An edge computing system for data analysis

Abstract

The application discloses an edge computing system for data analysis, and belongs to the technical field of Internet of Things.The application comprises the following steps: a data acquisition module acquires data and transmits the data to a centralized management module, wherein the data comprises video streams or audio streams; the centralized management module intercepts the data into frames and transmits the frames to an edge computing module; the edge computing module comprises a plurality of edge computing devices capable of running different identification algorithms, and when an edge computing device with insufficient performance appears, a new edge computing device is intelligently scheduled to assist in running; a remote management module deploys identification algorithms to the edge computing module and transmits the identified data to a storage module; a power module is used for detecting the power consumption of the edge computing module and controlling the power supply of the edge computing module; and a network transmission module is used for network transmission of data.The application realizes a method for scheduling algorithms and network transmission control of a plurality of edge computing devices.

Description

Technical Field

[0001] This invention patent belongs to the field of IoT technology, specifically, it relates to an algorithm and device for edge computing device scheduling and network transmission control. Background Technology

[0002] Edge computing has begun to be vigorously promoted by the industry and has entered a phase of rapid growth. With the rise of edge computing and the digital context of the Internet of Things, the integration of edge technology with network technologies such as cloud computing has become an important force driving the implementation of edge computing technology. The application of edge computing technology has developed steadily and entered the stage of industry implementation. However, the image processing efficiency of a single edge computing device is limited, and it is also prone to downtime due to overuse of a single edge computing device, resulting in certain losses.

[0003] For example, Chinese patent application No. 202110205681.5, published on July 13, 2021, discloses a scheduling method and edge device cluster based on edge computing. This scheme discloses a scheduling method and edge device cluster based on edge computing, including receiving an orchestration task from a control center and creating a container matching the orchestration task; detecting device information of each edge device and, based on the device information, selecting a target edge device suitable for the container from among the edge devices; scheduling the container to the target edge device and binding the container to the target edge device to establish a mapping relationship between the container and the corresponding cache in the target edge device. However, when the orchestration task from the control center is complex and there is no suitable edge device among the edge devices, the orchestration task runs on the edge device of the most suitable container at present, thus the orchestration task cannot be completed in the most efficient way. This invention first selects the most suitable edge device for the current orchestration task, and then continuously monitors the operating status of the edge device during task execution, and schedules new edge devices to assist in task execution based on the operating status, thus solving the problem of not having an edge device that is most suitable for the container.

[0004] Chinese patent application No. 202111094500.2, published on January 7, 2022, discloses an allocation method for an edge computing system. This method includes: obtaining data collected by each edge computing application within a unit of time; calculating the total data change of each edge computing application within a unit of time based on the collected data; calculating the ratio of the actual data processed by each edge computing application within a unit of time to its maximum data processing volume within a unit of time based on the total data change of each edge computing application within a unit of time; and allocating the data currently being processed by edge computing applications whose calculated ratio does not meet the standard to the remaining edge computing applications. However, this method only schedules edge computing applications based on the ratio of the maximum total data change of each edge computing application within a unit of time to the current actual data processing volume. It does not consider that when the ratio of the current actual data processing volume to the maximum data processing volume of an edge computing application meets the standard, the hardware status, such as memory, exceeds the predetermined standard, causing a decrease in the edge computing application's ability to continuously process data for a long time, and making it more likely to cause the edge computing application to crash. This invention continuously monitors the operating status of edge computing devices during task processing. When the task processing capacity of an edge computing device decreases or its memory usage exceeds a predetermined standard, a new edge computing device is promptly scheduled to jointly process the unfinished tasks. This ensures both the continuous task processing capability of the edge computing devices and prevents them from crashing due to hardware operating conditions.

