A heterogeneous signal creation architecture simulation integrated computing management platform and method

Abstract

The application relates to an isomeric signal creation architecture simulation integrated computing management platform and method, belongs to the technical field of computing cluster management, and solves the problems that signal creation compliance and computing performance are difficult to be considered in the prior art, cross-architecture software adaptation and unified scheduling are absent, and cross-environment simulation business processes are fragmented. The platform comprises a signal creation core control layer and a non-domestic computing power execution layer, the signal creation core control layer is arranged on a domestic ARM architecture server, and the non-domestic computing power execution layer is arranged on a non-domestic x86 architecture cluster. The signal creation core control layer obtains a user task and matches a target node to generate a general task instruction; a cross-architecture scheduling adaptation module translates the general task instruction into an execution instruction recognizable by a corresponding scheduler of the target node; and the non-domestic computing power execution layer receives and executes the execution instruction. The signal creation compliance and computing performance are considered, and the unified scheduling of cross-architecture computing power resources and the full-process closed-loop management of simulation businesses are realized.

Description

Technical Field

[0001] This invention relates to the field of computing cluster management technology, and in particular to a heterogeneous information technology innovation architecture simulation integrated computing management platform and method. Background Technology

[0002] In the research and development of high-end equipment manufacturing, multi-disciplinary and multi-scale simulation calculations are required. This process relies on high-performance computing clusters, various types of professional simulation software, and graphical visualization workstations. Currently, with the advancement of information technology application innovation, core management processes need to be migrated to domestically produced platforms. However, high-precision simulation calculations and complex model graphics rendering place extremely high demands on computing power performance and software compatibility, necessitating the continued use of technologically mature non-domestic computing environments. This has created a deployment requirement for the heterogeneous coexistence of domestic and non-domestic computing power environments.

[0003] For the construction of simulation computing platforms, CN121277719A proposed a CAE software management system, which realizes the collaborative scheduling of CAE software and background computing resources through modules such as graphics rendering virtualization, job management and monitoring; CN104899404B further integrates the simulation cloud platform with HPC workstation clusters, realizing the decomposition and unified scheduling of simulation tasks. However, these solutions are all based on homogeneous hardware architecture design, and the scheduler adaptation is only for different scheduling systems under the same CPU architecture. They do not consider the heterogeneous deployment problems caused by the physical separation of domestic and non-domestic computing power environments and different CPU instruction sets.

[0004] Regarding heterogeneous software loading, CN119938159A proposes a mirror encapsulation model of "core components + multiple types of peripheral components," enabling the reuse and loading of the same software across different types of computing instances. CN121389470A achieves the embedding and integration of simulation software into the support platform through variable organization and interface adaptation. While these technologies have addressed the adaptation issues of cross-environment software deployment, their loading timing is all at the static deployment stage. They lack the ability to dynamically select corresponding components based on the target node's CPU architecture during task execution, and they do not address the issue of how to securely transmit simulation data generated in non-domestic computing environments to domestically controlled environments during task execution.

[0005] In addition, in professional fields such as materials calculation, CN111274667B proposed a cross-scale materials calculation software integration system, which realizes the unified calling of multiple types of calculation software and reduces the threshold for software use. However, this solution focuses on the integration of software functions and does not involve the problem of resource abstraction and unified scheduling in cross-architecture environments.

[0006] In summary, existing technologies, when deployed in heterogeneous scenarios where domestic ARM architecture and non-domestic x86 architecture coexist, cannot achieve unified scheduling of heterogeneous computing resources by the management end, nor can they achieve full-process cross-environment tracking of simulation tasks from application to result review. It is difficult to balance compliance requirements for domestic IT innovation with simulation computing performance. Therefore, there is an urgent need for an integrated computing platform and method that can bridge the gap between domestic IT innovation management environments and non-domestic computing execution environments. Summary of the Invention

[0007] Based on the above analysis, the embodiments of the present invention aim to provide a heterogeneous information technology innovation architecture simulation integrated computing management platform and method to solve the problems in the prior art that make it difficult to balance information technology innovation compliance and computing performance, lack of cross-architecture software adaptation and unified scheduling, and fragmentation of cross-environment simulation business processes.

[0008] On one hand, embodiments of the present invention provide a heterogeneous information technology innovation architecture simulation integrated computing management platform, including: a unified business control module, a cross-architecture scheduling and adaptation module, a heterogeneous computing power management module, a multi-software integration management module, and a distributed data security management module; The unified business management module is used to acquire user tasks, parse user tasks to obtain computing power requirements and software requirements, output computing power requirements to the heterogeneous computing power management module, output task software requirements to the multi-software integration management module, receive node matching results output by the heterogeneous computing power management module and software matching results output by the multi-software integration management module, generate general task instructions and target nodes and send them synchronously to the cross-architecture scheduling adaptation module and the distributed data security management module. The heterogeneous computing power management module is used to filter target nodes based on computing power requirements and real-time load, and output the node matching results to the unified business management module; The cross-architecture scheduling adaptation module is used to receive general task instructions and target nodes, translate the general task instructions into execution instructions that can be recognized by the scheduler corresponding to the target node, and send the execution instructions to the non-domestic computing power execution layer. The multi-software integration management module is used to receive software requirements matching software and return the software matching result to the unified business control module; The distributed data security management module is used to extract task data to be processed from the core control layer of the domestic IT innovation system and send it to the non-domestic computing power execution layer based on general task instructions and target node information.

[0009] Based on further improvements to the above platform, the cross-architecture scheduling adaptation module includes a scheduler adaptation unit, which includes a core terminal. The core terminal is used to convert general task instructions into execution instructions that can be recognized by the non-domestic computing power execution layer. The core of the scheduler adaptation unit receives the general task instruction issued by the unified business management module and executes the following translation process: S1: Parse the general task instructions and extract task metadata; S2: Parse the task metadata into a generic intermediate instruction model that is independent of the scheduler type; S3: Based on the scheduler type of the target node, query the built-in standardized adaptation rule library to obtain the corresponding general-native instruction mapping rules; S4: According to the general-native instruction mapping rules, the general intermediate instruction model is converted into the native execution script of the scheduler corresponding to the target node, and used as the execution instruction.

[0010] Based on further improvements to the platform, the adaptation rule base stores the mapping relationship between different scheduler types and the parameters of the intermediate instruction model; the core of the scheduler adaptation unit calls the corresponding mapping relationship from the adaptation rule base according to the scheduler type actually used by the target node, maps each parameter in the intermediate instruction model to job submission parameters that the scheduler can recognize, and combines them to generate the execution instruction.

