Distributed resource optimization scheduling method for ship cavitation simulation analysis

Abstract

The invention discloses a distributed resource optimization scheduling method for ship cavitation simulation analysis, and relates to the field of ship computer-aided engineering, and the method comprises the steps: carrying out the physical field coupling analysis of a ship cavitation simulation model, dynamically recognizing strong and weak coupling regions, and aggregating the strong coupling regions into independent fragments, fragmenting the weak coupling region based on a load balancing principle; constructing a plurality of levels of computing resource units, automatically matching the computing resource units with opposite resource granularities according to strong and weak coupling fragmentation results, and executing subtasks on the distributed computing resource units; dynamically adjusting the current computing resource unit based on the subtask prediction time consumption correction result to adapt to the resource requirements of each subtask; and outputting and integrating ship cavitation simulation calculation results. According to the method, the problems of low fragmentation quality, resource waste and large prediction deviation in cavitation simulation of a traditional method are solved, the efficiency and precision of ship cavitation simulation analysis are improved, and reliable technical support is provided for ship design and performance optimization.

Description

Technical Field

[0001] The present invention relates to the field of ship computer-aided engineering (CAE), and in particular to a distributed resource optimization scheduling method for ship cavitation simulation analysis. Background Art

[0002] The ship cavitation phenomenon is one of the key issues in the study of ship fluid mechanics, involving the analysis of the cavitation effects of propellers, rudders, hulls and other components under high-speed water flow. Cavitation simulation usually requires a high-precision computational fluid dynamics (CFD) model, combined with multi-physics field coupling (such as fluid-solid coupling, gas-liquid two-phase flow) for simulation, resulting in a large-scale calculation and extremely high complexity. For example, cavitation simulation often requires processing tens of millions of grid divisions, transient flow analysis and complex phase change processes, which places extremely high demands on computing resources. The traditional single-node computing mode can no longer meet the real-time and computational accuracy requirements of ship cavitation simulation. Although distributed computing can expand computing power through multi-node collaboration, the existing scheduling method has exposed significant defects in the strong coupling scenario unique to cavitation simulation:

[0003] Conflict between slicing strategies and physical field coupling characteristics: Traditional uniform meshing methods fail to fully account for the strong coupling effects between flow and structural fields in cavitation simulations (e.g., differences in physical properties between cavitated and non-cavitated areas on a propeller surface). This results in significant differences in the subtask sizes in strongly coupled areas (up to tens of times), exacerbating load imbalance.

[0004] Rigid resource adaptation: Static or experience-driven resource allocation strategies cannot adapt to the dynamic changes in cavitation simulation tasks (such as the dynamic evolution of the cavitation area over time), resulting in both resource waste and performance bottlenecks. In typical scenarios, CPU utilization is less than 60%.

[0005] ● Limited prediction accuracy: Existing resource prediction models rely on historical data or static characteristics and do not fully consider the nonlinear iterative convergence characteristics of cavitation simulation (such as the transient solution process of gas-liquid two-phase flow). This results in an error rate of more than 25% in resource allocation decisions, and there is a significant deviation between the prediction results and actual dynamic demand, which affects computational efficiency. Summary of the Invention

[0006] To address these challenges and meet these technical requirements, the inventors have proposed a distributed resource optimization and scheduling method for ship cavitation simulation analysis. This technical solution aims to optimize resource scheduling strategies in a distributed computing environment, improve the efficiency and accuracy of ship cavitation simulation analysis, and provide reliable technical support for ship design and performance optimization. This invention addresses the issue of strong coupling of physical fields and dynamic resource adaptation in ship cavitation simulation analysis scenarios, and proposes the following core, practical methods:

[0007] (1) Physical field-aware dynamic slicing method: In view of the strong coupling characteristics of the flow field and structural field in cavitation simulation (such as the highly dynamic changes in the cavitation area on the propeller surface), the model is dynamically divided into slices based on the physical field coupling strength (such as the energy transfer characteristics of the fluid-solid interface). Strongly coupled areas are aggregated into independent slices, and weakly coupled areas are sliced ​​according to the load balancing principle. The output slice results are compatible with the grid format of mainstream CAE software, avoiding the surge in communication overhead caused by traditional uniform slicing. This method can be directly integrated into the model preprocessing process without modifying the existing modeling tool chain, reducing deployment costs.

