JP7809946B2 - Grid Computing System - Google Patents

Description

The technology disclosed here belongs to the technical field of grid computing systems.

Patent Documents 1 and 2 disclose technology related to a distributed processing system that includes multiple distributed processing nodes, aggregates numerical data from each distributed processing node to generate aggregated data, and distributes the aggregated data to each distributed processing node.

Patent Document 3 discloses technology related to a distributed processing system that links a counting processing node with multiple distributed processing nodes to train a neural network.

In addition, as shown in Patent Document 4, in recent years, in order to make effective use of the computing resources installed in vehicles, there has been consideration of connecting multiple vehicles to a wireless network using Wi-Fi communication or the like and using this as grid computing.

Japanese Patent Application Laid-Open No. 2019-215603 Japanese Patent Application Laid-Open No. 2020-67687 Japanese Patent Application Laid-Open No. 2019-219714 Japanese Patent Application Laid-Open No. 2020-160661

In systems that perform distributed processing of a single neural network, deep neural networks, for example, have a large number of parameters, which necessitates a high computational load and results in a significant increase in the power consumption required for the computation. On the other hand, in grid computing that utilizes vehicles and other mobile objects (hereinafter simply referred to as "mobile objects"), it is desirable to minimize the power consumption of each mobile object, as the batteries installed in the mobile objects have a finite storage capacity. In particular, when a user performs grid computing-related computations (hereinafter referred to as "grid computations") while traveling, it is anticipated that the remaining battery charge may not be very high. Even for such mobile objects, it is necessary to perform grid computations while minimizing the limitations and hindrances on the mobile object's next driving distance due to a lack of remaining battery charge.

The technology disclosed here was developed in light of these issues, and aims to provide a grid computing system that minimizes the impact of grid calculations on the next driving distance.

To solve the above problem, a first aspect of the present disclosure is directed to a grid computing system configured with mobile objects operating as processing nodes for the job, wherein the mobile object comprises a measurement unit that measures a tolerance parameter serving as an index of the mobile object's tolerance for computational load, and a first communication unit that transmits the tolerance parameter measured by the measurement unit to the management device; the management device comprises a second communication unit that receives the parameter, a memory unit that stores multiple computation parameters that differ in computation accuracy and computational load in the job's computation, and a control unit that selects setting parameters to be used by the mobile object from the multiple computation parameters based on the tolerance parameter and transmits job data for the job and the setting parameters to the mobile object via the second communication unit; and the mobile object further comprises a memory unit that stores the job data and the setting parameters received from the management device, a computation unit that includes multiple computation elements configured in a dataflow format and a bypass route that bypasses a computation route selected among the multiple computation elements, and a computation control unit that enables some or all of the bypass route based on the setting parameters and causes the computation object to compute the job data.

In the above aspect, the grid computing system executes a series of processes: (1) a tolerance parameter, which is an index of the tolerance of the computational load, is measured in the mobile body; (2) a management device selects a setting parameter to be used for the mobile body from among multiple computation parameters based on the tolerance parameter; and (3) the mobile body enables a bypass route for the computation element selected based on the setting parameter and performs computation in the computation unit. As a result, the mobile body executes grid computation with a computational load based on the tolerance parameter, minimizing restrictions and obstacles on the vehicle's driving distance after the grid computation.

Specifically, for example, when the remaining battery charge, as a tolerance parameter, is low, setting parameters are selected that reduce the calculation load compared to when the remaining battery charge is sufficient. Then, by applying these setting parameters, the number of bypass routes that are activated increases; in other words, the grid calculation is performed with an increased number of bypassed calculation elements. This makes it possible to minimize restrictions and obstacles on the vehicle's driving distance after the grid calculation.

As explained above, the technology disclosed herein allows mobile bodies to perform calculations with a computational load that corresponds to their tolerance, thereby minimizing restrictions and obstacles to the vehicle's driving distance after grid calculations.

Conceptual diagram of a grid computing system A block diagram illustrating the configuration of a vehicle Block diagram showing an example of the configuration of a management server Block diagram illustrating the configuration of a grid computing system A diagram showing an example of a neural network structure. 1 is a flowchart illustrating an example of the operation of a grid computing system. FIG. 10 is a diagram showing an example of the relationship between the junction temperature rise and the number of driven arithmetic elements. A block diagram showing another example of the configuration of a grid computing system.