[0005] Therefore, there is an urgent need to find an edge computing system for data analysis that employs an edge computing device scheduling algorithm to enable the system to autonomously select suitable edge computing devices to assist in operation when a single edge computing device cannot meet the demand, thereby alleviating the operational pressure on a single edge computing device and improving data processing efficiency. Summary of the Invention

[0006] 1. The problem to be solved

[0007] To address the low image processing efficiency of existing edge computing devices, an edge computing system for data analysis is provided. This system integrates multiple edge computing devices into an edge computing module and includes an edge computing device scheduling method, thereby improving data processing efficiency and ensuring the stability of the edge computing devices.

[0008] 2. Technical Solution

[0009] To solve the above problems, the technical solution adopted by the present invention is as follows:

[0010] An edge computing system for data analysis, characterized in that it comprises a data acquisition module, a centralized management module, an edge computing module, a remote management module, a storage module, a power supply module, and a network transmission module; wherein:

[0011] The data acquisition module collects data and transmits it to the centralized management module via the network transmission module. The data includes video streams or audio streams; the data acquisition module is a camera or a recorder.

[0012] The centralized management module extracts data into frames (images or audio segments) and transmits them to the edge computing module. The centralized management module then selects the most suitable edge computing device to assist in the operation, reducing the workload of the edge computing device and improving the task processing efficiency.

[0013] The edge computing module contains several edge computing devices that can run different recognition algorithms. The operating status information of the edge computing devices is fed back to the centralized management module. When an edge computing device with insufficient performance appears, a new edge computing device is intelligently scheduled to assist in its operation.

[0014] The remote management module deploys the recognition algorithm to the edge computing module and transmits the recognized data to the storage module;

[0015] The power module is used to detect the power consumption of the edge computing module and control the power supply of the edge computing module;

[0016] The network transmission module is used for network transmission of data.

[0017] As a further improvement to the present invention, the edge computing devices are arranged as follows: the box is box-shaped with empty spaces on the left and right; the fan is placed at the bottom of the box, the physical buttons are installed at the rear of the box, and the regulated power supply is installed at the rear of the box; the edge computing devices are arranged sequentially inside the box, maintaining a spacing of 2cm to 3cm.

[0018] As a further improvement of the present invention, the remote management module deploys the identification algorithm to high-priority edge computing devices through the network transmission module, wherein the method for determining the priority of edge computing devices is as follows:

[0019] Step 1: When the identification algorithm is deployed, record the number of times it is deployed to edge computing devices, m1, m2, ..., m. k k is the number of edge computing devices;

[0020] Step 2: While the edge computing device is running, count the number of times the algorithm detects anomalies on the edge computing device, n1, n2, ..., n. k k is the number of edge computing devices;

[0021] Step 3: P(i,j) represents the probability function that the j-th recognition algorithm successfully runs on the i-th edge computing device, m i,j n represents the number of times the j-th image recognition algorithm runs on the i-th edge computing device. i,j This represents the number of times the j-th recognition algorithm fails to run on the i-th edge computing device, where i is the ID of the edge computing device, j is the ID of the recognition algorithm, i∈[1,m], j∈[1,s], m represents the maximum number of edge computing devices that the edge computing module can accommodate, and s represents the total number of recognition algorithms.

[0022]

[0023] Step 4: Prioritize the identification algorithms during deployment based on the value of P(i,j). Larger P(i,j) indicates higher priority, and smaller P(i,j) indicates lower priority. m,s S indicates the priority of the recognition algorithm for edge computing devices. m,s S is a positive integer. m,s ∈[1,k], where

[0024]

[0025] As a further improvement of this invention, the deployment of the recognition algorithm and the allocation of video or audio stream data are as follows: The data acquisition module sends the video or audio stream data to the centralized management module through the network transmission module. The centralized management module extracts the video or audio stream data into z-frame images or audio and distributes them to the edge computing devices with the highest priority in the edge computing module. When multiple edge computing devices are idle, the recognition algorithm and video or audio stream data are sent to the highest priority edge computing device for execution by default. z∈[1,max], where max represents the number of frames that can be extracted from the video.

