Space hyper-intelligence fusion computing power system

Abstract

The invention relates to the field of intelligent computing, and discloses a space hyper-intelligence fusion computing power system, which comprises a computing subsystem used for determining computing resources, receiving target jobs and processing the target jobs based on the computing resources if the computing resources meet the computing requirements of the target jobs; the storage subsystem is used for determining read-write characteristics and a corresponding read-write path if the target job is executed, and storing data corresponding to the target job based on the read-write path; the communication subsystem is used for establishing a first communication channel between the subsystems and a second communication channel between the subsystems and an external system; the global management subsystem is used for monitoring the running state of each subsystem, if the running state of at least one subsystem is a fault state, determining each task instruction corresponding to the fault state based on a rule engine, and distributing the task instructions to the corresponding intelligent agents, the effect of integrating powerful computing power, efficient storage, stable communication, reliable energy supply and heat dissipation and a solid hardware foundation is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent computing, and particularly to a space super-intelligent fusion computing power system. Background Art

[0002] In astronomy and cosmology research, it is necessary to process a vast amount of celestial data to explore the mysteries of the universe, such as analyzing galaxy evolution, dark matter distribution, etc.; Earth observation and environmental monitoring rely on powerful computing capabilities to perform real-time analysis of data such as satellite images in order to promptly grasp natural disasters, climate change, etc.; fields such as space science experiments and simulations, deep space exploration science, etc. also pose extremely high requirements for computing performance.

[0003] However, existing space computing power system technologies have many computing bottlenecks in terms of computing power, storage, communication, etc., and face profound technical contradictions between high performance and high reliability, heat dissipation and power consumption, communication latency and autonomous intelligence, and it is difficult to perform super-intelligent computing fully and effectively in the space environment.

[0004] In view of this, the present invention is specifically proposed. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a space super-intelligent fusion computing power system, achieving the effect of integrating powerful computing capabilities, efficient storage, stable communication, reliable energy supply and heat dissipation, and a solid hardware foundation.

[0006] An embodiment of the present invention provides a space super-intelligent fusion computing power system, which includes:

[0007] A computing subsystem, configured to determine computing resources within the currently deployed space, receive a target job, and process the target job based on the computing resources when the computing resources meet the computing requirements of the target job;

[0008] A storage subsystem, configured to determine the read-write characteristics corresponding to the target job when the target job is executed, and determine a read-write path according to the read-write characteristics, and store the data corresponding to the target job based on the read-write path; wherein, the read-write characteristics are determined according to the read-write throughput, read-write burstiness, and source data operation volume of the target job;

[0009] A communication subsystem, configured to establish a first communication channel between each subsystem and a second communication channel between the space super-intelligent fusion computing power system and an external system;

[0010] A global management subsystem, configured to monitor the operating states of each subsystem, and when there is at least one subsystem in a fault state, determine each task instruction corresponding to the fault state based on a rule engine, and execute each task instruction according to each intelligent agent corresponding to each task instruction to change the operating state to a normal state.

[0011] The embodiments of the present invention have the following technical effects: The computing subsystem determines the computing resources in the current deployment space, receives a target job, and processes the target job based on the computing resources when the computing resources meet the computing requirements of the target job. When the target job is executed, the storage subsystem determines the read-write characteristics corresponding to the target job, determines the read-write path according to the read-write characteristics, and stores the data corresponding to the target job based on the read-write path. The communication subsystem establishes a first communication channel between each subsystem and a second communication channel between the space super-intelligent fusion computing power system and an external system. The global management subsystem monitors the operating states of each subsystem, and when the operating state of at least one subsystem is a fault state, based on the rule engine, determines each task instruction corresponding to the fault state, and executes each task instruction according to each intelligent agent corresponding to each task instruction to change the operating state to a normal state, achieving the effect of integrating powerful computing capabilities, efficient storage, stable communication, reliable energy supply and heat dissipation, and a solid hardware foundation in space, effectively improving the performance of space intelligent computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0013] Figure 1 It is a schematic structural diagram of a space super-intelligent fusion computing power system provided by an embodiment of the present invention;

[0014] Figure 2 It is a schematic diagram of the operation mode of a space super-intelligent fusion computing power system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0016] The space super-intelligent integrated computing power system provided by the embodiments of the present invention is mainly applicable to the situation of fully performing calculations, storage, communication, and other processing on target operations in a space environment. The space super-intelligent integrated computing power system provided by the embodiments of the present invention can be deployed in space stations or satellites of different structural types.