[0008] (2) Dynamic Adaptation Method for Elastic Resource Pools: In view of the dynamic changes in tasks in cavitation simulation (such as the transient evolution of cavitation areas), by constructing multi-level computing resource units (from micro to ultra-large), the resource granularity of the corresponding computing resource units is automatically matched according to the subtask scale, and dynamic adjustment of resource units (such as resource upgrades and upgrades) is supported to adapt to sudden changes in computing power demand. This mechanism is compatible with mainstream resource management frameworks such as Kubernetes, and implements resource expansion through lightweight plug-ins, avoiding the rigid configuration problem of traditional fixed thread pools.

[0009] (3) Hybrid prediction of subtask time consumption correction to drive resource adjustment: Combining the nonlinear iterative convergence characteristics of cavitation simulation (such as the transient solution process of gas-liquid two-phase flow), combined with parameter experience (such as grid size, material parameters) and pre-iteration calculation (residual convergence trend), the resource demand prediction value is dynamically corrected. The relationship between the subtask time consumption ratio and the threshold is automatically triggered to add task resources to avoid overall progress delays. The pre-iteration calculation time consumption is controlled within 5% of the total task duration, achieving a balance between prediction accuracy and efficiency to ensure that the main calculation process is not disturbed.

[0010] The beneficial technical effects of the present invention are:

[0011] Focusing on the characteristics of ship CAE cavitation virtual simulation scenarios, this application optimizes communication overhead, improves resource utilization, and speeds up key task chains through the synergy of three technologies: physical field coupling sharding rules, lightweight elastic resource scheduling, and low-overhead hybrid prediction. It solves the problems of low sharding quality, serious resource waste, and large prediction deviation in cavitation simulation caused by traditional methods. The entire process is compatible with industrial-grade tool chains and supports rapid verification and deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of the distributed resource optimization scheduling method for ship cavitation simulation analysis provided by this application.

[0013] Figure 2 This is a flow chart of the physical field-aware dynamic sharding method provided in this application.

[0014] Figure 3 This is a flow chart of the method for correcting the time consumption of hybrid prediction subtasks provided in this application.

[0015] Figure 4 This is a flow chart of the method for dynamically adjusting the current computing resource unit provided by this application. DETAILED DESCRIPTION

[0016] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0017] An embodiment of the present application provides a distributed resource optimization scheduling method for ship cavitation simulation analysis. The method takes optimizing resource scheduling strategies in a distributed computing environment and improving the efficiency and accuracy of ship cavitation simulation analysis as its core goal. It is built around three core technical points: dynamic sharding based on physical field perception, dynamic adaptation of elastic resources, and time-consuming correction of hybrid prediction subtasks to drive resource adjustment. It covers the entire process from model input to result output. The overall method is mainly divided into three stages: model preprocessing, resource allocation, dynamic scheduling, and computational execution. Each stage is linked through data flow and feedback mechanism. The overall process is referenced. Figure 1 As shown, specifically including the following:

[0018] (1) Model preprocessing stage

[0019] The system receives a ship cavitation simulation model (including geometric meshes, material parameters, boundary conditions and other data), and the sharding module performs physical field coupling analysis on the model, dynamically identifies strong and weak coupling areas, and aggregates strong coupling areas (such as the cavitation area on the propeller surface and the interface between the hull and the fluid) into independent shards to avoid the cross-node communication bottleneck caused by traditional uniform sharding. Weakly coupled areas (such as the internal structure of the hull and isolated fluid domains) are sharded based on the principles of grid density and load balancing. The sharding results are output as a set of subtasks with dependencies, which can be represented by a directed acyclic graph (DAG) in this embodiment. It is compatible with the grid formats of mainstream CFD software such as ANSYS Fluent and STAR-CCM+, ensuring that it can be directly integrated into the existing tool chain without reconstructing the model.