[0026] As a further improvement of this invention, during the transmission of video or audio stream data through the network transmission module, the relationship between the network transmission rate decision and the network transmission rate in the network transmission module is continuously observed and controlled, and the network transmission rate is automatically adjusted to adapt to the network environment. The adjustment is divided into two cases: a=1 and a=2, to accelerate or decelerate the network transmission rate. The steps are as follows:

[0027] Step 1: Define each network monitoring period as t seconds. Within each monitoring period, first use V... a As the data packet transmission rate, where t = ∈ [2, 4]

[0028]

[0029] g represents the rate adjustment parameter, g∈(0,1); v represents the current network transmission rate.

[0030] Step 2: Calculate the rate adjustment function Y when a=1 and a=2 respectively. a (v a ),

[0031]

[0032] L a T represents the packet loss rate detected by the current network. a This indicates the current actual network throughput;

[0033] Step 3: Compare Y when a=1 and a=2 a (v a The size of ) is used to adjust the current network transmission rate according to the larger value of 'a'.

[0034] As a further improvement of the present invention, the network transmission module adopts an intelligent TCP congestion control algorithm to adaptively control the video data transmission rate in the network environment.

[0035] As a further improvement of the present invention, the operation status detection is as follows: Each edge computing device in the edge computing module is equipped with a task detection module, which is used to track the task execution status, that is, edge computing device operation status detection, which monitors the memory usage, GPU usage, disk usage and the recognition speed of target data (image or audio segment) of the edge computing device, and transmits the detection data to the centralized management module.

[0036] As a further improvement of the present invention, the centralized management module collects the detection data of the task detection module in each edge computing device at regular intervals of t1 seconds, where t1∈[1,3].

[0037] As a further improvement to this invention, power consumption monitoring is implemented: the power module detects the power consumption of the edge computing module using an AC power sensor to monitor the real-time power consumption of the edge computing devices. When issues arise such as excessive power consumption of a single device or excessive total power consumption of the edge computing devices, the power control process addresses the problem to prevent overloading and excessive power consumption from causing safety issues. When excessive load or power consumption occurs in the edge computing devices, some edge computing devices are shut down to reduce the overall power consumption of the device, and edge computing devices are periodically shut down. Furthermore, the power module performs timed hard-start and hard-shutdown of edge computing devices, and physical buttons are provided to control the shutdown and forced restart of the edge computing devices.

[0038] Targeted solutions for edge computing device outages, false outages, and resource information transmission failures.

[0039] As a further improvement of the present invention, the handling method for edge computing device downtime is as follows:

[0040] Step 1: Set a local clock for each edge computing device. During the operation of each edge computing device, continuously compare the difference between the local clock of the edge computing device and the local clock of the central management module. If the time difference between the two exceeds t2 seconds for N consecutive times, the edge computing device is determined to be down, where N∈[3,+∞]; t2∈[2,+∞).

[0041] Step 2: After determining that the edge computing device has crashed, the centralized management module sends a crash fault command to the centralized management module. The centralized management module records the number of failures at the fault node and calls the power module to restart the edge computing device.

[0042] Step 3: If an edge computing device restarts more than M times within t3 minutes, the power module will shut it down. The centralized management module will then report this information to the remote management module for manual intervention. Here, t3 ∈ [3,5] and M ∈ [3,+∞].

[0043] As a further improvement of this invention, when edge computing devices experience outages, false outages, or resource information transmission failures, it is determined that an edge computing device with insufficient performance has occurred, and the result is determined according to the probability function P. x (i,j) autonomously selects a new edge computing device to assist in operation, P x (i,j) represents the probability function for the x-th deployment of the image recognition algorithm, where x∈[1,+∞], and:

[0044]

[0045] ξ (i,j) =1 / Surplus

[0046] m (i,j) ξ(i,j) represents the number of times the j-th identification algorithm runs on the i-th edge computing device; ξ(i,j) represents the expected value of the j-th identification algorithm autonomously selecting the i-th edge computing device; S x This represents the set of edge computing devices currently selected autonomously; Surplus represents the set of edge computing devices that can be selected and used in the current device.

