Computing method based on parallel computing power dynamic allocation under cloud edge fusion
By dynamically adjusting the resource allocation parameters of edge computing nodes in the cloud-edge fusion computing system, the problems of limited resources and task changes in the edge computing system are solved, the task processing speed and simulation computing efficiency are improved, and the dynamic changes of the system are adapted to the dynamic changes.
Patent Information
- Application Number
- CN202411077152.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-07
AI Technical Summary
The existing cloud-edge converged computing system fails to effectively consider the limited computing resources and dynamic tasks of edge computing systems in the resource allocation algorithm, resulting in a decline in computing performance and the parallel computing algorithm is not compatible with the characteristics of edge computing environments.
Using a calculation method based on dynamic allocation of parallel computing power under cloud edge fusion, the calculation nodes of the edge computing system are divided into two groups evenly, and the resource allocation parameters are dynamically adjusted, and the blocking probability is calculated based on the task arrival time and processing time, and the resources are optimized. Assign to ensure the system's parameter configuration during the pre-allocation and stable operation stage.
It improves the speed of the system processing tasks, reduces the communication delay of the task in the cloud computing system, reduces the computing time of the simulation platform, and adapts to changes in task type and arrival frequency.
Smart Images

Figure CN119011504A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of edge computing technology, and specifically relates to a computing method based on dynamic allocation of parallel computing power under cloud-edge fusion. Background Art
[0002] Edge computing systems are deployed near users of devices such as the Internet of Things to provide computing services. Compared with cloud computing systems, the communication latency of tasks in edge computing is very low; however, the storage and computing resources of edge computing systems are limited and cannot serve all arriving tasks. Therefore, the integration of edge computing and cloud computing can serve more tasks and improve the performance of cloud-edge integrated computing systems, which can effectively alleviate the simulation computing bottleneck of emergency power supply situation simulation involving massive scenarios.
[0003] Considering the huge differences between cloud computing and edge computing, simple system fusion cannot maximize the performance of the fusion system, so it is necessary to reasonably allocate the computing resources of the fusion system. Current resource allocation algorithms mainly focus on optimizing the resource allocation and task scheduling management between edge and cloud. However, these algorithms do not take into account the following issues: some computing nodes in the fusion system face problems such as power management, software and hardware failures, and computing resource sharing, which leads to a decrease in system computing performance.
[0004] Using parallel computing methods in cloud computing is a way to solve the above problems. Parallel computing algorithms allocate multiple computing nodes to perform calculations simultaneously for each arriving task. This algorithm can integrate scattered computing resources and combine methods such as encoding and replication to avoid the impact of some slower tasks on the computing system, thereby reducing the computing time of tasks and improving the utilization of computing resources. However, the introduction of parallel computing into edge computing systems faces the following problems: when designing resource allocation algorithms, the objective condition of limited computing resources in edge computing systems must be taken into account; parallel computing algorithms in cloud computing mainly optimize the computing time and cost of each task, and usually do not consider whether the introduction of redundancy will cause more tasks to be blocked, while cloud-edge fusion computing systems need to optimize the entire system; edge computing environments often face large changes in computing task types. When designing resource allocation algorithms, it must be considered that the algorithm needs to be able to dynamically adjust the allocation of computing resources. Summary of the invention
[0005] In order to solve the above technical problems, the present invention provides a computing method based on dynamic allocation of parallel computing power under cloud-edge fusion, which is applied to the edge computing system. Through the algorithm, several computing nodes are allocated to each task to perform parallel computing, and the resource allocation parameters can be adaptively adjusted.