[0017] Figure 1 It is a schematic structural diagram of a space super-intelligent integrated computing power system provided by the embodiments of the present invention. Refer to Figure 1 , the space super-intelligent integrated computing power system specifically includes: a computing subsystem 110, a storage subsystem 120, a communication subsystem 130, and a global management subsystem 140.

[0018] The computing subsystem 110 is used to determine the computing resources in the current deployment space, receive the target operation, and process the target operation based on the computing resources when the computing resources meet the computing requirements of the target operation.

[0019] Among them, the current deployment space is the space where the space super-intelligent integrated computing power system is deployed, which can be the cabin of a space station or a satellite. The computing resources are the resources that the space super-intelligent integrated computing power system can mobilize and use for computing. The target operation is the operation that requires the space super-intelligent integrated computing power system to perform super-intelligent computing processing. The computing requirements are the various resources required to complete the processing of the target operation.

[0020] Specifically, the computing subsystem 110 can determine the computing resources in the current deployment space in real time and can receive the target operation that needs to be processed in space. After receiving the target operation, it is necessary to determine whether the computing resources in the current deployment space meet the computing requirements of the target operation. If they meet, the target operation can be executed in the current deployment space, that is, the target operation can be processed based on the computing resources in the current deployment space. If they do not meet, it is necessary to cooperate with the ground scheduling server for re-scheduling and execution of the target operation.

[0021] The computing subsystem 110 provides powerful computing capabilities and is a core functional component of the space ultra-intelligent integrated computing power system. In the ground-based and integral construction methods, the computing subsystem 110 consists of multiple computing modules. Each computing module contains network server blades, computing server blades, monitoring blades, and a power supply backplane. The network server blades are generally located in the middle of the module, and the monitoring blades are at the bottom. Among them, the network server blades are used to interact the data of the computing boards within the module to the core switch component, and then interact with other computing modules or components; the computing server blades are set in a heterogeneous form within the board and are composed of high-performance computing chips such as CPUs (Central Processing Units) and GPUs (Graphics Processing Units). Liquid cooling covers core components such as CPU, GPU chips, and memory to provide ultra-intelligent integrated computing capabilities. Each independent blade server has a traditional monitoring architecture that can detect internal and external performance indicators, which are uniformly collected by the monitoring blades within the module at regular intervals. At the same time, each node is equipped with high-capacity HBM (High Bandwidth Memory) and CXL SSD (Compute Express Link Solid-State Drive), but the accessories are optional, and an optical communication network based on RDMA (Remote Direct Memory Access) is adopted.

[0022] In the management of nodes, to ensure the high availability and reliability of the cluster, different computing queues can be set up to build a computing resource pool according to the different characteristics of space computing applications. A resource scheduling method with job characteristic awareness is adopted. According to the job characteristics, appropriate resources are allocated. For example, the computing resources are divided into different computing queues such as a computing area, a training area, and an inference area.

[0023] The storage subsystem 120 is used to determine the read-write characteristics corresponding to the target job when the target job is executed, and determine the read-write path according to the read-write characteristics, and store the data corresponding to the target job based on the read-write path.

[0024] Among them, the read-write characteristics are determined according to the read-write throughput, read-write burstiness, and source data operation volume of the target job. The read-write characteristics are used to characterize the data stability, that is, the characteristics of whether it needs to be frequently called and updated. The read-write path is the path for reading and writing the data corresponding to the target job, which can be understood as different storage hardware and the read-write process.

[0025] Specifically, when the storage subsystem 120 executes a target job, it can first determine the read-write characteristics of the target job by combining the read-write throughput, read-write burstiness, and source data operation volume of the target job, and the read-write characteristics can be obtained through calculation, model processing, etc. According to the read-write characteristics, in the pre-constructed correspondence between the read-write characteristics and the read-write paths, the read-write characteristics corresponding to the target job are determined. Furthermore, based on the read-write path, the data corresponding to the target job, such as initial data, data generated during the execution process, etc., is stored.