[0011] Based on further improvements to the aforementioned platform, the scheduler adaptation unit also includes an execution end; the core end is deployed on the core management and control layer of the domestic IT innovation system, and the execution end is deployed on the non-domestic computing power execution layer; the core end is also used to send the execution instruction carrying the unique identifier of the task to the execution end; after receiving the native execution instruction, the execution end of the scheduler adaptation unit first queries the local execution record according to the unique identifier of the task. If the instruction corresponding to the identifier has been executed, it directly returns the recorded result without resubmitting; if it has not been executed, it performs syntax verification on the execution instruction and submits it to the local scheduler; after successful submission, the execution end of the scheduler adaptation unit obtains the job number returned by the local scheduler and binds it with the unique identifier of the task, and then sends it back to the core end of the scheduler adaptation unit. At the same time, it monitors the job status in real time and converts it into a standardized status before sending it back.

[0012] Based on further improvements to the aforementioned platform, the multi-software integration management module includes a core terminal deployed in the core control layer of domestic IT innovation and an execution terminal deployed in the non-domestic computing power execution layer. The core of the multi-software integration management module extracts the main program and general dependency library of each simulation software into a general kernel layer that is shared across architectures. It then splits the runtime dependency components required by the software under different CPU architectures into differentiated peripheral layers corresponding to different architectures. Finally, it associates and packages the general kernel layer with each differentiated peripheral layer into a composite image and stores it. After receiving the software requirements output by the unified business control module, the core of the multi-software integration management module matches the corresponding target software according to the software identifier in the software requirements, generates a software matching result, and sends the software matching result to the execution ends of the unified business control module and the multi-software integration management module respectively; the software matching result includes the composite image storage address of the target software and the correspondence between its supported CPU architecture type and differentiated peripheral layer; The execution end of the multi-software integration management module loads the general kernel layer from the composite image storage address according to the CPU architecture type of the target node issued by the cross-architecture scheduling and adaptation module, and only loads the differentiated peripheral layer corresponding to the CPU architecture type, and combines them to construct the local executable software runtime environment of the target node and start the target software.

[0013] Based on further improvements to the above platform, the heterogeneous computing power management module includes an intelligent scheduling unit deployed in the core control layer of domestic IT innovation and a monitoring agent deployed in each node of the non-domestic computing power execution layer. The monitoring agent collects the load indicators of the node it is located at a preset period and reports them to the intelligent scheduling unit. The intelligent scheduling unit receives the load indicators of each node and the computing power requirements output by the unified business management module. Based on the resource conditions in the computing power requirements, it selects candidate nodes from each node, sorts them by comprehensive score according to the current load of each candidate node, determines the node with the best score as the target node, and returns the node matching result to the unified business management module.

[0014] Based on further improvements to the platform, the comprehensive score ranking is performed according to the following formula: S i =ω c ·C match +ω m ·M match +ω g ·G match -ω l ·L i in, S i Assign a comprehensive fit score to the i-th candidate node. oh c , oh m , oh g , oh l These are the weighting coefficients for CPU, memory, GPU, and node load, respectively. C match ,M match , G match These represent the degree to which the node's CPU, memory, and GPU resources meet the task requirements. L i This represents the current overall load rate of the node.

[0015] Based on further improvements to the aforementioned platform, the distributed data security management module includes a data synchronization unit deployed in the core control layer of the domestic IT innovation system and a data synchronization agent deployed in the execution layer of non-domestic computing power. Before the cross-architecture scheduling and adaptation module issues the execution instruction, the data synchronization unit sends the task data to be processed to the data synchronization agent deployed on the non-domestic computing power execution layer according to the data list in the general task instruction. The data synchronization agent then sends the data to be processed to the target node. After the task is executed, the data synchronization agent obtains the calculation result file from the target node and sends it back to the data synchronization unit.

[0016] Based on further improvements to the aforementioned platform, the platform also includes a data isolation unit. The data isolation unit is deployed on the non-domestic computing power execution layer. Before the target node executes the task, the data isolation unit creates an independent isolated running environment for the task through operating system-level virtualization technology, restricts the process space, file access path, and network communication of the task to the isolated running environment, and binds the task to the specified CPU core and GPU device through cgroups technology.

[0017] On the other hand, embodiments of the present invention provide a heterogeneous information technology innovation architecture simulation integrated computing management method, including the following steps: The unified business management module generates tasks, outputs computing power requirements to the heterogeneous computing power management module, and outputs software requirements to the multi-software integration management module. The heterogeneous computing power management module filters target nodes based on computing power requirements and real-time load, and outputs the node matching results to the unified business management module; The multi-software integration management module matches software based on software requirements and outputs the software matching results to the unified business control module; The unified business management module generates general task instructions based on task, node matching results and software matching results. The unified business management module then synchronously sends the general task instructions and target nodes to the cross-architecture scheduling adaptation module and the distributed data security management module. The cross-architecture scheduling adaptation module receives general task instructions and target nodes, translates the general task instructions into execution instructions that can be recognized by the scheduler corresponding to the target node, and sends the execution instructions to the non-domestic computing power execution layer. The distributed data security management module extracts task data to be processed from the core control layer of domestic IT innovation and sends it to the non-domestic computing power execution layer based on general task instructions and target node information. After receiving the execution instructions and the data to be processed, the non-domestic computing power execution layer starts the task execution; The distributed data security management module sends the execution results to the unified business control module, which triggers the result review process. Once the review is passed, the project is officially archived.

[0018] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. Adopting a hierarchical and heterogeneous deployment architecture of the core control layer of domestic IT innovation and the non-domestic computing power execution layer, core modules such as unified business control, cross-architecture scheduling and adaptation, heterogeneous computing power management, distributed data security management and full-link security audit are deployed on domestic ARM architecture servers equipped with Phytium CPUs to meet the compliance requirements of domestic IT innovation. At the same time, high-performance simulation computing tasks are scheduled to be executed on non-domestic x86 architecture clusters to give full play to their computing power performance and compatibility advantages with commercial simulation software. Thus, while ensuring the localization of core control links, the high-performance requirements of simulation computing are taken into account, solving the technical problem of difficulty in balancing domestic IT innovation compliance and computing power performance in existing technologies. 2. Through the cross-architecture scheduling adaptation module, general task instructions are abstracted into intermediate instruction models that are independent of the scheduler type. Then, based on the scheduler type of the target node, the corresponding mapping rules are selected from the adaptation rule base to restore the native execution script that the target scheduler can recognize. This enables the core control layer of the information technology innovation to uniformly call different scheduling systems. At the same time, the multi-software integration management module adopts a composite image encapsulation method that separates the general kernel layer and the differentiated peripheral layer. This allows the simulation software to be encapsulated once and loaded on demand under both ARM and x86 architectures. There is no need to maintain software images separately for different architectures, which greatly reduces the software deployment and maintenance costs in heterogeneous environments. 3. Through the coordinated operation of the unified business control module, cross-architecture scheduling and adaptation module, and distributed data security management module, the flow of the same business object associated with the task is realized between the modules by using the unique identifier of the task. This enables closed-loop management of the entire process of simulation tasks across environments, from application, approval, decomposition, resource matching, instruction translation, data synchronization, isolated execution to result feedback, review and archiving. At the same time, the full-link security audit module ensures the integrity of audit logs with cryptographic chain digests and digital signatures, making cross-environment business processes traceable and solving the problem of business link breakage after the physical separation of the control end and the execution end.

