Unlock instant, AI-driven research and patent intelligence for your innovation.

Task scheduling method, device, device and computer storage medium

A task scheduling and task technology, applied in the field of deep learning technology, can solve problems such as unreasonable operator task scheduling and insufficient storage resources, and achieve the effect of alleviating insufficient storage resources and reasonable task scheduling

Active Publication Date: 2022-06-28
BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF11 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] However, there are usually some limited storage resources in hardware, such as video memory resources. Once the operator task scheduling is unreasonable, it will face the problem of insufficient storage resources.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Task scheduling method, device, device and computer storage medium
  • Task scheduling method, device, device and computer storage medium
  • Task scheduling method, device, device and computer storage medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0027] figure 1 The flowchart of the main method provided in the first embodiment of the present disclosure, in many application scenarios, it is necessary to perform multi-threaded parallel scheduling on the target task on the device, so as to improve the computing efficiency. The above-mentioned device may be a server device, a computer device with relatively strong computing power, or the like. The present disclosure can be applied to the above-mentioned devices. like figure 1 As shown in, the method may include the following steps:

[0028] In 101 , according to the hardware execution cost of each operator task in the target task obtained by pre-analyzing, prioritize the concurrently executable operator tasks.

[0029] The target task can be any computationally intensive task. A typical target task is a training task or an application task within a deep learning framework, that is, a training task or an application task based on a deep learning model.

[0030] In the ...

Embodiment 2

[0038] figure 2 A detailed method flow chart provided in the second embodiment of the present disclosure, such as figure 2 As shown in, the method may include the following steps:

[0039] In 201, hardware occupancy information of each operator task in the target task is predetermined.

[0040] The hardware occupancy information may include newly added hardware resource occupancy, reclaimed resources after execution, and hardware execution cost.

[0041] This step can be performed in but not limited to the following two ways:

[0042] The first way: in the compilation phase of the target task, the hardware occupation information of each operator task is determined according to the size of the specified input data and the dependencies between each operator task in the target task.

[0043] For example, after the user builds a deep learning model, the size of the input data and output data of each operator task can be deduced in the model compilation stage according to the ...

Embodiment 3

[0074] image 3 A schematic structural diagram of the task scheduling device provided in the third embodiment of the present disclosure, the device may be an application located on the server side, or may also be a plug-in or a functional unit such as a software development kit (Software Development Kit, SDK) in the application located on the server side, Alternatively, it may also be located at a computer terminal, which is not particularly limited in this embodiment of the present invention. like image 3 As shown in , the apparatus 300 may include: a sorting unit 301 and a scheduling unit 302 , and may further include a first analyzing unit 303 , a second analyzing unit 304 and a determining unit 305 . The main functions of each unit are as follows:

[0075] The sorting unit 301 is configured to prioritize the concurrently executable operator tasks according to the hardware execution cost of each operator task in the target task obtained by pre-analysis.

[0076] The sch...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The disclosure discloses a task scheduling method, device, device and computer storage medium, and relates to deep learning technology in the technical field of artificial intelligence. The specific implementation plan is: according to the hardware execution cost of each operator task in the target task obtained in advance, sort the operator tasks that can be executed concurrently; Subtasks are scheduled; wherein, the scheduling includes: determining whether to execute the current operator task to be scheduled according to the amount of new hardware resources required by the current operator task to be scheduled and the amount of hardware resources available in the current system. The disclosure realizes reasonable scheduling of tasks to alleviate the problem of insufficient storage resources.

Description

technical field [0001] The present disclosure relates to the field of computer application technology, and in particular, to deep learning technology in the field of artificial intelligence technology. Background technique [0002] The deep learning framework is one of the basic technologies for the development of artificial intelligence technology. In the deep learning framework, both the training and application of deep learning models require a large number of tasks to complete. As the amount of computation increases, the processing method of concurrent execution gradually emerges. That is, according to the dependencies between operators in the deep learning model, the operator tasks are scheduled concurrently as much as possible. [0003] However, there are usually some limited storage resources in hardware, such as video memory resources. Once the operator task scheduling is unreasonable, it will face the problem of insufficient storage resources. SUMMARY OF THE INV...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06F9/48G06F9/50
CPCG06F9/4881G06F9/5011G06F9/5044G06F9/5022G06F9/5038G06F2209/484G06F2209/5021
Inventor 陈秋良刘红雨蓝翔
Owner BEIJING BAIDU NETCOM SCI & TECH CO LTD