JPH0442345A - Distributed processing system - Google Patents

Abstract

PURPOSE:To improve the availability of each processor by making an arithmetic unit which inputs the comparison processing result of data groups finish the comparison of these data groups before each arithmetic unit starts the comparison of data groups and outputs these comparison processing results to the next arithmetic device. CONSTITUTION:The time t1 needed for the comparison processing of a processor P1 is equal to the time t01 needed for a processor P0 to start its comparison processing and then outputs the synthetic extraction data to the processor P1. Meanwhile the time t2 needed for the comparison processing of a processor P2 is equal to the time t12 needed for the processor P1 to start its comparison processing and then outputs the synthetic extraction data to the processor P2. Each processor inputs the synthetic extraction data received from its precedent processor right after the end of the comparison processing and at the same time performs the duplication eliminating synthetic processing to output the synthetic extraction data to the next processor or a main processor 2. Thus each processor is not required to wait for the next processing after the end of the comparison processing. As a result, the availability is improved for each processor.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field to which the invention pertains) The present invention relates to a distributed processing method, and particularly to a distributed processing method that improves the utilization efficiency of each of a plurality of arithmetic units.

(Prior art) Distributed processing methods divide a group of data to be processed among multiple processing units and process the data, so the processing speed is faster and the processing time is shorter than the centralized processing method in which data processing is performed by a single processing unit. It is used in various systems that require processing huge amounts of data.

One type of system that requires processing a huge amount of data in a short period of time is an expert system whose purpose is to obtain answers to required problems. This system determines the disease name by comparing the symptoms of the patient with data constructed based on the knowledge of experts, and displays the answer on the screen.

Conventionally, an expert system for medical diagnosis using a distributed processing method has been configured as shown in FIG. 2, for example.

In the figure, 1 is an inference section, and the main processor 2
and processor Pf! to P2 are connected in a loop, and the main processor 2 is also connected to the memory 3, keyboard 4, and screen display section 5. Further, the knowledge base 6 is a memory that stores matching data in which the knowledge necessary for medical diagnosis by the above-mentioned experts is made into rules, and the processors P2 and P2 are connected to form a system. A program for determining a disease name from a patient's symptoms is written in a memory 3 connected to the main processor 2.

The medical diagnosis expert system configured in this manner operates as follows.

First, according to the program in the memory 3, the main processor 2 asks the patient via the screen display unit 5 whether or not he/she has symptoms necessary to determine the name of the disease.

The patient or operator can use the keyboard 4 according to the patient's symptoms.
When the main processor 2 answers the above question by operating , the main processor 2 outputs the results (hereinafter referred to as inspection data S) to each of the processors PH1 to Pa, and also outputs the matching data group a selected based on the inspection data S. a [! It is divided equally into three parts from a2 to a2, and each is processed by a processor Pl.
Assign to I- to P2.

The processor Pf1 sequentially compares the inspection data S and each data of the matching data group a[], extracts matching data that is equal to the inspection data S, and then removes duplicate data from the extracted data and combines the data. Output to processor P1. The processor P1 sequentially compares the inspection data S and each data of the matching data group a1, and after extracting the matching data that matches the inspection data S, compares this extracted data with the combined extracted data of the processor PlI, and compares the matching data with the matching data group a1. The data is removed, combined, and output to the processor P2. The processor P2 sequentially compares the inspection data S and each data of the matching data group a2, and after extracting matching data that is equal to the inspection data S, compares this extracted data with the combined extracted data from the processor P1, The data is removed, combined, and output to the main processor 2. In addition, the main processor 2 outputs the next inspection data S' to each of the processors Pal to Pa in accordance with the synthesized extracted data, and also outputs the matching data group a° corresponding to the inspection data S° from a'9 to a'. 2, and each is divided into three parts, each of which is processed by processors Pal to P.
2, each processor performs comparison processing and deduplication synthesis processing, and transfers the synthesized extracted data to main processor 2.
Output to. Further, the main processor 2 repeats the above-mentioned processing a required number of times in accordance with the synthetic extraction data from the processor P2, determines the disease name for the patient's symptoms, and displays it on the screen display section 5.