[0006] The computing method based on dynamic allocation of parallel computing power under cloud-edge fusion described in the present invention comprises the following steps:
[0007] Step 1: Divide the computing nodes of the edge computing system into two groups evenly, and use different allocation parameters for each group of nodes;
[0008] Step 2: The edge computing system assigns computing nodes to the arrived tasks, and calculates the average task arrival time, the average task processing time of each group of nodes, and the average cloud computing processing time during the observation period;
[0009] Step 3: Using the time obtained in step 2, calculate the task blocking probability of the edge computing system under each group of allocation parameters, and then calculate the task completion time of each group under the corresponding allocation parameters;
[0010] Step 4: By comparing whether the task completion time of each group decreases with the allocation parameters, it is determined whether to end the pre-allocation phase, and the parameters are dynamically optimized and configured accordingly;
[0011] Step 5: Use the optimal allocation parameters in the pre-allocation phase to calculate and update the average task arrival time, the average task processing time of the node, and the average cloud computing processing time during the observation period;
[0012] Step 6: Based on the time obtained in step 5 and the historical data of the average task processing time under different allocation parameters in step 3, the task blocking probability of the edge computing system under the current allocation parameters and adjacent allocation parameters is calculated, and the task completion time under the corresponding allocation parameters is calculated;
[0013] Step 7: By comparing the task completion time under the current allocation parameters and the adjacent allocation parameters, determine whether the stable operation phase is ended. If so, re-enter the pre-allocation phase to perform dynamic parameter optimization configuration; if not, complete the computing resource allocation.
[0014] Furthermore, in step 1, the computing nodes of the edge computing system are evenly divided into two groups. Each node in the first group processes one task alone, that is, the allocation parameter is m1=1; every two nodes in the second group process one task, that is, the allocation parameter is m2=2.
[0015] Furthermore, in step 2, the edge computing system allocates computing nodes to the arriving tasks, specifically:
[0016] Randomly select computing nodes based on the number of assigned parameters from any group of idle nodes. If there are no idle nodes, send the task to the cloud computing system for processing; if the number of idle nodes is less than the parameters m1 and m2, send the task to the cloud computing system;
[0017] The observation period refers to the edge computing system calculating the average task arrival time 1 / μ, the average task processing time t of each group of nodes after processing s tasks.j , average cloud computing processing time t c , the parameters are calculated using the arithmetic mean.
[0018] Furthermore, in step 3, the computing system uses the task blocking probability P(m) of the edge computing system under each group of allocation parameters, and the formula is:
[0019]
[0020] Among them, a = N / m, is the number of tasks that the edge computing system can calculate simultaneously; N is the total number of edge computing nodes; m is the general term for the parameters of the node allocation method, when calculating the task blocking probability of the first group, m = m1, when calculating the task blocking probability of the second group, m = m2; μ is the average task arrival rate, that is, the inverse of the average task arrival time; i is the intermediate variable of the summation, and its value is an integer [0, a];
[0021] Calculate the task completion time T(m) of each group under the corresponding allocation parameters, and the formula is:
[0022] T(m)=(1-P(m))t j +t c (2).
[0023] Furthermore, step 4 is specifically as follows:
[0024] If T(m1)≤T(m2), the pre-allocation phase ends and the current optimal allocation parameter is m. o =m1; where T(m1) is the task completion time of the first group of computing nodes, and T(m2) is the task completion time of the second group of computing nodes;
[0025] If T(m1)>T(m2), the pre-allocation phase continues, and the allocation parameter of the first group is adjusted to m1=m1+1, and the allocation parameter of the second group is adjusted to m2=m2+1.
[0026] Furthermore, in step 5, the optimal allocation parameter m obtained in step 4 is used. o Assign idle nodes to the arriving tasks. If there are no idle nodes, the tasks are sent to the cloud computing system for processing. If the number of idle nodes is less than the optimal allocation parameter m, o , the remaining idle nodes are assigned to the same task, and the task does not participate in the subsequent statistics of related time;
[0027] In step 5, the same method as in step 2 is used to calculate the average task arrival time 1 / μ and the average task processing time t j and the average cloud computing processing time t c Calculate and update.
[0028] Furthermore, in step 6, the historical data of the average task processing time refers to the allocation parameters m before the end of the pre-allocation phase. o -1 and m o The average task processing time at +1.
[0029] Furthermore, step 7 is specifically as follows:
[0030] If T(m o )≤T(m o -1) and T(m o )≤T(m o +1), then continue the stable operation phase and use the optimal allocation parameter m o , where T(m o ) is the task completion time under the optimal allocation parameters;
[0031] Otherwise, all computing nodes in edge computing are re-divided into two groups equally, where if T(m o )>T(m o -1), then the first group parameter m1 = m o -1, the second group parameter m2 = m o ; If T(m o )>T(m o +1), then the parameter m1 of the first group = m o , the second group parameter m2 = m o +1.