[0026] The storage subsystem 120 includes an accelerated storage layer, a persistent storage layer, and a cold data storage layer; the number of storage blade servers in the persistent storage layer is less than the number of storage blade servers in the accelerated storage layer;

[0027] Among them, the read-write path includes a first path and a second path; the accelerated storage layer and the persistent storage layer together serve as the first path, and the persistent storage layer serves as the second path; when the read-write path is the first path, a parallel file system is constructed based on the accelerated storage layer, data is stored based on the parallel file system, and after the target job is completed, the data in the parallel file system is copied to the persistent storage layer, and the parallel file system is destroyed;

[0028] The cold data storage layer is used to store the packaged data corresponding to the generated data in the persistent storage layer when the stable time of the stored data in the persistent storage layer reaches the first time threshold.

[0029] Among them, the accelerated storage layer is a storage structure for fast reading and writing and can be used to store hot data. The persistent storage layer is a storage structure for reading and writing data and having a large storage space and can be used to store warm data. The cold data storage layer is a storage structure for storing data that is not frequently called and is used to store cold data. The stable time is the time obtained by timing since the last read or write of the data in the persistent storage layer, that is, the time when the data has not been read or written continuously. The first time threshold is a time value used to determine whether the data needs to be transferred and stored in the cold data storage layer.

[0030] Specifically, when the read / write path is the first path, it is necessary to improve the data read / write efficiency through the accelerated storage layer. Thus, a parallel file system can be constructed based on the accelerated storage layer, data can be stored based on the parallel file system, and after the target job is completed, the use of the accelerated storage layer is stopped, that is, the data in the parallel file system is copied to the persistent storage layer, and the parallel file system is destroyed. When the read / write path is the second path, data is directly stored through the persistent storage layer. For the stored data in the persistent storage layer, it is not re-timed after each read / write to obtain a new stable time. When the stable time reaches the first time threshold, the data is determined to be cold data, and the corresponding packaged data is generated and sent to the cold data storage layer for storage.

[0031] The storage subsystem 120 further includes: a software image library.

[0032] The software image library is used to store the software images corresponding to each application program so that when the application program cannot run, the corresponding software image can be pulled from the software image library; when the update completion duration reaches the preset update period, the software image library obtains a new software image from the ground data server through the communication subsystem 130 to update the software image library.

[0033] Among them, the update completion duration is the duration after the software image library completed the last update and started timing. The preset update period is the period used to trigger the update of the software image library. The ground data server is a server set on the ground for storing various software images and can be transmitted through the communication subsystem 130.

[0034] Exemplarily, an accelerated storage layer including a computing node HBM, a persistent storage layer with multiple backups of the computing node local CXL SSD, and a cold data storage layer based on optical discs are constructed on the hardware of the storage subsystem 120 to form a multi-level storage structure. Inside the computing node, HBM is used to accelerate the data processing efficiency, and the local high-speed CXL SSD is mounted. The storage subsystem 120 sets a separate storage module and installs it in the storage blade server. The storage server blade can be loaded with a large-capacity NVME SSD (Non-Volatile Memory Express Solid-State Drive).

[0035] Each space module loads the same number of storage blade servers. Metadata management and configuration information management services are deployed on each storage blade server. The persistent storage capacity owned by all space modules is evenly divided into multiple storage resource pools (N) with equal storage capacity. Then, 1 of the storage resource pools is selected as the accelerated storage layer, 1 as the cold data storage layer, and the remaining N - 2 as the persistent storage layer to build a multi-copy distributed file system and mount it as needed. The metadata management and configuration information management services of all space modules share metadata and configuration information.

[0036] At the software level, the accelerated storage layer can build an accelerated layer that is generated and destroyed according to the target job cycle. The allocation and management of multi-level storage media are responsible for by a unified storage media controller. When the target job is ready to execute, the storage media controller adjusts the read and write paths according to the read and write characteristics of the target job, selects a resource with a suitable capacity from the resource pool for the target job. If the accelerated storage layer is used, a parallel file system is built, and the original data is copied into the parallel file system. When the target job ends, the data is copied to the persistent storage layer and the file system is destroyed.