[0019] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0020] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0021] Figure 1 This is a schematic diagram of a heterogeneous information technology innovation architecture simulation integrated computing management platform in Embodiment 1 of the present invention. Detailed Implementation

[0022] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0023] Example 1, A specific embodiment of the present invention discloses a heterogeneous information technology innovation architecture simulation integrated computing management platform, such as... Figure 1 As shown, it includes: a unified business management module, a cross-architecture scheduling and adaptation module, a heterogeneous computing power management module, a multi-software integration management module, and a distributed data security management module; The unified business management module is used to acquire user tasks, parse user tasks to obtain computing power requirements and software requirements, output computing power requirements to the heterogeneous computing power management module, output task software requirements to the multi-software integration management module, receive node matching results output by the heterogeneous computing power management module and software matching results output by the multi-software integration management module, generate general task instructions and target nodes and send them synchronously to the cross-architecture scheduling adaptation module and the distributed data security management module. The heterogeneous computing power management module is used to filter target nodes based on computing power requirements and real-time load, and output the node matching results to the unified business management module; The cross-architecture scheduling adaptation module is used to receive general task instructions and target nodes, translate the general task instructions into execution instructions that can be recognized by the scheduler corresponding to the target node, and send the execution instructions to the non-domestic computing power execution layer. The multi-software integration management module is used to receive software requirements matching software and return the software matching result to the unified business control module; The distributed data security management module is used to extract task data to be processed from the core control layer of the domestic IT innovation system and send it to the non-domestic computing power execution layer based on general task instructions and target node information.

[0024] It should be noted that the core inventive concept of this application lies in adopting a hierarchical and heterogeneous architecture of a domestically developed core control layer and a non-domestic computing power execution layer. The domestically developed core control layer is deployed on a domestically produced ARM architecture server equipped with Phytium series CPUs, adapted to the domestically produced Kylin operating system, and carries the platform's core business control, data security storage, and compliance auditing functions. The non-domestic computing power execution layer is deployed on a non-domestic x86 architecture high-performance computing cluster and visualization server, responsible for high-performance simulation computing and complex model graphics rendering. The two layers achieve cross-architecture and cross-environment interconnection and interoperability through an encrypted communication bus based on the national cryptographic SM2 / SM4 encryption algorithm.

[0025] The core management and control layer of the domestic IT innovation platform also deploys a domestic IT innovation adaptation base module, which is used to adapt the platform's core services to Phytium series ARM architecture CPUs and the Kylin operating system, and to build a cross-architecture compatible operating environment for ARM and x86 architectures, providing underlying technical support for cross-architecture communication and scheduling. The unified business management and control module is also deployed in the core management and control layer of the domestic IT innovation platform. As the core management and control hub of the platform, it is used to realize the unified management of the entire simulation business process and the synchronization of cross-environment status. Internally, it uses task records as the data exchange carrier, and associates business data generated by modules such as cross-architecture scheduling adaptation, heterogeneous computing power management, multi-software integration management, and distributed data security management through unique task identifiers, project identifiers, and user identifiers, thereby connecting the complete business link from task generation to result archiving.

[0026] The cross-architecture scheduling adaptation module, heterogeneous computing power management module, multi-software integration management module, and distributed data security management module all adopt a core-end and execution-end collaboration approach: the core end is located in the domestic IT innovation core control layer, responsible for saving configurations, generating task contexts, issuing control commands, and summarizing status; the execution end or execution agent is located in the non-domestic computing power execution layer, responsible for collecting local resources, verifying commands, calling local schedulers or software, executing data synchronization, and returning results. The core end and execution end are associated with the same business object through unique task identifiers, unique project identifiers, and unique user identifiers to avoid inconsistent execution statuses across environments. The unified business control module is deployed in the domestic IT innovation core control layer, serving as the platform's core control hub, without distinguishing between the core end and the execution end. Through the above-mentioned hierarchical heterogeneous deployment and end-to-end collaboration mechanism, this application fully guarantees the computing performance and software compatibility of industrial simulation while meeting the domestic IT innovation compliance requirements.

[0027] Furthermore, the cross-architecture scheduling adaptation module includes a scheduler adaptation unit, which includes a core terminal. The core terminal is used to convert general task instructions into execution instructions that can be recognized by the non-domestic computing power execution layer. The core of the scheduler adaptation unit receives the general task instruction issued by the unified business management module and executes the following translation process: S1: Parse the general task instructions and extract task metadata; S2: Parse the task metadata into a generic intermediate instruction model that is independent of the scheduler type; S3: Based on the scheduler type of the target node, query the built-in standardized adaptation rule library to obtain the corresponding general-native instruction mapping rules; S4: According to the general-native instruction mapping rules, the general intermediate instruction model is converted into the native execution script of the scheduler corresponding to the target node, and used as the execution instruction.

[0028] It's important to note that S1 and S2 abstract common task instructions into an intermediate instruction model independent of the scheduler type. This model shields the syntactic differences between different schedulers, allowing upper-layer business logic to disregard the specific type of the underlying scheduler. The common intermediate instruction model includes fields such as unique task identifier, user identifier, software identifier and version, target architecture type, number of CPU cores, memory capacity, working directory, entry script, parameter list, and input file list. S3 and S4, in turn, select the corresponding mapping rules from the adaptation rule base based on the actual scheduler type of the target node, restoring the intermediate instruction model into a native execution script recognizable by the target scheduler. This abstraction-then-restoration translation mechanism allows the core of the cross-architecture scheduling adaptation module to support multiple schedulers by maintaining only a unified intermediate instruction model and an extensible adaptation rule base. When a new scheduler type needs to be added, only the corresponding mapping rules need to be added to the rule base; no modification to the upper-layer business logic is required, demonstrating excellent scalability.