Embodiments will be described in detail with reference to the drawings. Identical or equivalent parts in the drawings will be designated by the same reference numerals, and repeated explanations may be omitted. Furthermore, the following embodiments will focus on configurations that are highly relevant to the contents of this disclosure.

The following embodiments are illustrative, and the contents of this disclosure are not intended to be limited by the presence or absence of descriptions or the numerical values exemplified. Furthermore, regardless of whether the terms "system," "unit," or "apparatus" are used in this disclosure, a system/unit/apparatus may be implemented using dedicated circuits such as an application-specific integrated circuit (ASIC) or a programmable logic array (PLA). Similarly, a system/unit/apparatus may be implemented using a processor circuit that executes computer-readable instructions (e.g., a program) to perform specific processing steps and thereby perform specific functions.

(Grid Computing System)
FIG. 1 illustrates an example of the configuration of a grid computing system 1 (hereinafter also simply referred to as "system 1") according to an embodiment.

This system 1 comprises multiple vehicles 10 and a management server 50. These components can communicate with each other via a global network. Each of the multiple vehicles 10 is equipped with a computing unit 17. The management server 50 is an example of a management device. The vehicles 10 are an example of a mobile object. Note that the management device may be implemented in the cloud. Furthermore, the mobile object is not limited to the vehicle 10.

[Grid Computing]
1 , in the system 1 of the embodiment, a grid computing G (hereinafter also simply referred to as a "grid G") is configured by a plurality of computing units 17. In the system 1, grid computing is executed in which an available computing unit 17 among the plurality of computing units 17 computes an application job (hereinafter also simply referred to as a "job").

When the vehicle 10 is moving, the computing power of the computing unit 17 is required to control the driving of the vehicle 10, and the computing unit 17 is in an operating state. On the other hand, for example, when the vehicle 10 is stopped and the power to the vehicle 10 is turned off, the computing power of the computing unit 17 for controlling the driving of the vehicle is essentially unnecessary. Therefore, grid computing calculations (hereinafter referred to as "grid calculations") are performed when the vehicle 10 is not in operation (for example, while parked). In this example, when the vehicles are not in operation, neural network calculations are performed using the computing resources (including the computing unit 17) installed in each vehicle as calculation nodes.

〔vehicle〕
2 , the vehicle 10 includes a battery 11, a measurement unit 12, a communication unit 15, a storage unit 16, a calculation unit 17, and an arithmetic and control unit 19. The functions of the measurement unit 12, the communication unit 15, the storage unit 16, the calculation unit 17, and the arithmetic and control unit 19 can be realized by, for example, an MPU (Micro-Processing Unit) mounted on the vehicle 10.

-battery-
The vehicle 10 is equipped with a battery 11. The power of the battery 11 is supplied to on-board devices such as a computing unit 17. The battery 11 may also be used as a power source for a drive motor of a mobile object. Examples of such a vehicle 10 include an electric vehicle and a plug-in hybrid vehicle.

- Measurement section -
The measurement unit 12 measures tolerance parameters when the calculation unit 17 performs grid calculations. The specific form of the measurement unit 12 is not particularly limited, but includes, for example, a battery remaining capacity detection unit 121 that detects the remaining capacity of the battery 11 and a temperature sensor 122 provided near the calculation unit 17. For example, the temperature sensor 122 can be a conventionally known temperature detector for measuring junction temperature that is generally built into a microcomputer.

The tolerance parameter is a parameter that indicates the tolerance of the computational load. The tolerance parameter is not particularly limited, but examples include remaining battery charge information, temperature information measured using the temperature sensor 122 built into the microcomputer, etc.

The measurement unit 12 may also determine the allowable number of drives of the processing element PE, which will be described later, and use this allowable number of drives as the tolerance parameter. The method for determining the allowable number of drives will be described later.

-Communications Department-
The communication unit 15 transmits and receives information and data to and from the communication unit 51 of the management server 50. Specifically, the communication unit 15 transmits the tolerance parameters measured by the measurement unit 12 to the management server 50 via the global network. The communication unit 15 is an example of a first communication unit.

The communication unit 15 receives job data D1 and setting parameters D17 for the application job to be calculated (hereinafter simply referred to as "job") from the management server 50. The information and data received by the communication unit 15 are sent to the calculation unit 17. The job data D1 and setting parameters D17 will be explained later.