[0047] 3. Beneficial effects

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] This invention discloses an edge device operating power supply, device scheduling algorithm, and apparatus. By integrating edge computing devices into an edge computing unit, it not only enables multiple image recognition algorithms to run on a single edge computing unit but also improves image processing efficiency. Furthermore, by using edge computing device scheduling, image recognition algorithms can autonomously select edge computing devices, thereby enhancing the operational stability of the edge computing unit. Attached Figure Description

[0050] Figure 1 This is a flowchart of the edge computing system of the present invention;

[0051] Figure 2 This is a screenshot of the remote control platform of the edge computing system of the present invention. Detailed Implementation

[0052] The present invention will be further described below with reference to specific embodiments.

[0053] Example

[0054] I. Fabrication of enclosure, assembly of equipment, and installation and configuration of systems:

[0055] When deploying the edge computing system in this embodiment, weather conditions need to be considered. Therefore, a waterproof, lightweight enclosure measuring 110cm x 70cm x 40cm is constructed. Four fans are installed at the bottom of the enclosure for heat dissipation. Four semi-circular buckles are installed at the four corners of the bottom of the enclosure for easy on-site installation. Two physical buttons are installed at the rear of the enclosure, used to restart and shut down all edge computing devices, respectively.

[0056] A Linux operating system is installed on the centralized management module, and an SSH port is opened for remote sessions. A 4G / 5G signal source is connected to the remote management module. The remote management module controls the edge computing devices to deploy image recognition algorithms and monitors and displays the operating status information of all edge computing devices through the SSH port.

[0057] Second, set different network speed adjustment parameters to ensure that the network speed can be adjusted automatically.

[0058] A network testing cycle is set to 5 seconds, the actual network throughput is set to 100Mbps, and the packet loss rate is set to 1%. During the first video transmission over the network, video information is transmitted to the centralized management module at a rate of 2Mbps. The network speed is adjusted by setting the network speed adjustment parameter g to 0.5 and 0.9 respectively.

[0059] 1) The network speed adjustment parameter g is set to 0.5:

[0060] V1=(1+0.5)*2=3Mbps V2=(1-0.5)*2=1Mbps

[0061] U1 = 100 * (1 / 2.099) = 0.4764

[0062] U2 = 100 * (1 / 2.099) = 0.4764

[0063] Y1(v1)=0.4764-3*1%=0.4464

[0064] Y2(v2) = 0.4764 - 1 * 1% = 0.3764

[0065] Because T1(v1) > Y2(v2), it is necessary to increase the network transmission rate to 3 Mbps.

[0066] 2) The network rate adjustment parameter g is 2:

[0067] V1 = (1 + 0.9) * 2 = 3.8 Mbps V2 = (1 - 0.9) * 2 = 0.2 Mbps

[0068] U1 = 100 * (1 / 2.009) = 0.4977

[0069] U2 = 100 * (1 / 2.009) = 0.4977

[0070] Y1(v1) = 0.4977 - 3.8 * 1% = 0.4597

[0071] Y2(v2) = 0.4977 - 0.2 * 1% = 0.4957

[0072] Since Y1(v1) < Y2(v2), it is necessary to reduce the network transmission rate to 0.2 Mbps.

[0073] III. The first image recognition algorithm is deployed 15 times in total. Among them, it runs 10 times on the 4th edge computing device, 2 times on the 5th edge computing device, 2 times on the 7th edge computing device, and 1 time on the 10th edge computing device; the current number of edge computing devices that can be selected for use is 15.