[0029] Furthermore, the adaptation rule base stores the mapping relationship between different scheduler types and the parameters of the intermediate instruction model; the core of the scheduler adaptation unit calls the corresponding mapping relationship from the adaptation rule base according to the scheduler type actually used by the target node, maps the parameters in the intermediate instruction model to job submission parameters that the scheduler can recognize, and combines them to generate the execution instruction.

[0030] It's important to note that the adaptation rule base stores the mapping relationships between different scheduler types and parameters of the general intermediate instruction model. The core of the scheduler adaptation unit, based on the actual scheduler type used by the target node, retrieves the corresponding mapping relationship from the adaptation rule base, mapping each parameter field in the intermediate instruction model to job submission parameters recognizable by that scheduler, and then combining them to generate a complete native execution script. For example, the number of CPU cores in the intermediate instruction model is mapped to the `cpus-per-task` parameter under the Slurm scheduler and the `select:ncpus` parameter under the PBS scheduler; memory capacity is mapped to the `mem` parameter under both Slurm and PBS; the working directory is mapped to the `chdir` parameter under Slurm and the `d` parameter under PBS. The adaptation rule base adopts a configurable design; when adding a new scheduler type, only the corresponding mapping rules need to be added to the rule base, without modifying the core code of the scheduler adaptation unit.

[0031] Furthermore, the scheduler adaptation unit also includes an execution end; the core end is deployed on the domestic IT innovation core management and control layer, and the execution end is deployed on the non-domestic computing power execution layer; the core end is also used to send the execution instruction carrying the unique identifier of the task to the execution end; after receiving the native execution instruction, the execution end of the scheduler adaptation unit first queries the local execution record according to the unique identifier of the task. If the instruction corresponding to the identifier has been executed, it directly returns the recorded result without resubmitting. If it has not been executed, it performs syntax verification on the execution instruction and submits it to the local scheduler; after successful submission, the execution end of the scheduler adaptation unit obtains the job number returned by the local scheduler and binds it with the unique identifier of the task, and then sends it back to the core end of the scheduler adaptation unit. At the same time, it monitors the job status in real time and converts it into a standardized status before sending it back.

[0032] It should be noted that the scheduler adaptation unit adopts a separate deployment architecture for the core and execution ends. The core end is located in the core control layer of the domestic IT innovation system, while the execution end is located in the execution layer of non-domestic computing power. The two interact through an SM2 / SM4 encrypted communication bus and connect via a RESTful API interface. When issuing execution instructions, the core end generates an idempotent key by combining the unique task identifier and the instruction sequence number. After receiving the instruction, the execution end first queries the local execution record based on the idempotent key. If the instruction has already been executed, it directly returns the recorded result without resubmitting, thus preventing duplicate submissions caused by network retransmissions. If the instruction has not been executed, the execution end performs signature verification, timestamp verification, random number verification, and script syntax validity verification on the execution instruction in sequence. After all verifications pass, it calls the native interface of the local scheduler to submit the job. After successful submission, the execution end obtains the job number returned by the scheduler, binds it with the unique task identifier, and sends it back to the core end. The execution end also continuously monitors the job status, converting the original status code returned by the scheduler into six standardized statuses: queued, running, paused, successful, failed, and canceled, and sends them back to the core end. At the same time, it retains the original status code and error information for auditing and traceability. Through the above mechanism, the core management and control layer of information technology innovation can track and transparently perceive the submission, execution and status changes of tasks in the non-domestic computing power layer throughout the entire process.

[0033] Furthermore, the multi-software integration management module includes a core terminal deployed in the core control layer of domestic IT innovation and an execution terminal deployed in the non-domestic computing power execution layer; The core of the multi-software integration management module extracts the main program and general dependency library of each simulation software into a general kernel layer that is shared across architectures. It then splits the runtime dependency components required by the software under different CPU architectures into differentiated peripheral layers corresponding to different architectures. Finally, it associates and packages the general kernel layer with each differentiated peripheral layer into a composite image and stores it. After receiving the software requirements output by the unified business control module, the core of the multi-software integration management module matches the corresponding target software according to the software identifier in the software requirements, generates a software matching result, and sends the software matching result to the execution ends of the unified business control module and the multi-software integration management module respectively; the software matching result includes the composite image storage address of the target software and the correspondence between its supported CPU architecture type and differentiated peripheral layer; The execution end of the multi-software integration management module loads the general kernel layer from the composite image storage address according to the CPU architecture type of the target node issued by the cross-architecture scheduling and adaptation module, and only loads the differentiated peripheral layer corresponding to the CPU architecture type, and combines them to construct the local executable software runtime environment of the target node and start the target software.

[0034] It should be noted that simulation software running on different CPU architectures relies on different underlying libraries and compilation environments. Maintaining a separate software image for each architecture would result in a significant increase in the number of images and make it difficult to guarantee the consistency of software versions across different architectures. To address this issue, the multi-software integration management module of this application adopts a composite image encapsulation method that separates the general kernel layer and differentiated peripheral layers. The core end uses the main program of the simulation software and the cross-architecture common basic dependency libraries as the general kernel layer, and separates the runtime dependency components required by different CPU architectures into corresponding differentiated peripheral layers. Then, the general kernel layer and each differentiated peripheral layer are associated and packaged into the same composite image for storage. When matching software, the core end matches the target software according to the software requirements, generates a software matching result containing the storage address of the composite image and the correspondence between each architecture and the peripheral layer, and sends it to the unified business control module and the execution end respectively. After receiving the software matching result, the execution end loads only the general kernel layer and the differentiated peripheral layer corresponding to the target node's CPU architecture type, and combines them locally into a complete executable software environment. This packaging method allows the same composite image to serve both ARM and x86 architecture computing nodes simultaneously. When new or updated software is needed, only one composite image needs to be maintained to cover the operational requirements of multiple architectures, which reduces storage and maintenance costs and avoids the problem of inconsistent software versions in a multi-architecture environment.

[0035] Furthermore, the heterogeneous computing power management module includes an intelligent scheduling unit deployed in the core control layer of domestic IT innovation and a monitoring agent deployed in each node of the non-domestic computing power execution layer; The monitoring agent collects the load indicators of the node it is located at a preset period and reports them to the intelligent scheduling unit. The intelligent scheduling unit receives the load indicators of each node and the computing power requirements output by the unified business management module. Based on the resource conditions in the computing power requirements, it selects candidate nodes from each node, sorts them by comprehensive score according to the current load of each candidate node, determines the node with the best score as the target node, and returns the node matching result to the unified business management module.