[0032] The beneficial effects described in the present invention are as follows: the method described in the present invention dynamically allocates parallel computing resources in the cloud-edge fusion computing system, and is compatible with parallel computing algorithms such as coding computing, thereby speeding up the system's processing of each task; through dynamic adjustment of computing resources, more computing tasks can be performed in the edge computing system, avoiding the large communication delay required to send to the cloud computing system, and greatly reducing the time for simulation platform deduction and calculation; the present invention can adaptively adapt to the impact of changes in task types and task arrival frequency in the system on system allocation parameters by dynamically switching between the pre-allocation stage and the stable operation stage. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 It is a schematic diagram of the application of the method of the present invention;
[0034] Figure 2 is a flow chart of the method of the present invention;
[0035] Figure 3 It is a schematic diagram of edge computing using parallel computing algorithms;
[0036] Figure 4 It is a schematic diagram of simulation results. DETAILED DESCRIPTION
[0037] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.
[0038] Combination Figure 1 The present invention describes a computing method based on dynamic allocation of parallel computing power under cloud-edge fusion, which is applied to the edge computing system. An algorithm is used to allocate several computing nodes to each task arriving at the edge computing system, and relevant parallel computing algorithms are adopted, such as replication and encoding. If there are no idle nodes in the system, the task will be sent to the cloud computing system for processing.
[0039] Combination Figure 2 The computing method based on dynamic allocation of parallel computing power under cloud-edge fusion described in the present invention comprises the following steps:
[0040] Step 1. Initialization: Combination Figure 3 , that is, the computing nodes of the edge computing system are evenly divided into two groups, in which each node in the first group processes one task alone, that is, the allocation parameter is m1=1; in the second group, every two nodes process one task, that is, the allocation parameter is m2=2. (The total number of edge computing nodes used in the algorithm of the present invention needs to be greater than 3).
[0041] Step 2: Allocation of computing resources (pre-allocation): The edge computing system allocates nodes to the arriving tasks by randomly selecting from any group of idle nodes. If there are no idle nodes, the task is sent to the cloud computing system for processing. If the number of idle nodes is less than the parameters m1 and m2, the task is sent to the cloud computing system. At the same time, the edge computing system calculates the average task arrival time 1 / μ and the average task processing time t of each group of nodes after processing s tasks. j , average cloud computing processing time t c , the calculation method of parameters is arithmetic mean. Combined with the simulation parameters in Table 1, the task is set to Poisson arrival, its rate is μ, the task processing time follows exponential distribution, and the rate parameter is 1 / t j , the cloud computing processing time is a fixed value t c .
[0042] Step 3: Calculate the task completion time (pre-allocation): that is, calculate the task blocking probability P(m) of the edge computing system under each set of allocation parameters using the time calculated in the second step. The formula is:
[0043]
[0044] Where a = N / m, is the number of tasks that the edge computing system can calculate simultaneously; N is the total number of edge computing nodes; m is the general term for the parameters of the node allocation method (when calculating the task blocking probability of the first group, m = m1; when calculating the task blocking probability of the second group, m = m2); then calculate the task completion time T(m) of each group under the corresponding allocation parameters, the formula is:
[0045] T(m)=(1-P(m))t j +t c (2)
[0046] Step 4: Determine the end of the pre-allocation phase: By comparing whether the task completion time of each group decreases with the allocation parameter, determine whether to end the pre-allocation phase, and dynamically optimize the parameters accordingly. If T(m1)≤T(m2), the pre-allocation phase ends, and the current optimal allocation parameter is m. o =m1; if T(m1)>T(m2), the pre-allocation phase continues, and the allocation parameters of the first group are adjusted to m1=m1+1, and the allocation parameters of the second group are adjusted to m2=m2+1.
[0047] Step 5: Computing resource allocation (stable): The system combines computing resources into one group and uses the optimal allocation parameter m obtained in the preprocessing stage. o Assign idle nodes to the arriving tasks. If there are no idle nodes, the tasks are sent to the cloud computing system for processing. If the number of idle nodes is less than the optimal allocation parameter m, o , the remaining idle nodes are assigned to the same task, and the task does not participate in the statistics of subsequent related time; at the same time, after the system processes each n tasks, it calculates the average task arrival time 1 / μ, the average task processing time t of the node, and the average task arrival time t of the node. j , average cloud computing processing time t c , the parameters are calculated using the arithmetic mean.