In this way, according to the medical diagnosis expert system using the conventional distributed processing method, as shown in FIG. 3(a), when a single processor performs comparison processing, tp [s]
The main processor 2 processes the matching data group that requires 3
The time required for this comparison process is tp/3 [S], and the matching data is divided into equal parts, and each of the equally divided matching data is compared in parallel by each of the processors P2 to Pa. It is possible to guess the disease name for a patient's symptoms in a short time using an expert system.

However, in an expert system using the distributed processing method described above, each processor performs comparison processing and compares the data extracted by the previous processor with the data extracted by the previous processor. After performing deduplication synthesis processing on data, it is output to the next processor, so if each processor requires time τd [S] when performing the deduplication synthesis processing, processor P1 performs comparison processing. Even if the comparison process is finished, the processor Pl has to wait for τd [s] until the processor P1 outputs the extracted data, and even if the processor P2 finishes the comparison process, it has to wait for 2 s until the processor P1 outputs the extracted data. τd
[There was a problem that the processor had to wait for several seconds, which increased the processing time required to output the extracted data to the main processor, reducing the efficiency of processor utilization.

(Objective of the Invention) 10- The present invention was made to solve the problem of the above-mentioned distributed processing method, and it is possible to improve the utilization efficiency of each processor by shortening the waiting time of each processor. The purpose is to provide a distributed processing method.

(Summary of the invention) In order to achieve the above-mentioned object, the present invention has the following configuration.

That is, the data group is divided and allocated to multiple processing units,
In the means for executing predetermined successive approximation processing by each arithmetic unit, the amount of data group to be allocated to each arithmetic unit is determined according to the time required for processing by each arithmetic unit and the order in which processing results are transmitted. Make it.

(Example) Hereinafter, the present invention will be described in detail based on an example shown in the drawings.

FIG. 1 is a block diagram showing a medical diagnosis expert system according to the distributed processing method of the present invention.

In the figure, reference numeral 1 denotes an inference section, and the system is constructed by connecting a knowledge base 6 in the same manner as a conventional expert system for medical diagnosis. A program is written in the memory 3 to control each component in order to determine a disease name from a patient's symptoms using the distributed processing method of the present invention. In the knowledge base 6, matching data is written in which disease names are defined as rules for patient symptoms.

The medical diagnosis expert system configured as described above operates as follows.

First, the main processor 2 displays the symptoms necessary for determining the disease name on the screen display 5 according to the program in the memory 3.
Ask the patient questions via. When the patient operates the keyboard 4 according to the patient's symptoms, the main processor 2 transmits the test data S according to the patient's symptoms to the processors P8 to P2.
As shown in FIG. 4(a), matching data groups corresponding to the inspection data S are output to each
e into three parts and each is assigned to a processor P[I to P2. Here, the data amount of the matching data group according to the test data S is set as MH[bit]-7=, and each processor combines the test data S and the data amount M[bit].
t] with each data of the matching data group ([S] is set to t=τ(M) −−−−−−−−−−” (1), and M=τ−′ ( t) −−−−1−−−−(2) The time required for each processor to perform overlap synthesis processing is τd
[S], the processor P has matching data bl] of the data amount Me [bit] as shown in the following equation.
Assign.

M[]−τ−1(τ(Mx/3)−τd)
--(3) Also, matching data b1 of data fiM+[b it :] as shown in the following equation is assigned to the processor PI.

M + = τ-1(τ(MX/3))
(4) Matching data b2 of data fiM2 [bit] as shown in the following equation is assigned to the processor P2.

M2=r-'(r(Mx/3)+r,+)-1(5
) The processor P8 sequentially compares the inspection data S and each data of the matching data group b1], extracts matching data that is equal to the inspection data S, removes duplicate data from this extracted data, synthesizes the extracted data, and outputs the matching data to the processor P8. P
Output to 1. The processor P1 sequentially compares the inspection data S and each data of the matching data group b2, extracts matching data that is equal to the inspection data S, and then compares this extracted data with the combined extracted data of the processor P[1]. Processor P removes and synthesizes the data
Output to 2. The processor P2 sequentially compares the inspection data S and each data of the matching data group b3, extracts matching data that is equal to the inspection data S, and then compares this extracted data with the combined extracted data of the processor P1 to find duplicate data. are removed, combined, and output to the main processor 2. The main processor 2 outputs the next test data S' to each of the processors Pit to P2 in accordance with this synthetic extraction data, and divides the matching data group corresponding to the test data S into three parts, each of which is sent to the processors P9 to P2. P2, the comparison process and the duplication removal/synthesis process are repeated as described above, and the disease name corresponding to the patient's symptoms is determined and displayed on the screen display section 5.