[0036] It should be noted that the core control layer of the domestic IT innovation system needs to have real-time awareness of the resource status of each node in the non-domestic computing power execution layer in order to make reasonable scheduling decisions. To this end, the heterogeneous computing power management module of this application deploys a resident lightweight monitoring agent on each node of the non-domestic computing power execution layer. Upon startup, the agent registers static attributes such as node identifier, CPU architecture, operating system, CPU and GPU models, total memory, and total storage with the core. During operation, it calls the operating system monitoring interface to collect dynamic indicators at preset intervals. Linux nodes read the proc and sys file systems and call NVML or nvidia-smi to obtain GPU indicators, while Windows visualization nodes obtain CPU, memory, disk, network, and graphics resource indicators through WMI or performance counters. The agent reports the collected raw monitoring data to the resource monitoring unit of the core control layer of the domestic IT innovation system via an encrypted communication link. If the reporting fails, it is temporarily stored in a local queue and retried. After receiving the data, the resource monitoring unit performs identity verification, duplicate data filtering, outlier filtering, and time window aggregation, outputting the real-time node status of each node, which is then processed by the resource abstraction unit. The resource abstraction unit uses a field mapping table to convert inconsistent monitoring fields in real-time node status, which vary due to different CPU architectures, GPU models, and operating systems, into standardized resource objects in a unified format. These objects include fields such as node identifier, CPU architecture, operating system type, number of available CPU cores, available memory, GPU model and available quantity, remaining storage, network bandwidth, current overall load rate, availability of software architecture components, recent heartbeat time, and health status. These objects are then written into a standardized resource pool, thereby shielding the differences in underlying hardware architecture and device models and providing consistent candidate node data for the intelligent scheduling unit.

[0037] After receiving the computing power requirements from the unified business management module, the intelligent scheduling unit first selects candidate nodes from the standardized resource pool that meet the hard constraints based on resource conditions such as task type, number of CPU cores, memory capacity, GPU model and quantity, maximum runtime, and software architecture component requirements. Then, for the selected candidate nodes, a comprehensive score is calculated and ranked according to the degree to which each node's current CPU, memory, and GPU resources meet the task requirements, as well as its current overall load rate. The scoring mechanism uses differentiated weighting coefficients for different task types, such as CPU-intensive and GPU-intensive tasks. The higher the task's demand for a certain type of resource, the greater the weight of that resource's satisfaction in the score. Simultaneously, the current overall load rate is included as a penalty in the score. The intelligent scheduling unit determines the node with the best score as the target node and returns the node matching result to the unified business management module. During task execution, the monitoring agent continues to report load and heartbeats. The intelligent scheduling unit updates the resource pool in real time accordingly. When a node fails to report a heartbeat for several consecutive cycles or the job exits abnormally, the candidate node selection and matching process is automatically re-executed, transferring the task to other available nodes. Through the above mechanism, this application can achieve optimal matching and dynamic load balancing of tasks and resources in heterogeneous clusters while meeting the task resource requirements.

[0038] Furthermore, the overall score ranking is performed according to the following formula: S i =ω c ·C match +ω m ·M match +ω g ·G match -ω l ·L i in, S i Assign a comprehensive fit score to the i-th candidate node. oh c , oh m , oh g , oh l These are the weighting coefficients for CPU, memory, GPU, and node load, respectively. C match , M match , G match These represent the degree to which the node's CPU, memory, and GPU resources meet the task requirements. L i This represents the current overall load rate of the node.

[0039] It should be noted that the formula uses the highest task resource requirement satisfaction and optimal load balancing among heterogeneous nodes in the cluster as dual optimization objectives. CPU satisfaction, memory satisfaction, and GPU satisfaction reflect the degree to which candidate nodes meet task requirements across various resources, with values ​​ranging from 0 to 1. Nodes failing to meet basic requirements are directly set to 0 and eliminated. The load rate represents the node's current overall load level and is deducted as a penalty from the weighted sum of the first three factors; higher loads result in greater deductions, thus guiding scheduling to favor nodes with relatively low workloads. The weighting coefficients for CPU, memory, GPU, and load rate are dynamically configured based on the task type, with a weighted sum of 1. For CPU-intensive tasks, CPU satisfaction has a higher weight, and for GPU-intensive tasks, GPU satisfaction has a higher weight. Through this formula, the intelligent scheduling unit unifies resource supply and demand matching and load balancing objectives within a single scoring framework, achieving optimal matching between tasks and nodes.

[0040] Furthermore, the distributed data security management module includes a data synchronization unit deployed in the core control layer of the domestic IT innovation system and a data synchronization agent deployed in the execution layer of non-domestic computing power; Before the cross-architecture scheduling and adaptation module issues the execution instruction, the data synchronization unit sends the task data to be processed to the data synchronization agent deployed on the non-domestic computing power execution layer according to the data list in the general task instruction. The data synchronization agent then sends the data to be processed to the target node. After the task is executed, the data synchronization agent obtains the calculation result file from the target node and sends it back to the data synchronization unit.

[0041] It should be noted that the core control layer of the domestic IT innovation system is physically isolated from the execution layer of the non-domestic computing power, and the data required for task execution cannot be directly accessed through shared storage. Therefore, the distributed data security management module of this application deploys a data synchronization unit in the core control layer of the domestic IT innovation system and a data synchronization agent in each node of the execution layer of the non-domestic computing power. During task scheduling, the data synchronization unit generates a synchronization plan based on the data list in the general task instruction, and sends it to the data synchronization agent of the target node via an encrypted communication link. The agent then writes the data to the local storage of the target node. During data synchronization, the data synchronization unit employs an adaptive block partitioning, rolling hashing for rapid location, and precise SM3 verification algorithm for difference comparison. First, the block size is adaptively set according to file type. The source and target files are then partitioned into blocks, and rolling hash values ​​and SM3 checksums are calculated separately. Difference blocks are located through this two-dimensional comparison, and identical data blocks are filtered out. Then, the differing blocks are extracted as incremental data. When network conditions and security policies support this, RDMA high-speed network protocol is used for transmission; otherwise, encrypted TCP or a dedicated data transmission channel is used. The receiving end confirms, resumes interrupted transmissions, and retries failures based on the block sequence number. After all data is written, file merging and consistency verification are performed. After task execution, the data synchronization agent reads the calculation result file from the target node and uses the same mechanism to send it back to the data synchronization unit in the core management layer of the IT innovation system. The data synchronization unit then writes it to the storage pool on the IT innovation side. Through this difference comparison and incremental synchronization mechanism, the amount of cross-environment data transmission is significantly reduced, while ensuring eventual consistency between the source and target data.

[0042] Furthermore, the platform also includes a data isolation unit, which is deployed on the non-domestic computing power execution layer. Before the target node executes the task, the data isolation unit creates an independent isolated running environment for the task through operating system-level virtualization technology, restricts the process space, file access path and network communication of the task to the isolated running environment, and binds the task to the specified CPU core and GPU device through cgroups technology.