[0074] The probability function of the first image recognition algorithm being deployed on the 5th edge computing device when selecting an auxiliary running node for deployment for the 16th time

[0075] The probability function of the first image recognition algorithm being deployed on the 7th edge computing device when selecting an auxiliary running node for deployment for the 16th time

[0076] The probability function of the first image recognition algorithm being deployed on the 10th edge computing device when selecting an auxiliary running node for deployment for the 16th time

[0077] Calculated, when the image algorithm makes an independent selection for deployment for the 16th time, the possibilities of being deployed on the 5th and 7th edge computing devices are the same. Considering the priority of the image recognition algorithm for edge computing devices, the 5th edge computing device is selected as the auxiliary running node.

[0078] IV. Operation

[0079] like Figure 2 The remote control platform of the edge computing system shown is set against the backdrop of a construction site. This embodiment uses the edge computing system to monitor the wearing of safety helmets by construction workers in real time. Video streams are captured by cameras. Based on the image processing efficiency, memory usage, and the likelihood of individual edge computing devices crashing during past use of the safety helmet recognition algorithm, the top three priority edge computing devices for the algorithm are edge computing devices 2, 5, and 9. Therefore, the safety helmet recognition algorithm is preferentially deployed on edge computing device 2. The remote management terminal invokes the safety helmet recognition algorithm on edge computing device 2. The video stream is divided into several frames by the centralized management module. The centralized management module then transmits these frames to edge computing device 2 via the network.

[0080] Edge computing device #2 started operating. The centralized management module continuously collected data on memory usage, disk storage usage, GPU usage, and image processing speed from the task detection module of edge computing device #2, and transmitted this data to the remote management terminal for display via the network. After approximately 3 minutes of operation, network fluctuations were detected. Following processing by the network transmission module, the image processing speed of edge computing device #2 decreased. After 20 seconds, the network returned to normal, and the image processing speed returned to normal.

[0081] After running for two days, edge computing device No. 2 experienced a decrease in image processing speed and an increase in memory usage. The centralized management module automatically started edge computing devices No. 4 and No. 1 to assist in processing the safety helmet recognition task. The centralized management module then transmitted several frames of images processed by the module to edge computing devices No. 1, No. 2, and No. 4.

[0082] After six days of operation, the edge computing device showed no abnormalities. The edge computing system was then shut down via the remote management module, and all edge computing devices were turned off.

[0083] The present invention has been described in detail above with reference to specific exemplary embodiments. However, it should be understood that various modifications and variations can be made without departing from the scope of the invention as defined by the appended claims. The detailed description and drawings should be considered illustrative only and not restrictive, and any such modifications and variations shall fall within the scope of the invention described herein. Furthermore, the background art is intended to illustrate the current state of development and significance of the technology and is not intended to limit the present invention or the scope of application of the present application.

Claims

1. An edge computing system for data analysis, characterized in that: It includes a data acquisition module, a centralized management module, an edge computing module, a remote management module, a storage module, a power supply module, and a network transmission module; among which: The data acquisition module collects data and transmits it to the centralized management module. The data includes video streams or audio streams. The centralized management module extracts data into frames and transmits them to the edge computing module; The edge computing module contains several edge computing devices that can run different recognition algorithms. When an edge computing device with insufficient performance appears, a new edge computing device is intelligently scheduled to assist in its operation. The remote management module deploys the recognition algorithm to the edge computing module and transmits the recognized data to the storage module; The power module is used to detect the power consumption of the edge computing module and control the power supply of the edge computing module; The network transmission module is used for network transmission of data, wherein: The remote management module deploys the identification algorithm to high-priority edge computing devices via the network transmission module. The method for determining the priority of edge computing devices is as follows: Step 1: When the identification algorithm is deployed, record the number of times it is deployed on edge computing devices. , … k is the number of edge computing devices; Step 2: While the edge computing device is running, count the number of times the identification algorithm malfunctions during edge computing device operation. , … k is the number of edge computing devices; Step 3: This represents the probability function that the j-th recognition algorithm successfully runs on the i-th edge computing device. This represents the number of times the j-th image recognition algorithm runs on the i-th edge computing device. This represents the number of times the j-th recognition algorithm failed to run on the i-th edge computing device, where i is the ID of the edge computing device and j is the ID of the recognition algorithm. , m represents the maximum number of edge computing devices that the edge computing module can accommodate, and s represents the total number of recognition algorithms. Step 4: According to The priority of the size ranking and recognition algorithm during deployment. The larger the value, the higher the priority. The smaller the value, the lower the priority. This indicates the priority of the recognition algorithm for edge computing devices. ,in 。 2. The edge computing system for data analysis according to claim 1, characterized in that: The data acquisition module sends video or audio stream data to the centralized management module via the network transmission module. The centralized management module then extracts the video or audio stream data into z-frame images or audio. This indicates the number of frames that can be extracted from the video and distributed to the edge computing devices in the edge computing module. When multiple edge computing devices are idle, the recognition algorithm and video or audio stream data will be sent to the highest priority edge computing device for execution by default.