[0037] Based on the optical disc cold storage layer, taking the job as the unit, according to the last access time and data volume of all output files (data) of the target job, dynamically set the data migration rules. For example: set that after a certain time t (the stable time reaches the first time threshold), the output data of the target job is packed and compressed, marked as cold data, and copied from the persistent storage layer to the cold data storage layer for storage. If the data of the target job is read or written again within the time t (the first time threshold), the timing is restarted and the set data migration rules are executed.

[0038] Since it is difficult to update software in space, the storage subsystem 120 sets up a software image library. When the application program cannot run due to the lack of system libraries, the scheduler automatically pulls from the software image library. However, when restarted, the kernel returns to its original state. The software image library needs to be updated regularly by interacting with the ground (ground data server).

[0039] The communication subsystem 130 is used to establish the first communication channel between each subsystem and the second communication channel between the space super-intelligent fusion computing power system and the external system.

[0040] Among them, the first communication channel is the channel for internal communication of the space super-intelligent fusion computing power system. The second communication channel is the channel for external communication of the space super-intelligent fusion computing power system. The external system can be the space super-intelligent fusion computing power system in other space stations or satellites in space, or it can be a ground system.

[0041] Based on the above example, the second communication channel includes an inter-satellite communication channel and a satellite-ground communication channel. The communication subsystem 130 is further configured to:

[0042] Transmit the data corresponding to the target job based on an encrypted time-sensitive network.

[0043] Among them, the encrypted time-sensitive network extends the encrypted information in the network identification number field of the data frame at the Ethernet link layer, and adds a security label field and an integrity check value field to the data frame. The inter-satellite communication channel is a channel for communicating with the space ultra-intelligent fusion computing power system in other space stations or satellites in space. The satellite-ground communication channel is a channel for communicating with the ground system.

[0044] Specifically, when using the communication subsystem 130 for data transmission, an encrypted time-sensitive network is used for transmission to ensure the timeliness and security of inter-satellite communication and satellite-ground communication.

[0045] The communication subsystem 130 actually includes two parts. One part is the external communication interface, which is responsible for inter-satellite communication and satellite-ground communication. The other part is the internal communication interface, which is mainly responsible for the internal low-latency high-speed communication of the large-scale intra-satellite ultra-intelligent fusion computing power system. The communication subsystem 130 is closely connected to the computing subsystem 110 and the storage subsystem 120, and collaborates with the global management subsystem 140 to ensure the stable and reliable communication of the system.

[0046] For inter-satellite communication and satellite-ground communication, an external communication module is set up, which includes a radio frequency communication module and a laser communication module, and performs adaptive matching according to performance and function requirements to achieve data upload, download, and command interaction. For intra-satellite communication, low-latency communication is adopted between chips, high-speed and reliable interconnection is adopted between nodes, and high-speed network interconnection technology is adopted between units to ensure the efficiency of data transmission. The cluster network structure adopts the Spine-Leaf (leaf-spine network) structure. For the homogeneous construction method, each space module will be equipped with two customized switches. All devices are first connected to the Leaf (leaf) customized switch, then the Leaf switch is interconnected with the Spine (spine) switch in the module, and finally the Spine switches between modules are fully interconnected, so as to achieve a modular effect. When a module needs to be replaced, first logically isolate all the links related to the module to be replaced, and after the load is completed, perform a physical break, and finally replace the module.

[0047] In terms of communication protocols, CTSN (Cybersecurity Time Sensitive Network or Encrypted Time Sensitive Network, encrypted time-sensitive network) is proposed to ensure the timeliness and security of inter-satellite communication and satellite-ground communication. Through means such as data encryption, communication protocols, and access control, communication security and efficiency are comprehensively guaranteed to prevent data leakage and illegal access. The CTSN network is an extension of the Ethernet link layer (Layer 2) protocol. By expanding the PCP (Priority Code Point) / DEI (Drop Eligible Indicator) / VID (VLAN Identifier) bits in the extended VLAN Tag (Virtual Local Area Network Tag, network identification number) field of the Layer 2 packet, and then adding SecTAG (Security Tag) and ICV (Integrity Check Value) fields, the frame content is encrypted.