[0020] (2) Resource allocation and forecasting stage

[0021] The allocation module constructs five levels of computing resource units (L1 micro to L5 super large). Its number of threads, memory quota, and CPU binding strategy are designed according to the typical requirements of ship cavitation simulation tasks. It automatically matches computing resource units with relatively large resource granularity based on the results of strong and weak coupling sharding, such as the physical field characteristics of the subtask (such as coupling strength and grid size) and computational complexity (such as the time consumption of tasks such as transient solution of gas-liquid two-phase flow and nonlinear iteration in the cavitation region). In this embodiment, strong coupling sharding results (such as cavitation regions) can be directly allocated high-level computing resource units (such as L4 / L5 resources) due to the need for frequent iterations, while weak coupling sharding results are determined by comprehensively determining the required computing resource unit level through physical field characteristics and computational complexity.

[0022] The initially allocated computing resource units can then be dynamically adjusted in real time based on the dynamic changes in the task's requirements and the revised subtask prediction times. These revised subtask prediction times are calculated based on the subtask's grid size, computational complexity, and the convergence trend of the pre-iteration residual. Pre-iteration computational time is controlled to within 5% of the total task duration by limiting the number of initial iterations (typically 1%-3% of the total number of iterations) to avoid excessive computational resource usage.

[0023] (3) Dynamic scheduling and computation execution phase

[0024] The processing module executes subtasks on the allocated computing resource units, and the monitoring module collects data such as task progress, resource utilization, and residual convergence trends in real time. During the monitoring process, the current computing resource unit is dynamically adjusted based on the subtask prediction time correction results to adapt to the resource requirements of each subtask. For example, when resources are insufficient, a resource addition strategy is triggered, or when resources are excessive, a resource splitting strategy is triggered. The processing module outputs the ship cavitation simulation calculation results and integrates them. The independent slices of the strongly coupled task (such as the cavitation area) have exclusive resource units, so their iterative results are directly output locally. The results of the weakly coupled slices are merged into the output results of the strongly coupled slices according to the DAG dependency relationship to ensure the continuity of the physical field output in the final output.

[0025] In the model preprocessing stage (1), a physical field-aware dynamic sharding method is used. This method aims at the strong interactive characteristics of the fluid-solid coupling region in the ship cavitation simulation analysis (such as the dynamic changes of the cavitation region on the propeller surface), and proposes a coupling strength index CSI (coupling strength index) based on energy transfer density. By quantifying the energy exchange characteristics of the physical field, the sharding boundaries are dynamically divided to avoid the communication bottleneck caused by uniform sharding. Figure 2 As shown, the specific implementation steps of this method include:

[0026] 1) Calculate the energy transfer characteristics of the fluid-solid interface of the ship cavitation simulation model. Specifically, for each grid cell i of the ship cavitation simulation model, calculate its energy transfer amount:

[0027]

[0028] in, is the pressure field gradient vector of grid cell i, reflecting the rate of change of fluid pressure; is the displacement field gradient vector at grid cell i, representing the structural deformation rate. The physical significance of calculating energy transfer is that the dot product operation captures the synergy between pressure and displacement changes. A positive value indicates energy transfer from the fluid to the structure (e.g., compression deformation of a ship hull), while a negative value indicates energy transfer in the opposite direction (e.g., structural vibration affecting the flow field).

[0029] 2) The energy transfer characteristics are normalized by region to obtain the coupling strength index (CSI) of each region. The CSI is a dynamic indicator that quantifies the energy transfer intensity in the fluid-structure coupling region.

[0030] The coupling strength index CSI of the fluid-solid region R on the ship surface is defined as R for:

[0031]

[0032] Among them, N R is the total number of grid cells in region R; E base As the benchmark energy density, the average energy density of the typical fluid-solid coupling area in the historical mission (such as E base =1.2×10 3 J / m 3 ).

[0033] 3) Sharding decision rules

[0034] Basis for value selection: Through statistics of 200+ groups of ship cavitation simulation cases, the CSI value of the strong coupling area (such as the propeller-cavitation interface) is generally >0.8, and the CSI of the weak coupling area (such as the internal beam structure of the hull) is <0.5. Therefore, this embodiment is set as follows: If CSI ≥ 0.8, the area is identified as a strong coupling area, and the greedy algorithm is used to aggregate adjacent high-CSI units into independent shards to ensure that the iterative calculation is completed within a single resource unit and avoid cross-node communication. If CSI < 0.5, the area is identified as a weak coupling area, and the METIS graph partitioning algorithm is used to evenly partition the weak coupling area. The objective function is to minimize the difference in the number of grids between shards (limiting the difference to ≤ 15%).

