by comprising: In a process allocation method for dynamically allocating a process including a plurality of subprocesses as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series,
Execution schedule information including the number of processors required by the process and the amount of data transferred between processors of the sub-process constituting the process for performing the dynamic process allocation is statically set in advance by the static scheduling means of the sub-process. The steps you're looking for,
A dynamic process allocating unit that receives the execution schedule information of the obtained sub-process and, based on the execution schedule information, performs a dynamic process allocation to each of the processors;
The static scheduling means of the subprocess is:
When performing static scheduling, a procedure for obtaining a sub-process schedule that minimizes the processing time for the process to finish with the specified number of processors,
A procedure for obtaining the processing time by performing an operation of dividing the total processing time of sub-processes by the number of target processors;
Allocation is attempted under the condition of the processing time T equal to or greater than the calculated processing time, and it is determined whether the allocation is possible. If the allocation is impossible, the processing time T is updated, and whether the allocation is possible again. The procedure to obtain a schedule that minimizes the processing time of the single process when it becomes possible to assign,
A process allocation method characterized by comprising: In a process allocation method for dynamically allocating a process including a plurality of subprocesses as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series,
Execution schedule information including the number of processors required by the process and the amount of data transferred between processors of the sub-process constituting the process for performing the dynamic process allocation is statically set in advance by the static scheduling means of the sub-process. The steps you're looking for,
A dynamic process allocating unit that receives the execution schedule information of the obtained sub-process and, based on the execution schedule information, performs a dynamic process allocation to each of the processors;
The static scheduling means of the subprocess is:
A procedure for dividing the process up to the number of sub-processes when performing static scheduling;
A procedure for performing scheduling on the divided processes and obtaining an execution time of each sub-process ;
It is determined whether or not the condition of the time constraint at this time is satisfied. If the condition is not satisfied, another division is tried again, and it is repeatedly determined whether or not the condition is satisfied, and the condition is satisfied. In this case, it is determined whether or not the maximum value of the data transfer amount between the divisions at that time is minimum, and if not, the process is repeated, and if it is minimum, the maximum value of the data transfer amount is minimum. Instructions for saving,
Performing the procedure for all the division patterns, and setting the stored maximum data transfer amount as a minimum as a sub-process schedule of the process;
A process allocation method characterized by comprising: In a process allocation method for dynamically allocating a process including a plurality of subprocesses as constituent elements to each processor of a linear array type multiprocessor system in which a plurality of processors are connected in series,
Execution schedule information including the number of processors required by the process and the amount of data transferred between processors of the sub-process constituting the process for performing the dynamic process allocation is statically set in advance by the static scheduling means of the sub-process. The steps you're looking for,
A dynamic process allocating unit that receives the execution schedule information of the obtained sub-process and, based on the execution schedule information, performs a dynamic process allocation to each of the processors;
When the sub-process static scheduling means is composed of a plurality of sub-process schedules determined in advance by a plurality of sub-process static scheduling means, Select and use process schedule results,
In the above procedure, a plurality of pieces of information including the process allocation status to the current processor, the data transfer amount between the current processors, and future process deallocation information are updated every time the process is allocated, and the next process is performed. When assigning, selecting and assigning optimal information from the plurality of information;
A process allocation method characterized by comprising:

Images