According to the expert system using the distributed processing method of the present invention that operates in this manner, the time required for comparison processing by each of Bromo Nosa P8 to P2 is ti+ [S] to t2.
[sl is expressed by the following equation from equations (3) to (5).

j9=τ (Mx/3) −τd −, ---(
6) tl=τ (MX/3)
(7) t2−τ (M/3)+τd −1−−
(8) Also, the time tII+ required from the time the processor P0 starts the comparison process until the output of the composite extraction data to the processor P1 is expressed by the following equation.

tH=ti++τd=r (Mx/3) --(9
) The time t required from when the processor P+ starts the comparison process until it outputs the composite extracted data to the processor P2
12 is expressed as in the following equation.

t 12 = t ++ τd = τ (M / 3) + τd (1
0) That is, in the distributed processing method of the present invention, as shown in FIGS. 4(b) to 4(d), the processor P1 transfers the synthesized extracted data to the processor P1 after the time t1 required for the comparison process and the time t1 required for the processor Pe to start the comparison process. The time ti11 required for the output is equal, and the time t2 required for the comparison process by the processor P2 is equal to the time t12 required for the processor P1 to output the composite extracted data to the processor P2 after starting the comparison process.

Therefore, according to an expert system using the above-mentioned distributed processing method, each processor immediately inputs the composite extraction data from the previous processor after completing the comparison process, performs deduplication composite processing, and calculates the next bromo. Alternatively, since the synthesized extracted data can be output to the main processor, each processor does not need to wait for the next process to be performed after the comparison process is completed, so that the utilization efficiency of each bromo/sosa can be improved.

In this embodiment, the main processor 2 is connected to the three processors in a loop, but the present invention is not limited to this. For example, as shown in FIG. The processor may be connected to the main Bromo Sosa 2, and the processing results may be sequentially output to a processor that is far from the main Bromo Sosa 2 or a processor that is near the main Bromo Sosa 2 to perform the processing as described above.

Further, in the above-described embodiment, the distributed processing method of the present invention is used in an expert system for medical diagnosis, but the present invention is not limited to this, but can also be applied to a system or device that handles data that can calculate the time required for processing according to the amount of data. Can be used for

Furthermore, it is obvious that the distributed processing system of the present invention is also applicable to a system in which all or some of the plurality of processors are replaced with independent computers.

(Effects of the Invention) As explained above, the present invention is such that from the time each arithmetic unit starts comparison processing of a data group to the time the processing result is output to the next arithmetic unit, the arithmetic unit that manually performs this process compares the data group. This is effective in providing a distributed processing method that improves the utilization efficiency of each processor in order to finish processing.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram of an expert system for medical diagnosis using the distributed processing method of the present invention, FIG. 2 is a configuration diagram of an expert system for medical diagnosis using the conventional distributed processing method, and FIGS. d) is a timing chart diagram showing the processing process of the processor in the conventional distributed processing method;
Figures (a) to (d) are timing charts showing the processing steps of the processor in the distributed processing system of the present invention.
The figure is a configuration diagram of an expert system for medical diagnosis showing another embodiment of the present invention. 1... Reasoning section, 2... Main processor, Pl
] to P2...processor, 3...memory, 4
···keyboard. 5... Screen display section, 6... Knowledge base, b... Matching data group, bll to b2... Divided matching data group. Patent applicant: Toyo Tsushinki Co., Ltd.

Claims (2)

[Claims] (1) Divide and allocate a data group to multiple processing units,
In the means for executing predetermined successive approximation processing by each arithmetic unit, the amount of data groups to be allocated to each arithmetic unit is determined according to the time required for processing by each arithmetic unit and the order in which processing results are transmitted. A distributed processing method characterized by: (2) Divide and allocate the data group to multiple processing units,
In the means for executing a predetermined successive approximation process by each arithmetic unit, the time required for each arithmetic unit to start comparison processing of a data group until it finishes the comparison process and outputs the processing result to the next arithmetic unit. A distributed processing method characterized in that the amount of data to be allocated to each arithmetic unit is determined so that the time required for the next arithmetic unit to process a data group comparison process is approximately equal.