-Storage part-
The storage unit 16 stores information and data. The specific configuration of the storage unit 16 is not particularly limited. For example, the storage unit 16 may be realized by a memory built into a chip, a hard disk drive (HDD), a solid state drive (SSD), or an optical disc such as a DVD or a BD.

In this example, the storage unit 16 stores vehicle information D10. The vehicle information D10 includes basic vehicle information D11, vehicle condition information D13, and operation information D15. The storage unit 16 also stores job data D1 and setting parameters D17 received from the management server 50.

<Vehicle basic information>
The vehicle basic information D11 includes vehicle identification information and resource information.

Vehicle identification information includes information for identifying the vehicle, such as the VIN, and user identification information that identifies the owner of the vehicle 10.

The resource information is information related to the computational resources 171, which will be described later. The resource information includes, for example, a computational resource ID assigned to each computational resource 171, and performance information indicating the performance of each computational resource 171. The performance of the computational resource 171 includes the computational capacity of the computational resource 171 (specifically, maximum computational capacity), the ratio of CPU to GPU in the computational resource 171, and the like. The computational capacity of the computational resource 171 is, for example, the amount of data that each computational resource 171 can compute per unit time.

<Vehicle status information>
The vehicle state information D13 is information indicating the state of the vehicle 10, and includes, for example, vehicle position information, driving history information, vehicle communication information, vehicle power supply information, measured temperature data, vehicle driving state information, etc. The vehicle state information D13 is used, for example, to detect the parking state of the vehicle.

Vehicle position information indicates the position (latitude and longitude) of the vehicle 10. For example, vehicle position information can be obtained using the Global Positioning System (GPS).

The driving history information may be, for example, information that associates vehicle driving information detected by a driving detector (not shown) with time, or information that associates the above-mentioned vehicle position information with time. In addition to the driving history information, driving schedule information that indicates future driving schedules for the vehicle 10 may also be included.

The vehicle communication information includes information indicating the communication status between the vehicle 10 and the global network, and information about the communication bandwidth between the vehicle 10 and the management server 50. The vehicle communication information is updated, for example, at predetermined time intervals. The vehicle communication information may also include information indicating the status of vehicle-to-vehicle communication with other vehicles 10.

Vehicle power information includes information indicating the power status of the vehicle 10, remaining battery capacity information, vehicle charging information, etc. For example, vehicle power information indicates whether the ignition power is on/off, whether the accessory power is on/off, etc. Battery capacity information is information indicating the remaining capacity of the battery 11 detected by the battery capacity detection unit 121. Battery capacity information is updated, for example, at predetermined time intervals. The information may also be updated based on a specific trigger, such as when the vehicle is detected as parked or when the remaining capacity falls below a predetermined threshold. Vehicle charging information indicates whether the vehicle 10 is being charged at a charging facility (not shown).

The temperature measurement data is measured using the temperature sensor 122 built into the microcomputer. The temperature measurement data is updated, for example, at predetermined time intervals. The information may also be updated based on a specific trigger, such as when a vehicle is detected as parked or when the margin for the junction temperature falls below a predetermined threshold.

Vehicle driving state information is information that indicates the current driving state of the vehicle. For example, vehicle driving state information includes engine on/off information, driving speed information, shift lever setting information, and various brake status information.

<Operation information>
The operation information D15 includes, for example, operation history information indicating the operation history of the computational resources 171 described below, and operation schedule information indicating the operation schedule of the computational resources 171.

Operation history information indicates, for example, the utilization rate of the computational resources 171 and/or the amount of job processing performed, in association with time. The operation history information includes normal operation history and grid operation history. Normal operation history indicates the history of operation of the computational resources 171 for user use, such as providing services such as vehicle driving, car navigation, and music playback. Grid operation history indicates the history of operation of the computational resources 171 to execute grid computing processes.

Operation schedule information includes, for example, usage schedule information indicating the future usage status of the computing resource 171.

-Arithmetic section-
The calculation unit 17 controls each part of the vehicle 10. In this example, the calculation unit 17 controls each actuator (not shown) in accordance with various information obtained by sensors (not shown).

The calculation unit 17 includes a processor, memory, etc. Examples of processors include a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The memory stores programs for operating the processor, information and data indicating the processing results of the processor, etc.

In this disclosure, for convenience of explanation, the resources available for grid computing calculations and processing within the calculation unit 17 will be referred to as "computational resources."

The computational resources 171 may also be resources used to control the vehicle 10, or may be CPUs or GPUs dedicated to grid computing that are not used to control the vehicle.