[0048] Step 6: Calculate the task completion time (stable): The time calculated in step 5 and the allocation parameters are m o -1 and m o +1, the corresponding latest statistical average task processing time historical data, according to formulas (1) and (2), calculate the current allocation parameters and adjacent allocation parameters (m o -1 and m o +1), and calculates the task completion time under the corresponding allocation parameters.
[0049] Step 7: Determine the end of the stable operation phase: By comparing the task completion time under the current allocation parameters and the adjacent allocation parameters, determine whether the stable operation phase is over and whether to perform dynamic optimization configuration of parameters. If T(mo )≤T(m o -1) and T(m o )≤T(m o +1), then continue the stable operation phase and use the current allocation parameter m o ; Otherwise, all computing nodes in edge computing are re-divided into two groups equally, where if T(m o )>T(m o -1), then the first group parameter m1 = m o -1, the second group parameter m2 = m o , if T(m o )>T(m o +1), then the parameter m1 of the first group = m o , the second group parameter m2 = m o +1.
[0050] The computing method based on dynamic allocation of parallel computing power under cloud-edge fusion described in the present invention is applied to the following edge computing system, and its simulation parameter settings are shown in Table 1. Figure 4 The simulation results verify that when the algorithm of the present invention is used, the task completion time is significantly shorter than when the algorithm of the present invention is not used. Especially when the task arrival rate is small, the algorithm of the present invention can further improve the performance of the edge computing system. Table 2 verifies the effectiveness of the dynamic computing resource allocation of the algorithm of the present invention. The edge computing system changes the task type or size every 500 tasks, making the average task processing time t j The size of the
[0051] Table 1 Simulation example parameter settings
[0052] Parameter name Parameter Value The total number of edge computing nodes N 20 Average task arrival time 1 / μ 1 / 7s Average task processing time tj 1s Average cloud computing processing time tc 10s Number of simulation tasks arrived 500 The number of tasks s for which the average time is calculated in the pre-allocation phase 10 The number of tasks n that calculate the average time during the stable operation phase 20
[0053] Table 2 Verification of the effectiveness of algorithm dynamic allocation
[0054] Average task processing time tj (seconds) 1 1.5 2 0.5 1.2 Using the algorithm of the present invention 0.341 0.795 1.717 0.225 0.563 Not used 1.042 1.596 2.228 0.516 1.169
[0055] In summary, the above computing resource allocation algorithm can effectively reduce the task computing time of the edge computing system.
[0056] The above description is only a preferred embodiment of the present invention and is not intended to be a further limitation of the present invention. All equivalent changes made using the contents of the present specification and drawings are within the protection scope of the present invention.
Claims
1. A computing method based on dynamic allocation of parallel computing power under cloud-edge fusion, It is characterized in that The following steps are involved: Step 1: Divide the computing nodes of the edge computing system into two groups evenly, and use different allocation parameters for each group of nodes; Step 2: The edge computing system assigns computing nodes to the arrived tasks, and calculates the average task arrival time, the average task processing time of each group of nodes, and the average cloud computing processing time during the observation period; Step 3: Using the time obtained in step 2, calculate the task blocking probability of the edge computing system under each group of allocation parameters, and then calculate the task completion time of each group under the corresponding allocation parameters; Step 4: By comparing whether the task completion time of each group decreases with the allocation parameters, it is determined whether to end the pre-allocation phase, and the parameters are dynamically optimized and configured accordingly; Step 5: Use the optimal allocation parameters in the pre-allocation phase to calculate and update the average task arrival time, the average task processing time of the node, and the average cloud computing processing time during the observation period; Step 6: Based on the time obtained in step 5 and the historical data of the average task processing time under different allocation parameters in step 3, the task blocking probability of the edge computing system under the current allocation parameters and adjacent allocation parameters is calculated, and the task completion time under the corresponding allocation parameters is calculated; Step 7: By comparing the task completion time under the current allocation parameters and the adjacent allocation parameters, determine whether the stable operation phase is ended; if so, re-enter the pre-allocation phase to perform dynamic optimization configuration of parameters; if not, complete the computing resource allocation.
2. According to the computing method based on dynamic allocation of parallel computing power under cloud-edge fusion according to claim 1, It is characterized in that In step 1, the computing nodes of the edge computing system are evenly divided into two groups. Each node in the first group processes a task independently, that is, the allocation parameter is m 1 =1; in the second group, every two nodes process one task, that is, the allocation parameter is m 2 =2.