[0035] Optionally, the independent shard aggregation algorithm can also use the DBSCAN clustering algorithm instead of the greedy algorithm to identify shard boundaries through density clustering, or use a graph convolutional network (GCN) to predict shard boundaries in strongly coupled areas. The specific implementation process of each method can refer to the existing algorithm implementation and will not be repeated here.

[0036] In this embodiment, the first CSI threshold is selected as 0.8 because of the significant sudden change in the fluid-structure coupling energy of the ship (such as the transient change in the cavitation area), and it is necessary to avoid misjudgment and excessive segmentation. Therefore, it is set higher than the conventional threshold of 0.6-0.7. base Calibration based on typical working conditions ensures the comparability of CSI between different models.

[0037] In the resource allocation and prediction phase (2), the elastic resource pool dynamic adaptation method is adopted, and the classification standards of the five-level computing resource units (L1-L5) are constructed as shown in the following table:

[0038] Table 1. L1-L5 computing resource unit classification standards

[0039]

[0040] The table shows NUMA (Non-Uniform Memory Access): a non-uniform memory access architecture. L4 / L5 resource units use a cross-NUMA domain binding strategy to reduce memory access latency. The strong-weak coupling sharding results are graded based on an analysis of over 200 ship cavitation simulation tasks, finding that 70% of tasks require 6-10 threads. The L5 level is reserved for the 5% of extreme-scale tasks (such as solving cavitation regions on meshes measuring tens of millions).

[0041] In the resource allocation and prediction stage (2), a hybrid prediction subtask time correction method is also used. This method combines the parameterized empirical model and the pre-iteration residual convergence trend. By correcting the subtask prediction time in real time, resource requirements can be dynamically adjusted in the dynamic scheduling and calculation execution stage (3), and the bottleneck acceleration strategy is used to prioritize the critical path tasks. Figure 3 As shown, the specific implementation steps of this method include:

[0042] 1) Build a parameter empirical model based on the grid size and computational complexity of the subtask to calculate the basic time consumption prediction value of the subtask:

[0043] T base =αN β +γC

[0044] Where: N is the number of grid nodes; C is the complexity of the computational domain (e.g., for two-phase gas-liquid flow, C = 3, and for the cavitation region, C = 5); α, β, and γ are regression coefficients obtained by fitting historical data. For example, α = 0.012, β = 1.15, and γ = 1.8.

[0045] 2) Considering the convergence trend of the pre-iteration residual, perform k = 3 initial iterations to calculate the pre-iteration residual decrease rate r:

[0046]

[0047] Where r k is the residual of the k-th iteration. In actual applications, measurements show that 3 iterations can cover more than 80% of the convergence trend patterns, and the time-consuming proportion is < 5%. Therefore, setting k = 3 in this example is reasonable.

[0048] 3) Combine the sub-task time-consuming correction rule for r: If r < w1 = 40%, it is determined that the convergence is slow, and then the predicted value of the basic time-consuming is corrected according to the r value as:

[0049]

[0050] If r ≥ w1 = 40%, then keep T pred = T base Finally, output T pred as the corrected value of the sub-task predicted time-consuming to the dynamic scheduling and calculation execution stage (3), as one of the conditions for resource dynamic adjustment judgment.

[0051] In the dynamic scheduling and calculation execution stage (3), a method of driving resource adjustment based on the corrected result of the sub-task predicted time-consuming is adopted. This method dynamically adapts to task requirements through vertical expansion (upgrading resource specifications) and horizontal splitting, and solves the problems of resource waste or insufficient performance caused by fixed resource granularity. As Figure 4 shown, the specific implementation steps of this method include:

[0052] 1) When it is monitored that the thread utilization rate M and the sub-task time-consuming ratio Y satisfy: M = (m o / m) ≥ w2%、Y ≥ threshold s and lasts for 5 minutes, upgrade the level of the current computing resource unit allocated to the sub-task (e.g., L2 → L3).