Furthermore, for example, there may be separate time periods during which use as computational resource 171 is permitted and time periods during which use as computational resource 171 is restricted. That is, a single CPU may be counted as computational resource 171 during certain time periods and not be counted as computational resource 171 during other time periods. Furthermore, if a CPU is implemented with a single or multiple cores, some of the multiple cores may be counted as computational resource 171, and the remaining cores may not be counted as computational resource 171. The same applies to GPUs.

As shown in Figure 4, the computational resource 171 comprises multiple processing elements PE configured in a dataflow format, a bypass path 31 that bypasses a computation path selected among the multiple processing elements PE, and a RAM (Random Access Memory) 172 for storing intermediate computation progress. In other words, in the computational resource 171, multiple processing elements PE form a dataflow architecture. In this disclosure, a "processing element PE" refers to a combinational circuit that performs multiply-and-accumulate operations. The block diagram in Figure 4 conceptually illustrates an example in which the neural network model shown in Figure 5, described below, has been converted into x dataflow architectures, 21 to 2x (x is a positive integer).

In addition, some or all of the computational resources 171 may be configured as reconfigurable circuits. That is, the circuit configuration of the computational resources 171 when controlling the vehicle may be different from the circuit configuration of the computational resources 171 when performing grid calculations. A reconfigurable circuit is a circuit configured to allow programmable hardware reconfiguration. Specifically, a reconfigurable circuit is a programmable hardware device that incorporates a wide variety of fine-grained computing elements (in this example, computing elements PE) and one or more memories (in this example, RAM 172), and can switch the internal connections between them. Examples of reconfigurable circuits include an FPGA (Field Programmable Gate Array) and a DRP (Dynamically Reconfigurable Processor).

- Arithmetic and control unit -
The calculation control unit 19 enables a part or all of the bypass route 31 based on the setting parameters stored in the storage unit 16, and causes the calculation unit 17 to calculate the job data D1. The specific operation of the calculation control unit 19 will be described later.

[Management Server]
The management server 50 manages the operation of grid computing. In other words, the system 1 includes the management server 50. The management server 50 is owned by the operator that operates the system 1.

As shown in FIG. 3, the management server 50 includes a communication unit 51, a memory unit 53, and a control unit 55.

-Communications Department-
The communication unit 51 is configured to be capable of two-way communication with the vehicle 10, and transmits and receives information and data to and from the connected vehicle 10 via a global network. The communication unit 51 is also connected to a client terminal (not shown) to be capable of two-way communication, and transmits and receives information about jobs, job data D1, calculation result data D2, etc., between them. The information and data received by the communication unit 51 are sent to the control unit 55. The communication unit 51 is an example of a second communication unit.

-Storage part-
The storage unit 53 stores information and data. The specific configuration of the storage unit 53 is not particularly limited. For example, the storage unit 53 may be realized by a memory built into a chip, a hard disk drive (HDD), a solid state drive (SSD), or an optical disc such as a DVD or a BD.

In this example, the memory unit 53 stores various tables and data, such as a vehicle information table D51, a job table D53, multiple calculation parameters D55, job data D1, and calculation result data D2.

<Vehicle Information Table>
The vehicle information table D51 is a table for managing vehicle information. The vehicle information table D51 stores a list of vehicle information D10 for each vehicle 10.

<Job Table>
The job table D53 is a table for managing jobs requested by clients. For each job, the job table D53 registers job information such as the reception number set for that job, the client ID set for the client that requested the job, and the name and content of the job. The job table D53 also registers for each job the operation type and processing conditions of the job data corresponding to that job, the required computing power that is the computing power required to calculate the job data, the delivery date set for that job, and the like.

<Job Data>
The job data D1 stored in the storage unit 53 is data of a job accepted from a client terminal, and is data to be processed to execute the job.

Job data D1 can be classified by operation type. Examples of operation types include CPU-based operation types and GPU-based operation types. Job data D1 of the CPU-based operation type tends to require complex operations with many conditional branches, such as simulation operations. Job data D1 of the GPU-based operation type tends to require a huge amount of simple operations, such as image processing and machine learning. In this example, job data D1 is assumed to be data used for distributed processing of a neural network.

<Calculation result data>
The computation result data D2 stored in the storage unit 53 is data of the computation result of a job computed by each neural network through grid computing processing, which will be described later.