[0043] It should be noted that when tasks are executed on target nodes of non-domestic computing power execution layers, multiple tasks may run simultaneously on the same node. Without isolation, the processes, files, and data of different tasks will be visible to each other, posing a risk of data leakage and mutual interference. Therefore, the data isolation unit creates an independent isolated runtime environment for each task using Linux Namespace technology before executing the task on the target node.

[0044] Specifically, the data isolation unit creates independent Mount, PID, IPC, Network, and User namespaces through the `unshare` system call or the `clone` function with corresponding flags, achieving full-dimensional isolation of file systems, processes, inter-process communication, networks, and user permissions between tasks. Specifically, the Mount namespace sets the root mount propagation attribute to private, uses tmpfs or a pre-configured root file system as the task root directory, binds the input data directory to the mount in read-only mode, and the output directory to the mount in write mode, and restricts the process visibility path to the task root directory using `pivot_root` or `chroot`. The User namespace maps platform users to users within the sandbox. The Network namespace does not configure external network interfaces by default, only accessing the controlled network through the veth device when a task needs to connect to a license server or audit interface, and using iptables or nftables whitelists to only open preset destination addresses and ports. Meanwhile, the data isolation unit uses Linux cgroups technology to bind tasks to specified CPU cores and GPU devices: CPU binding uses the cpuset subsystem of cgroups to configure the cpuset.cpus and cpuset.mems parameters, binding the sandbox process to a specified CPU thread and corresponding memory node; GPU binding uses the devices subsystem of cgroups to control the access permissions of GPU device nodes, combined with the CUDA_VISIBLE_DEVICES or ROCR_VISIBLE_DEVICES environment variables to limit the GPU cores visible to the task. Before the task starts, the data isolation unit verifies the namespace identifier, mount table, directory owner and permissions, cgroup members, and GPU-visible devices. Once all verifications are successful, the sandbox is ready. After the task completes, it sequentially terminates any remaining processes within the cgroup, refreshes the output file, unbinds the device and CPU, unmounts the task directory, destroys the namespace, and cleans up the temporary directory to prevent subsequent tasks from accessing data left over from the previous task. Through the above isolation and binding mechanisms, this application achieves secure isolated execution and exclusive resource access for multiple tasks on a shared physical node.

[0045] It should be noted that the platform also includes a full-link security audit module, deployed at the core control layer of the domestic IT innovation platform. This module includes a user operation audit unit and an authorization audit unit. The user operation audit unit records all user actions such as login / logout, file download / deletion, software startup, and task submission, and highlights high-risk operations such as large-scale file downloads. The authorization audit unit records all operations performed by permission administrators on user roles and permissions, enabling full traceability of permission changes.

[0046] The end-to-end security audit module records the event identifier, task identifier, user identifier, node identifier, operation type, operation object, operation result, timestamp, source address, and digest value for each audit event. It forms a sequential audit chain according to the formula Hn=SM3(Hn-1||event_data||timestamp||nonce). Every preset number of events or preset time periods, the digest of the current chain head is signed using the SM2 private key. When querying the audit log, the event digest, preceding digest, and signature are verified simultaneously to detect log deletion, insertion, or modification. For behaviors such as batch downloads, unauthorized access, and continuous authentication failures within a short period, alarms are generated based on configurable thresholds and associated with the corresponding task identifier or user identifier.

[0047] The end-to-end security audit module establishes audit event connections with the domestic IT innovation adaptation base module, unified business control module, heterogeneous computing power management module, multi-software integration management module, cross-architecture scheduling adaptation module, and distributed data security management module. Core components deployed in the core control layer of the domestic IT innovation system call the audit event collection interface through the unified audit software development kit. Execution agents, node monitoring agents, software execution components, data synchronization agents, and task sandbox execution components deployed on non-domestic computing nodes and visualization servers send events to the audit collection gateway through the encrypted channel established by the standardized communication unit. End-to-end auditing is achieved through event generation, unified collection, asynchronous transmission, correlation analysis, and tamper-proof storage. The audit processing service deduplicates events based on event identifiers and strings together project approval, software and computing power matching, data synchronization, instruction dispatch, sandbox creation, software execution, result feedback, and archiving events into a complete business chain based on task identifiers, thereby achieving traceable auditing of the entire simulation task process from generation to archiving.

[0048] For example, in a specific deployment instance of this application, the core management and control layer of the domestic IT innovation system uses a domestic server equipped with a Phytium S5000C ARM architecture processor and the operating system is the Galaxy Kylin Advanced Server Operating System. The unified business management and control module is developed based on a Django and Vue front-end and back-end separation architecture, providing a unified web operation portal to the outside world. The non-domestic computing power execution layer consists of two parts: a high-performance computing cluster and a visualization server. The high-performance computing cluster consists of seven x86 architecture computing nodes, one management node, and one storage node. The CPU of each computing node is an AMD x86 architecture processor, and the storage node is configured with a Lustre parallel storage system. The cluster operating system is CentOS, using the Slurm scheduling system. The visualization server consists of two x86 architecture graphics workstations equipped with NVIDIA professional graphics cards, operating on Windows Server and deploying RDS remote desktop services to support concurrent visualization access by multiple users. The core management and control layer of the domestic IT innovation system and the non-domestic computing power execution layer are connected through a fiber optic network and a national cryptographic encryption gateway and firewall are deployed to achieve one-way controllable access and encrypted communication. R&D terminals and management terminals access the unified web portal of the Phytium domestic server through the unit's intranet and do not directly access the non-domestic computing nodes. The distributed deployment of each module is as follows: Phytium domestic servers deploy the core ends of the domestic IT innovation adaptation base module, unified business management module, cross-architecture scheduling adaptation module, heterogeneous computing power management module, multi-software integration management module, distributed data security management module, and end-to-end security audit module; the management nodes of non-domestic high-performance computing clusters deploy the execution ends of the cross-architecture scheduling adaptation module, heterogeneous computing power management module, and distributed data security management module; each computing node deploys the execution end of the multi-software integration management module and the data isolation unit of the distributed data security management module; two visualization servers respectively deploy the execution ends of the cross-architecture scheduling adaptation module, heterogeneous computing power management module, multi-software integration management module, and distributed data security management module.