3. The edge computing system for data analysis according to claim 2, characterized in that: During the transmission of video or audio stream data through the network transmission module, the relationship between the network transmission rate decision and the network transmission rate in the control module is continuously observed, automatically adapting to the network environment. and The steps to speed up or slow down network transmission in two different situations are as follows: Step 1: Define each network monitoring period as t seconds. Within each monitoring period, first... As the data packet transmission rate, where... Indicates the rate adjustment parameter. Indicates the current network transmission rate; Step 2: Calculate the values ​​when... and Time-rate adjustment function , = This indicates the packet loss rate detected by the current network. This indicates the current actual network throughput; Step 3: Compare when and hour The size is determined by the larger value. This value adjusts the current network transmission rate.

4. The edge computing system for data analysis according to claim 1 or 3, characterized in that: The network transmission module employs an intelligent TCP congestion control algorithm to adaptively control the video data transmission rate in the network environment.

5. The edge computing system for data analysis according to claim 1, characterized in that: Each edge computing device in the edge computing module is equipped with a task detection module, which is used to monitor the memory usage, GPU usage, disk usage and target data recognition speed of the edge computing device, and transmit the detection data to the centralized management module.

6. The edge computing system for data analysis according to claim 5, characterized in that: The centralized management module is timed The system collects detection data from the task detection module in each edge computing device every second. .

7. The edge computing system for data analysis according to claim 1, characterized in that: The power module detects the power consumption of the edge computing module, uses an AC power sensor to monitor the real-time power consumption of the edge computing device, and handles edge computing device downtime.

8. The edge computing system for data analysis according to claim 7, characterized in that: When an edge computing device fails, the following steps should be taken to handle the situation: Step 1: Set a local clock for each edge computing device. During the operation of each edge computing device, continuously compare the time difference with the local clock of the central management module. If the time difference between the two is N consecutive times... If the time exceeds a certain threshold, the edge computing device is considered to have crashed. ; ; Step 2: After determining that the edge computing device has crashed, the centralized management module sends a crash fault command to the centralized management module. The centralized management module records the number of failures at the fault node and calls the power module to restart the edge computing device. Step 3: In If an edge computing device restarts more than M times within a few minutes, the power module will shut it down. This information will then be relayed from the centralized management module to the remote management module for manual intervention. ; .

9. The edge computing system for data analysis according to claim 7 or 8, characterized in that: When an edge computing device experiences a crash, a false crash, or a failure in transmitting resource information, it is determined that the edge computing device is underperforming, based on a probability function. The system can independently select new edge computing devices to assist in its operation. This represents the probability function for the x-th deployment of the image recognition algorithm. ,in: This indicates the number of times the j-th recognition algorithm runs on the i-th edge computing device; This represents the expected value of the j-th recognition algorithm autonomously selecting the i-th edge computing device; This represents the set of edge computing devices currently selected autonomously; Surplus represents the set of edge computing devices that can be selected and used in the current device.

Images