[0053] Where: m o is the number of active threads in the current computing resource unit, m is the total number of threads in the current computing resource unit; w2% is less than 100%. In this example, it is set as w2% = 85% to reserve 15% margin to handle sudden loads and avoid immediate overload after upgrading; the Y value is defined as the ratio of the corrected value of the sub-task predicted time-consuming to the predicted time-consuming of the overall task, reflecting the degree of blocking of the current task to the overall progress.

[0054] Optionally, set different thresholds s for different stages of task execution (early, mid-term, and final) to implement bottleneck acceleration strategies. For example, s = 0.25 for the early stage, s = 0.35 for the mid-term, and s = 0.4 for the final stage. A single task can trigger acceleration only three times within 30 minutes to avoid resource fluctuations.

[0055] 2) When it is detected that the computing resource level of the subtask is high (such as L4 / L5) and the thread utilization M and the subtask time ratio Y meet the following conditions: M = (m o / m)≤w3%, Y≥threshold s and lasts for 5 minutes, the current computing resource unit allocated to the subtask is horizontally split (horizontal splitting is only for L4 and L5 level resources, such as L5→L2, L3).

[0056] The w3% parameter should not exceed 50%. In this example, w3% = 50%. Horizontal splitting technology uses Kubernetes Custom Resource Definitions (CRDs) to implement resource unit splitting: ① Copy-on-Write memory sharing: During a split, only dirty pages (modified memory pages) are copied, reducing copy overhead (measured memory copying reduced by 60%). ② Split latency constraint: The time from triggering a split to the new unit being ready is ≤ 50ms, ensuring real-time performance.

[0057] Alternatively, horizontal splitting can also use the Apache Mesos or Slurm schedulers to implement similar hierarchical resource pool management through their APIs, or implement horizontal splitting based on the built-in service scaling mechanism of Docker Swarm Mode. The specific implementation process of each method can be referenced by existing algorithm implementations and will not be repeated here.

[0058] 3) If the above conditions 1) and 2) are not triggered, the current computing resource unit remains unchanged.

[0059] The above technical solution achieves significant performance improvements in large-scale ship cavitation simulation scenarios through the synergistic effect of three core inventions: physical field-aware sharding, dynamic adaptation of elastic resource pools, and hybrid prediction-driven allocation. Based on a virtual simulation case of propeller cavitation, with a grid size of 12 million and a computing cluster size of 50 nodes, the following key effects and field verification results are presented:

[0060] (1) Communication overhead optimization. After independently sharding the strongly coupled region (CSI ≥ 0.8) based on physical field-aware sharding technology, the frequency of data synchronization between nodes was reduced from once per iteration to once every five iterations. The weakly coupled region was balanced sharded based on METIS, and the grid difference between shards was controlled at 12%-15%, reducing the waiting time caused by uneven load. The final technical effect was: cross-node communication volume was reduced by 38%-42%, and the iteration time in the strongly coupled region was reduced by more than 50%.

[0061] (2) Improved resource utilization. The elastic resource pool is divided into L1-L5 resource units, covering lightweight preprocessing (L1) to large-scale solving (L5), and the task granularity matching is improved; a dynamic upgrade and downgrade mechanism is set: resource units trigger resource upgrades according to thread utilization (M ≥ 85%) and subtask time consumption ratio (Y ≥ threshold s). If the subtask computing resource level is L4 / L5 and the thread utilization (M ≤ 50%) and subtask time consumption ratio (Y ≥ threshold s) trigger resource splitting and downgrade, flexible resource scheduling is used to avoid resource idleness. The final technical effect is: the average utilization of computing resources is increased from 65% of the traditional solution to 92%, and the waste of idle resources is reduced by 80%.

[0062] (3) Critical path acceleration. By using a hybrid prediction model to correct pre-iteration time, the resource demand forecast error was reduced from ±25% to ±8%. The threshold s was dynamically adjusted according to the task phase (0.25 in the initial phase → 0.4 in the final phase), prioritizing the supply of resources for critical tasks. The resulting technical effect was a 46% reduction in overall task completion time and a 52% reduction in critical path time (the longest subtask chain).

[0063] The above description is only a preferred embodiment of the present application, and the present invention is not limited to the above embodiment. It is understood that other improvements and variations directly derived or imagined by those skilled in the art without departing from the spirit and concept of the present invention should be considered to be included in the scope of protection of the present invention.