[0049] Example 2, Another embodiment of the present invention discloses a heterogeneous information technology innovation architecture simulation integrated computing management method based on the platform implemented in Embodiment 1, comprising the following steps: The unified business management module generates tasks, outputs computing power requirements to the heterogeneous computing power management module, and outputs software requirements to the multi-software integration management module. The heterogeneous computing power management module filters target nodes based on computing power requirements and real-time load, and outputs the node matching results to the unified business management module; The multi-software integration management module matches software based on software requirements and outputs the software matching results to the unified business control module; The unified business management module generates general task instructions based on task, node matching results and software matching results. The unified business management module then synchronously sends the general task instructions and target nodes to the cross-architecture scheduling adaptation module and the distributed data security management module. The cross-architecture scheduling adaptation module receives general task instructions and target nodes, translates the general task instructions into execution instructions that can be recognized by the scheduler corresponding to the target node, and sends the execution instructions to the non-domestic computing power execution layer. The distributed data security management module extracts task data to be processed from the core control layer of domestic IT innovation and sends it to the non-domestic computing power execution layer based on general task instructions and target node information. After receiving the execution instructions and the data to be processed, the non-domestic computing power execution layer starts the task execution; The distributed data security management module sends the execution results to the unified business control module, which triggers the result review process. Once the review is passed, the project is officially archived.

[0050] It should be noted that the unified business management module serves as the starting point of the process, simultaneously outputting computing power requirements and software requirements to the heterogeneous computing power management module and the multi-software integration management module, respectively. These two matching processes execute in parallel, thereby shortening task response time. The heterogeneous computing power management module selects the optimal target node based on the real-time load of each node, while the multi-software integration management module matches the corresponding target software and architectural components based on the software identifier. After the matching results are returned to the unified business management module, it assembles the task, target node, and software information to generate a general task instruction. This general task instruction uses a format independent of the specific scheduler type and is uniformly sent to the cross-architecture scheduling adaptation module and the distributed data security management module. These two modules process the task in parallel from two dimensions: instruction scheduling and data synchronization. The cross-architecture scheduling adaptation module is responsible for translating the general instruction into an execution instruction recognizable by the scheduler and sending it to the target node, while the distributed data security management module is responsible for pre-synchronizing the task's pending data to the target node. Task execution can only begin after both are ready. The non-domestic computing power execution layer only starts the task after the execution instruction arrives and the data is ready. During the execution process, the unified business management module tracks the task status in real time through the cross-architecture scheduling and adaptation module. After the execution is completed, the distributed data security management module sends the result back, and the unified business management module triggers the review and archiving, thereby realizing the closed-loop management of the entire process from task generation, resource matching, instruction translation, data synchronization, isolated execution to result feedback.

[0051] The method of this application will be explained below with reference to a specific industrial application scenario.

[0052] In a specific application example of this application, taking the simulation of heat treatment process for pressure vessels as a scenario, the platform administrator first creates user accounts and assigns role permissions in the unified business management module of the core management layer of the information technology innovation platform. The accounts of R&D engineers, simulation engineers, and simulation supervisors are synchronized to the non-domestic computing cluster and visualization server, and corresponding software usage permissions and computing resource quotas are configured. The platform administrator completes the composite image packaging and deployment of Abaqus and JMatPro software required for heat treatment process simulation in the multi-software integration management module, and configures the parameterized job submission template. R&D engineers access the platform portal of the Phytium domestic server through a web browser, complete identity authentication, and log in to the system. In the unified business management module, the R&D engineer initiates a heat treatment process simulation commission application for pressure vessels, selects the platform's preset heat treatment simulation commission form template, fills in basic project information, material grade, heat treatment process requirements, simulation objectives, and completion deadline, uploads the 3D geometric model and material performance test report, and submits for approval. The system pushes the commission form to the simulation supervisor for approval through a custom workflow engine. After the simulation supervisor approves, the system automatically creates the corresponding simulation project. The simulation supervisor decomposes the project into tasks, using a drag-and-drop workflow editing tool to break it down into two sequential task nodes: material thermodynamics calculation and heat treatment process simulation calculation. A corresponding simulation engineer is assigned to each node. After receiving the material thermodynamics calculation task, the material calculation engineer selects JMatPro software in the multi-software integration management module. The cross-architecture scheduling and adaptation module loads the software environment on demand based on the CPU architecture type of the target node. The engineer then performs material thermodynamic calculations using this software. Upon completion, the calculation result file is transmitted back to the unified storage pool of the core control layer of the information technology innovation system via the distributed data security management module and associated with the corresponding project task node. The system synchronously pushes the material calculation results to downstream task nodes. After receiving the heat treatment process simulation task, the simulation engineer selects and launches the Abaqus software in the multi-software integration management module. The cross-architecture scheduling adaptation module sends the software launch command to the non-domestic visualization server. The execution end allocates an independent remote desktop instance to the user through the RDS component and launches the Abaqus software. The engineer completes the pre-processing of the heat treatment case through the web-based remote desktop, then selects the computing node and number of cores to submit the Abaqus heat treatment process simulation task. The cross-architecture scheduling adaptation module translates the task instructions into execution instructions that the Slurm scheduler can recognize and sends them to the target computing node for execution. After completing the post-processing and analysis of the calculation results, the simulation engineer compiles a simulation analysis report and uploads it to the platform for result review. The simulation supervisor reviews and confirms online that the results meet the requirements before approving them. The client can view the report online.After the project is completed, it is archived. Users publish the geometric model, mesh model, material parameters, boundary conditions, calculation scripts and analysis reports of the pressure vessel heat treatment simulation to the computational knowledge base. The system triggers the case publication approval process. After approval, the case is included in the platform case library, realizing the accumulation and reuse of industrial simulation experience and data.

[0053] Those skilled in the art will understand that all or part of the processes implementing the methods of the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0054] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A heterogeneous information technology innovation architecture simulation integrated computing management platform, characterized in that, include: Unified business management module, cross-architecture scheduling and adaptation module, heterogeneous computing power management module, multi-software integration management module, and distributed data security management module; The unified business management module is used to acquire user tasks, parse user tasks to obtain computing power requirements and software requirements, output computing power requirements to the heterogeneous computing power management module, output task software requirements to the multi-software integration management module, receive node matching results output by the heterogeneous computing power management module and software matching results output by the multi-software integration management module, generate general task instructions and target nodes and send them synchronously to the cross-architecture scheduling adaptation module and the distributed data security management module. The heterogeneous computing power management module is used to filter target nodes based on computing power requirements and real-time load, and output the node matching results to the unified business management module; The cross-architecture scheduling adaptation module is used to receive general task instructions and target nodes, translate the general task instructions into execution instructions that can be recognized by the scheduler corresponding to the target node, and send the execution instructions to the non-domestic computing power execution layer. The multi-software integration management module is used to receive software requirements matching software and return the software matching result to the unified business control module; The distributed data security management module is used to extract task data to be processed from the core control layer of the domestic IT innovation system and send it to the non-domestic computing power execution layer based on general task instructions and target node information.