<Calculation parameters>
A plurality of calculation parameters D55 used in the grid calculation process are stored in the storage unit 53. In this example, three calculation parameters, first to third calculation parameters, are prepared as shown in Fig. 4. The calculation parameters D55 refer to network structure information in the neural network.

Figure 5 shows an example of the configuration of a network model indicated by calculation parameter D55. In Figure 5, solid lines represent networks that execute processing, and dashed lines represent networks that bypass processing.

The left diagram in Figure 5 shows an example of an original network model, which for ease of explanation will be referred to as the "first calculation parameter." The first calculation parameter is set to have the greatest number of calculations, i.e., the highest calculation load, among the three calculation parameters D55, and is characterized by the highest calculation accuracy. In other words, the first calculation parameter is the parameter that results in the densest network.

The right diagram in Figure 5 shows a network model in which calculations have been thinned out and made sparse for the first calculation parameter. In this example, for ease of explanation, this is referred to as the "third calculation parameter." A sparse network model exhibits changes in characteristics compared to the original network, such as a reduced number of calculations and a slight deterioration in calculation accuracy. In other words, the third calculation parameter is set to have the fewest number of calculations, i.e., the lightest calculation load, among the three calculation parameters D55, but is characterized by a correspondingly lower calculation accuracy. In other words, the third calculation parameter is the parameter that makes the network sparsest.

Although not shown, the degree of sparsification of the second calculation parameter (the proportion of the dashed line in Figure 5) is set to be approximately halfway between the first calculation parameter and the third calculation parameter.

The degree of sparsity of the calculation parameters D55 may be selected from multiple calculation parameters D55 prepared with a preset degree of sparsity. Alternatively, calculations may be performed internally according to the degree of sparsity set by the user, and multiple calculation parameters D55 may be prepared. Note that the calculation parameters D55 may be updated as machine learning continues.

This disclosure is characterized in that the original network model is realized using multiple processing elements PE configured as a dataflow type, and each processing element PE is provided with a bypass path 31, allowing the degree of sparsification to be adjusted by enabling or disabling this bypass path 31. In other words, the calculation parameter is a parameter that indicates which of the processing elements PE configured as a dataflow type should have its bypass path 31 enabled when performing the calculation of job data D1.

Here, enabling the bypass path 31 refers to turning on the bypass path provided in the processing element PE that executes the process to be bypassed, so that the input signal flows to the output without passing through the processing element PE. This allows some network processing within the data flow process to be bypassed, thereby reducing the power consumption of the bypassed processing element PE. Note that stopping the supply of clocks to processing elements PE for which the bypass path 31 is enabled can further enhance the power consumption reduction effect.

In other words, this disclosure is characterized in that by selecting a calculation parameter D55 according to the tolerance for the calculation load as a setting parameter and applying it to the calculation resource 171, the calculation load of the grid calculation can be adjusted to a load that corresponds to that tolerance. For example, when the remaining battery charge is low, setting parameters are selected that reduce the calculation load compared to when the remaining battery charge is sufficient. This makes it possible to minimize restrictions and obstacles on the vehicle's driving distance after grid calculation. Specific operational examples will be explained later.

Note that bypass paths 31 do not necessarily need to be set for all processing elements PE, and processing elements PE for which bypass paths 31 are not set may be included. Also, a setting parameter may be provided to bypass an entire data flow column (for example, the processing element PE enclosed by the box 22 in Figure 4). Also, a bypass path 31 may be provided that bypasses multiple processing elements PE together.

-Control section-
In this example, the control unit 55 has the function of executing a series of controls and processes related to the operation and management of grid computing. For example, it executes the controls and processes in the flow chart of Fig. 6 described below. Note that in the following explanation, for the sake of convenience, the operations and processes are described with the management server 50 as the main body, but the control unit 55 may contribute to the processes and controls.

The control unit 55 stores information and data received from the client terminal in the storage unit 53. For example, when the control unit 55 receives job data D1 from the client terminal, it saves the job data D1 in the storage unit 53.

The control unit 55 stores information and data received from each vehicle 10 in the memory unit 53. For example, when the control unit 55 receives vehicle information D10 (including vehicle driving information and resource information) from a vehicle 10, it registers it in the vehicle information table D51 in the memory unit 53.