Claims

1. A distributed resource optimization scheduling method for ship cavitation simulation analysis, characterized in that: The method comprises: Perform physical field coupling analysis on the ship cavitation simulation model, dynamically identify strong and weak coupling areas, aggregate the strong coupling areas into independent slices, and slice the weak coupling areas based on the load balancing principle; Constructing multi-level computing resource units, wherein the number of threads, memory quota, and CPU binding strategy of the computing resource units are hierarchically designed according to the typical requirements of ship cavitation simulation tasks; Automatically match computing resource units with relatively large resource granularity based on the strong and weak coupling sharding results, and execute subtasks on the allocated computing resource units; Dynamically adjust the current computing resource unit based on the subtask prediction time correction results to adapt to the resource requirements of each subtask; Output and integrate ship cavitation simulation results.

2. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 1 is characterized in that: The physical field coupling analysis of the ship cavitation simulation model is performed to dynamically identify strong and weak coupling areas, including: Calculating the energy transfer characteristics of the fluid-solid interface of the ship cavitation simulation model and normalizing them by region to obtain a coupling strength index for each region; If the coupling strength index is not less than a first threshold, the region is identified as a strong coupling region; if the coupling strength index is less than a second threshold, the region is identified as a weak coupling region; The first threshold is greater than the second threshold.

3. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 2 is characterized in that: Calculating the energy transfer characteristics of the fluid-solid interface of the ship cavitation simulation model and normalizing them by region includes: For each grid cell i of the ship cavitation simulation model, its energy transfer is calculated: in, is the pressure field gradient vector of the grid cell i, is the displacement field gradient vector of the grid unit i; The coupling strength index CSI of the fluid-solid region R on the ship surface is defined as R for: Among them, N R is the total number of grid cells in region R, E base is the base energy density.

4. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 1 is characterized in that: The method further comprises: The prediction time correction result of the subtask is determined based on the grid size, computational complexity and pre-iteration residual convergence trend of the subtask.

5. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 4 is characterized in that: Based on the grid size, computational complexity, and pre-iteration residual convergence trend of the subtask, the predicted time correction result of the subtask is determined, including: Calculate a basic time-consuming prediction value of the subtask based on the grid size and computational complexity of the subtask; Perform k initial iterations and calculate the pre-iteration residual reduction rate r; If r is less than w1%, the basic time consumption prediction value is corrected according to the r value and output as the subtask predicted time consumption correction value; otherwise, the basic time consumption prediction value is directly output as the subtask predicted time consumption correction value.

6. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 5 is characterized in that: The calculation formula for the basic time-consuming prediction value of the subtask is: T base =αN β +γC Where N is the number of grid nodes, C is the computational complexity, and α, β, and γ are the regression coefficients obtained by fitting historical data.

7. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 5 is characterized in that: The calculation formula for correcting the basic time-consuming prediction value according to the r value is: Among them, T base The estimated value of the basic time consumption for calculation.

8. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 5 is characterized in that: The number of initial iterations k is limited to 1%-3% of the total number of iterations to control the pre-iteration calculation time to within 5% of the total task duration and avoid excessive use of computing resources.

9. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 1 is characterized in that: The dynamically adjusting the current computing resource unit based on the subtask predicted time-consuming correction result includes: When the thread utilization M and subtask time ratio Y are monitored to meet the following conditions: M=(m o / m)≥w2%, Y is not less than the threshold s and lasts for a certain period of time, the current computing resource unit allocated to the subtask is upgraded; When the subtask resource level is monitored to be high, and the thread utilization M and subtask time ratio Y meet the following conditions: M = (m o / m)≤w3%, Y is not less than the threshold s and lasts for a certain period of time, the current computing resource unit allocated to the subtask is split; If the condition is not triggered, the current computing resource unit is maintained; Among them, m o is the number of active threads in the current computing resource unit, m is the total number of threads in the current computing resource unit, w2% is less than 100%, w3% is not greater than 50%, and Y is the ratio of the subtask predicted time correction value to the overall task predicted time.

10. The distributed resource optimization scheduling method for ship cavitation simulation analysis according to claim 9 is characterized in that: The method further comprises: Set the corresponding threshold s according to the different stages of task execution.

Images