3. According to the computing method based on dynamic allocation of parallel computing power under cloud-edge fusion according to claim 2, It is characterized in that In step 2, the edge computing system allocates computing nodes for the arriving tasks, specifically: Randomly select a computing node based on the number of assigned parameters from any group of idle nodes. If there is no idle node, send the task to the cloud computing system for processing; If the number of idle nodes is less than the parameter m 1 and m 2 , then the task is sent to the cloud computing system; The observation period refers to the edge computing system calculating the average task arrival time 1 / v and the average task processing time t of each group of nodes after processing s tasks. j , average cloud computing processing time t c , the parameters are calculated using the arithmetic mean.
4. According to the computing method based on dynamic allocation of parallel computing power under cloud-edge fusion according to claim 3, It is characterized in that In step 3, the computing system uses the task blocking probability P(m) of the edge computing system under each group of allocation parameters, and the formula is: Where a = N / m, which is the number of tasks that the edge computing system can calculate simultaneously; N is the total number of edge computing nodes; m is the general term for the parameters of the node allocation method. When calculating the task blocking probability of the first group, m = m 1 , when calculating the blocking probability of the second group of tasks, m = m 2 ; μ is the average task arrival rate, which is the inverse of the average task arrival time; i is the intermediate variable of the summation, and its value is an integer in [0, a]; Calculate the task completion time T(m) of each group under the corresponding allocation parameters, and the formula is: T(m)=(1-P(m))t j +t c (2)。 5. According to the computing method based on dynamic allocation of parallel computing power under cloud-edge fusion according to claim 4, It is characterized in that Step 4 is as follows: If T(m 1 )≤T(m 2 ), then the pre-allocation phase ends, and the current optimal allocation parameter is m o =m 1 ; Among them, T(m 1 ) is the task completion time of the first group of computing nodes, T(m 2 ) is the task completion time of the second group of computing nodes; If T(m 1 )>T(m 2 ), then continue the pre-allocation phase, and the allocation parameters of the first group are adjusted to m 1 =m 1 +1, the allocation parameter of the second group is adjusted to m 2 =m 2 +1.
6. According to the computing method based on dynamic allocation of parallel computing power under cloud-edge fusion according to claim 4, It is characterized in that In step 5, the optimal allocation parameter m obtained in step 4 is used o Assign idle nodes to the arriving tasks. If there are no idle nodes, the tasks are sent to the cloud computing system for processing. If the number of idle nodes is less than the optimal allocation parameter m, o , the remaining idle nodes are assigned to the same task, and the task does not participate in the subsequent statistics of related time.
7. According to the computing method based on dynamic allocation of parallel computing power under cloud-edge fusion according to claim 2, It is characterized in that In step 6, the historical data of average task processing time refers to the time before the end of the pre-allocation phase, when the allocation parameters are m o -1 and m o The average task processing time at +1.
8. According to the computing method based on dynamic allocation of parallel computing power under cloud-edge fusion according to claim 7, It is characterized in that Step 7 is as follows: If T(m o )≤T(m o -1) and T(m o )≤T(m o +1), then continue the stable operation phase and use the optimal allocation parameter m o , where T(m o ) is the task completion time under the optimal allocation parameters; Otherwise, all computing nodes in edge computing are re-divided into two groups equally, where if T(m o )>T(m o -1), then the parameter m of the first group 1 =m o -1, the second group's parameter m 2 =m o ; If T(m o )>T(m o +1), then the parameter m of the first group 1 =m o , the second group parameter m 2 =m o +1.
Citation Information
Patent Citations
LTE power wireless private network task unloading and resource allocation method based on cloud side cooperation
CN111585916A
Experiment scheme parallel deduction system based on dynamic task allocation
CN113255165A
Calculation power scheduling method, system and device based on cloud edge fusion and storage medium
CN117539619A
Dynamic channel reservation method considering channel quality in cognitive radio network
CN118042532A
Cited By
Computing method for unloading blocked task based on user retransmission mechanism under cloud edge fusion
CN119364434A
A computational method for unloading blocked tasks based on user retransmission mechanism in cloud-edge fusion
CN119364434B