The control unit 55 also selects setting parameters to be used for the vehicle 10 from among multiple calculation parameters based on the tolerance parameters received from that vehicle 10. Then, it transmits the job data for the job to be requested of that vehicle and the selected setting parameters to that vehicle 10 via the communication unit 51. An example of a specific operation of the control unit 55 is described below in "Example of Operation of a Grid Computing System."

[Grid computing system operation example]
An example of the operation of the system 1 will be described below with reference to the flowchart of Fig. 5. In the example of Fig. 5, the operation of the management server 50 and the vehicle 10 and the exchange of information therebetween will be mainly described.

In this example, the job requested by the client (hereinafter also referred to as the "requested job") is assumed to be distributed processing of a neural network. Therefore, the memory unit 53 of the management server 50 stores data used for distributed processing of the neural network as job data D1. Here, for ease of explanation, the vehicle to which the job is requested will be referred to as the "target vehicle 10."

--Step S21--
In step S1, the management server 50 registers a plurality of calculation parameters D55 in the storage unit 53 in response to an input operation by the administrator.

There are no particular limitations on the method for registering the calculation parameters D55. For example, (1) the administrator inputs the degree of sparsity requested by the client (e.g., a sparsity of 5%, 10%, or 20%), or (2) the control unit 55 selects a calculation parameter D55 corresponding to each degree of sparsity from multiple pre-stored calculation parameters D55 and registers it in the storage unit 53. Alternatively, multiple calculation parameters D55 corresponding to each job may be obtained in advance from the client terminal, and the administrator may register them in the storage unit 53 as the calculation parameters D55 for that job.

--Steps S11 to S13, S22--
When it is detected that the conditions for starting grid calculation are satisfied in the target vehicle 10, the measurement unit 12 transmits the most recently measured tolerance parameters to the management server 50 via the communication unit 15 (steps S11 to S13).

Specifically, in step S11, the grid calculation start condition is determined to be satisfied when, for example, parking is detected using a sensor (not shown) provided on the target vehicle 10 and it is detected that the occupant has locked the vehicle from the outside. Furthermore, for example, when the target vehicle 10 is parked and locked, the grid calculation start condition is determined to be satisfied when a notification of permission to start grid calculation is received from the owner of the target vehicle 10.

In step S12, the measurement unit 12 of the target vehicle 10 measures the tolerance parameter. In this example, the battery remaining capacity detection unit 121 measures the remaining capacity of the battery 11 as the tolerance parameter. Note that if the remaining battery capacity is measured periodically and stored in the memory unit 16, the most recent measurement data may be used instead of a new measurement. Note that measurement data from the temperature sensor 122 may be used as the tolerance parameter in addition to or instead of the remaining battery capacity.

In step S13, the tolerance parameters are transmitted from the communication unit 15 of the target vehicle 10 to the communication unit 51 of the management server 50. The tolerance parameters received by the management server 50 are stored in the memory unit 53 under the control of the control unit 55 (step S22).

--Step S23--
In step S23, the control unit 55 of the management server 50 selects setting parameters to be used for the target vehicle 10 from the multiple calculation parameters stored in the memory unit 53 based on the tolerance parameters received from the target vehicle 10.

For example, when remaining battery charge information for the battery 11 is received as the tolerance parameter, the higher the remaining battery charge, the higher the upper limit of the calculation load for the calculation parameter D55 that can be selected as the setting parameter is set. In other words, the control unit 55 includes calculation parameters D55 with higher calculation accuracy in the options and selects the optimal calculation parameter D55 from within that range of options. On the other hand, when the remaining battery charge of the battery 11 is low, the upper limit of the calculation load is set low, and the control unit 55 selects calculation parameters D55 with a low calculation load, even if it means sacrificing some calculation accuracy.

For example, when measurement data from temperature sensor 122 is received as the tolerance parameter, the difference between the temperature of the semiconductor elements constituting the computational resource 171 and the maximum rated junction temperature is estimated based on that measurement data, and the upper limit of the computation load of computation parameters D55 selectable as setting parameters is set based on the number of computation elements driven corresponding to that difference. In other words, when the control unit 55 estimates that the temperature difference between the maximum rated junction temperature and the temperature of the semiconductor elements is large, it includes computation parameters D55 with higher computation accuracy in the options and selects the optimal computation parameter D55 within that range of options. On the other hand, when the temperature difference is estimated to be small, the upper limit of the computation load is set low, and the control unit 55 selects computation parameters D55 with a relatively small computation load within the upper limit of the number of computation elements driven, even if it means sacrificing some computation accuracy.