2. The platform according to claim 1, characterized in that, The cross-architecture scheduling adaptation module includes a scheduler adaptation unit, which includes a core terminal. The core terminal is used to convert general task instructions into execution instructions that can be recognized by the non-domestic computing power execution layer. The core of the scheduler adaptation unit receives the general task instruction issued by the unified business management module and executes the following translation process: S1: Parse the general task instructions and extract task metadata; S2: Parse the task metadata into a generic intermediate instruction model that is independent of the scheduler type; S3: Based on the scheduler type of the target node, query the built-in standardized adaptation rule library to obtain the corresponding general-native instruction mapping rules; S4: According to the general-native instruction mapping rules, the general intermediate instruction model is converted into the native execution script of the scheduler corresponding to the target node, and used as the execution instruction.

3. The platform according to claim 2, characterized in that, The adaptation rule base stores the mapping relationship between different scheduler types and the parameters of the intermediate instruction model; the core of the scheduler adaptation unit calls the corresponding mapping relationship from the adaptation rule base according to the scheduler type actually used by the target node, maps the parameters in the intermediate instruction model to job submission parameters that the scheduler can recognize, and combines them to generate the execution instruction.

4. The platform according to claim 2, characterized in that, The scheduler adaptation unit also includes an execution end; the core end is deployed in the domestic IT innovation core management and control layer, and the execution end is deployed in the non-domestic computing power execution layer; the core end is also used to send the execution instruction carrying the unique identifier of the task to the execution end; after receiving the native execution instruction, the execution end of the scheduler adaptation unit first queries the local execution record according to the unique identifier of the task. If the instruction corresponding to the identifier has been executed, it directly returns the recorded result without resubmitting. If it has not been executed, it performs syntax verification on the execution instruction and submits it to the local scheduler. After successful submission, the execution end of the scheduler adaptation unit obtains the job number returned by the local scheduler and binds it with the unique identifier of the task before sending it back to the core end of the scheduler adaptation unit. At the same time, it monitors the job status in real time and converts it to a standardized status before sending it back.

5. The platform according to claim 1, characterized in that, The multi-software integration management module includes a core terminal deployed in the core control layer of domestic IT innovation and an execution terminal deployed in the non-domestic computing power execution layer; The core of the multi-software integration management module extracts the main program and general dependency library of each simulation software into a general kernel layer that is shared across architectures. It then splits the runtime dependency components required by the software under different CPU architectures into differentiated peripheral layers corresponding to different architectures. Finally, it associates and packages the general kernel layer with each differentiated peripheral layer into a composite image and stores it. After receiving the software requirements output by the unified business control module, the core of the multi-software integration management module matches the corresponding target software according to the software identifier in the software requirements, generates a software matching result, and sends the software matching result to the execution ends of the unified business control module and the multi-software integration management module respectively; the software matching result includes the composite image storage address of the target software and the correspondence between its supported CPU architecture type and differentiated peripheral layer; The execution end of the multi-software integration management module loads the general kernel layer from the composite image storage address according to the CPU architecture type of the target node issued by the cross-architecture scheduling and adaptation module, and only loads the differentiated peripheral layer corresponding to the CPU architecture type, and combines them to construct the local executable software runtime environment of the target node and start the target software.

6. The platform according to claim 1, characterized in that, The heterogeneous computing power management module includes an intelligent scheduling unit deployed in the core control layer of domestic IT innovation and a monitoring agent deployed in each node of the non-domestic computing power execution layer; The monitoring agent collects the load indicators of the node it is located at a preset period and reports them to the intelligent scheduling unit. The intelligent scheduling unit receives the load indicators of each node and the computing power requirements output by the unified business management module. Based on the resource conditions in the computing power requirements, it selects candidate nodes from each node, sorts them by comprehensive score according to the current load of each candidate node, determines the node with the best score as the target node, and returns the node matching result to the unified business management module.

7. The platform according to claim 6, characterized in that, The overall score ranking is performed according to the following formula: S i =ω c ·C match +oh m ·M match +oh g ·G match -oh l ·L i in, S i Assign a comprehensive fit score to the i-th candidate node. ω c , ω m , ω g , ω l These are the weighting coefficients for CPU, memory, GPU, and node load, respectively. C match , M match , G match These represent the degree to which the node's CPU, memory, and GPU resources meet the task requirements. L i This represents the current overall load rate of the node.

8. The platform according to claim 1, characterized in that, The distributed data security management module includes a data synchronization unit deployed in the core control layer of the domestic IT innovation system and a data synchronization agent deployed in the execution layer of non-domestic computing power. Before the cross-architecture scheduling and adaptation module issues the execution instruction, the data synchronization unit sends the task data to be processed to the data synchronization agent deployed in the non-domestic computing power execution layer according to the data list in the general task instruction. The data synchronization agent then sends the data to be processed to the target node. After the task is completed, the data synchronization agent retrieves the calculation result file from the target node and sends it back to the data synchronization unit.

9. The platform according to claim 8, characterized in that, The platform also includes a data isolation unit, which is deployed on the non-domestic computing power execution layer. Before the target node executes the task, the data isolation unit creates an independent isolated running environment for the task through operating system-level virtualization technology, restricts the process space, file access path and network communication of the task to the isolated running environment, and binds the task to the specified CPU core and GPU device through cgroups technology.

10. A heterogeneous information technology innovation architecture simulation integrated computing management method, characterized in that, Includes the following steps: The unified business management module generates tasks, outputs computing power requirements to the heterogeneous computing power management module, and outputs software requirements to the multi-software integration management module. The heterogeneous computing power management module filters target nodes based on computing power requirements and real-time load, and outputs the node matching results to the unified business management module; The multi-software integration management module matches software based on software requirements and outputs the software matching results to the unified business control module; The unified business management module generates general task instructions based on task, node matching results and software matching results. The unified business management module then synchronously sends the general task instructions and target nodes to the cross-architecture scheduling adaptation module and the distributed data security management module. The cross-architecture scheduling adaptation module receives general task instructions and target nodes, translates the general task instructions into execution instructions that can be recognized by the scheduler corresponding to the target node, and sends the execution instructions to the non-domestic computing power execution layer. The distributed data security management module extracts task data to be processed from the core control layer of domestic IT innovation and sends it to the non-domestic computing power execution layer based on general task instructions and target node information. After receiving the execution instructions and the data to be processed, the non-domestic computing power execution layer starts the task execution; The distributed data security management module sends the execution results to the unified business control module, which triggers the result review process. Once the review is passed, the project is officially archived.

Images