--Steps S24, S14--
In step S24, the control unit 55 transmits the job data D1 of the job to be requested of the target vehicle 10 and the setting parameters set in step S23 to the target vehicle 10 via the communication unit 51. When the target vehicle 10 receives the job data D1 and the setting parameters from the management server 50, they are stored in the memory unit 16 (step S14).

--Step S15--
In step S15, the calculation control unit 19 enables the bypass path 31 based on the setting parameters. As described above, enabling the bypass path 31 bypasses part of the network processing. Specifically, for example, in the configuration of FIG. 4, when the calculation element PE 22b is arranged after the calculation element PE 21a and the setting parameters indicate that the calculation element PE 21a should be bypassed, the calculation control unit 19 enables the bypass path 31a. In this case, the input is input to the calculation element PE 22b and processed therein without passing through the calculation element PE 21a. At this time, if the clock of the calculation element PE 21a is stopped, power consumption can be further reduced, thereby suppressing battery drain.

--Steps S16, S17, S25--
In step S15, the calculation control unit 19 causes the calculation unit 17 to execute grid calculation using the job data D1 while enabling the bypass route 31 based on the setting parameters. Then, when the grid calculation using the job data D1 is completed, the calculation unit 17 transmits calculation result data D2 to the management server 50. Then, the calculation result data D2 is received by the management server 50 and stored in the storage unit 53, and the process ends.

[Effects of the embodiment]
As described above, the system 1 of the above embodiment executes a series of processes: (1) in the target vehicle 10, a tolerance parameter (for example, the remaining battery charge or the temperature around the computational resources 171) that is an index of the tolerance of the computational load is measured; (2) in the management server 50, a setting parameter to be used for the target vehicle 10 from among a plurality of computational parameters based on the tolerance parameter; and (3) in the target vehicle 10, the bypass path 31 of the computation element PE selected based on the setting parameter is enabled and, in this state, the computational resources 171 are caused to execute grid computation. As a result, in the target vehicle 10, grid computation is executed using the computational resources 171 that are set to have a computational load according to the tolerance parameter, and therefore, it is possible to minimize restrictions and obstacles on the vehicle's traveling distance after the grid computation.

The above embodiments are essentially preferred examples and are not intended to limit the scope of the technology disclosed herein, its applications, or its uses. In other words, the above embodiments are merely examples and should not be interpreted as limiting the scope of the present disclosure. The scope of the present disclosure is defined by the claims, and all modifications and variations that fall within the equivalent range of the claims are within the scope of the present disclosure.

For example, as shown in FIG. 8, the measurement unit 12 may have an estimation unit 123 that estimates the allowable driving number of the processing element PE, and the allowable driving number estimated by the estimation unit 123 may be used as the tolerance parameter.

Below, we will show two examples of how to determine the allowable number of drives.

<Method for determining the allowable driving number (1)>
First, as characteristics of the MPU itself, a table, a function, or the like showing the relationship between the rise in junction temperature of the semiconductor elements constituting the processing element PE (hereinafter simply referred to as "junction temperature") and the number of processing elements PE being driven is stored in advance in the storage unit 16. Fig. 7 is a graph showing an example of the relationship between the rise in junction temperature and the number of processing elements being driven.

The measurement unit 12 uses the temperature sensor 122 to measure the junction temperature of the semiconductor element when the processing element PE is not driven, for example, using the graph in Figure 7 after power-on when grid calculations are about to begin.

The estimation unit 123 then calculates the standard value for the number of drivable computing elements from the difference between the maximum rated junction temperature and the measured junction temperature of the semiconductor element.

For example, if the junction temperature of the processing element PE in a non-driven state after the power supply is turned on is 60°C, and the maximum rated junction temperature of the MPU (computational resource 171) is 105°C, the difference is 45°C. Therefore, from the graph in Figure 7, the allowable number of driven processing elements PE (allowable number of processing elements) is determined to be 50.

In this case, in step S13 of FIG. 6, information on the allowable number of drives of the processing element PE is transmitted from the target vehicle 10 to the management server 50 as a tolerance parameter. Then, in step S23, the control unit 55 of the management server 50 selects setting parameters corresponding to the allowable number of drives of the processing element PE based on that allowable number of drives. In other words, the control unit 55 sets an upper limit on the calculation load of the calculation parameters that can be selected as setting parameters based on the allowable number of drives (number of drives possible) of the processing element PE, and selects setting parameters corresponding to that upper limit.

<Method for determining the allowable driving number (2)>
First, the vehicle owner sets a standard remaining charge threshold value for the vehicle, which indicates "up to what percentage of the remaining charge of the battery 11 the grid calculation can be performed." Since the usage status of the battery depends on how the vehicle 10 is used, allowing the vehicle owner or the like to set the standard remaining charge threshold value in advance can increase convenience and reduce restrictions and obstacles on the vehicle's driving distance after grid calculation.

In the vehicle, the characteristics of the relationship between the number of calculations driven per unit time by the calculation resource 171 and the amount of battery consumption per unit time depending on the number of calculation elements driven are registered in advance as data in the memory unit 16.

Then, the estimation unit 123 calculates the allowable number of drives and executable time after the grid calculation is completed based on the remaining battery power measured by the battery remaining power detection unit 121 so that the remaining power of the battery 11 does not fall below the remaining power threshold standard value. Note that the executable time may be set in advance by the user, for example, to a time period during which grid calculations can be used.

In this case, in step S13 of FIG. 6, information on the allowable number of drives and executable time of the processing element PE is transmitted from the target vehicle 10 to the management server 50 as tolerance parameters. Then, in step S23, the control unit 55 of the management server 50 selects setting parameters corresponding to the allowable number of drives based on the allowable number of drives and executable time of the processing element PE.

As explained above, the grid computing system disclosed here is extremely useful because it can minimize restrictions and obstacles to vehicle mileage after grid calculations.

1 Grid computing system 10 Vehicle (mobile object)
12 Measurement unit 15 Communication unit (first communication unit)
16 Storage unit 17 Calculation unit 19 Calculation control unit 31 Bypass route 50 Management server (management device)
51 Communication unit (second communication unit)
53 Storage unit 55 Control unit D1 Job data D17 Setting parameters PE Processing element

Claims (6)

A grid computing system including a management device that manages jobs and a mobile device that operates as a processing node for the jobs,
The moving body is
a measurement unit that measures a predetermined attribute value that serves as an index of the tolerance of the computational load in the moving object;
a first communication unit that transmits the attribute value measured by the measurement unit to the management device;
The management device
a second communication unit that receives the attribute value ;
a storage unit storing a plurality of calculation parameters having different calculation accuracies and calculation loads in the calculation of the job;
a control unit that selects setting parameters to be used for the mobile body from the plurality of calculation parameters based on the attribute value , and transmits job data of the job and the setting parameters to the mobile body via the second communication unit;
The moving body is
a storage unit in which the job data and the setting parameters received from the management device are stored;
an arithmetic unit including a plurality of arithmetic elements that execute data flow processing and a bypass path that bypasses an arithmetic path selected from among the plurality of arithmetic elements;
a calculation control unit that enables a part or all of the bypass paths based on the setting parameters and causes the calculation unit to calculate the job data. In the moving body,
the measurement unit includes a battery remaining capacity detection unit that detects a remaining capacity of a battery mounted on the moving object as the attribute value ,
2. The grid computing system according to claim 1, wherein the management device selects the setting parameters to be used for the mobile object based on the remaining battery power measured by the remaining battery power detector. The grid computing system described in claim 2, wherein the management device sets a higher upper limit on the computational load of the computation parameters selectable as the setting parameters as the remaining battery charge increases. In the moving body,
the measurement unit includes a temperature sensor that measures the ambient temperature of the calculation unit as the attribute value ,
In the management device,
3. The grid computing system according to claim 1, wherein the control unit selects the setting parameters to be used in the mobile body based on a junction temperature of a semiconductor element constituting the computing element estimated based on temperature data measured by the temperature sensor. The grid computing system described in claim 4, wherein the management device sets a higher upper limit on the computational load of the computation parameters selectable as the setting parameters the greater the difference between the maximum rated junction temperature and the estimated junction temperature. In the moving body,
The measurement unit includes a temperature sensor for measuring the ambient temperature of the calculation unit;
an estimation unit that estimates, as the attribute value , the number of drivable processing elements that can be driven among the plurality of processing elements based on the measurement result of the temperature sensor,
In the management device,
3. The grid computing system according to claim 1, wherein the control unit sets an upper limit of the computation load of the computation parameters selectable as the setting parameters based on the number of computation elements that can be driven.