[0048] Exemplarily, the original Layer 2 packet format:

[0049] │Dest MAC│Src MAC│Type / Length│VLAN Tag│Payload│FCS│;

[0050] CTSN Layer 2 packet format:

[0051] │Dest MAC│Src MAC│Type / Length│SecTAG│VLAN+Payload (encrypted)│ICV│FCS│;

[0052] Among them, Dest MAC (Destination MAC Address), Src MAC (Source MAC Address), Type / Length (type / length field), FCS (Frame Check Sequence).

[0053] The computing subsystem 110 is also used to: in the case where the computing resources do not meet the computing requirements of the target job, send the computing resources and the computing requirements to the communication subsystem;

[0054] The communication subsystem 130 is further configured to: receive computing resources and computing requirements, and send the computing resources and computing requirements to the ground scheduling server, so that the ground scheduling server determines a space collaboration system according to the computing resources and computing requirements; receive the space collaboration system corresponding to the computing requirements fed back by the ground scheduling server, and establish a second communication channel with the space collaboration system, so as to collaboratively process the target job based on the second communication channel, the space super-intelligent fusion computing power system in the current deployment space, and the space collaboration systems in other deployment spaces.

[0055] The space collaboration system is a space super-intelligent fusion computing power system in other deployment spaces, and is used to collaborate with the space super-intelligent fusion computing power system in the current deployment space to complete the target job. The ground scheduling server is a scheduling server on the ground that is used to execute the target job for the space super-intelligent fusion computing power system in the current deployment space and allocate the collaborating space collaboration system.

[0056] Specifically, if it is determined by the computing subsystem 110 that the computing resources within the current system do not meet the computing requirements of the target job, it indicates that the systems in other deployment spaces need to collaborate to process the target job. Therefore, the computing resources and computing requirements are sent to the communication subsystem 130. The communication subsystem 130 receives the computing resources and computing requirements sent by the computing subsystem 110, and sends the computing resources and computing requirements to the ground scheduling server, so that the ground scheduling server determines whether the space super-intelligent fusion computing power systems in other deployment spaces can collaborate to complete the target task according to the computing resources and computing requirements, obtains the space collaboration systems that can collaborate, and feeds them back to the communication subsystem 130 of the space super-intelligent fusion computing power systems in the current deployment space. It can receive the space collaboration system corresponding to the computing requirements fed back by the ground scheduling server, and can establish a second communication channel with the space collaboration system to communicate based on the second communication channel, and collaboratively process the target job based on the space super-intelligent fusion computing power system in the current deployment space and the space collaboration systems in other deployment spaces, so as to achieve the purpose of multi-system collaboration.

[0057] Exemplarily, a unified scheduling server (ground scheduling server) is set up on the ground. Each deployment space is used as a unit resource pool and is added to the total resource pool managed by the unified resource scheduling server. Each unit resource pool can independently provide computing services. When there is a new target job ready to run in any unit resource pool, the unit resource pool first determines whether this job requires the computing services of other resource pools. If so, it submits to the unified resource scheduler, which determines the specific resource pool location and prepares the data. If the communication is abnormal or no other resource pools are required, its own computing resources are allocated for computing.

[0058] The global management subsystem 140 is used to monitor the operating status of each subsystem. When there is at least one subsystem in a fault state, based on the rule engine, it determines each task instruction corresponding to the fault state, and according to each agent corresponding to each task instruction, executes each task instruction to change the operating state to the normal state.

[0059] Among them, each subsystem includes a computing subsystem 110, a storage subsystem 120, and a communication subsystem 130. The operating state is used to describe whether the subsystem can work normally, including the normal state and the fault state. The rule engine receives data input, interprets rules, and outputs decision results, and is often used to process complex and frequently changing logics. The task instruction is an instruction to solve the current fault state. Each agent is an agent with embodied intelligent functions for processing different task instructions, and different agents have different functions. For example, it can include a system perception agent, a strategy planning agent, a command operation execution agent, and a hardware replacement track robot, etc.

[0060] Specifically, the global management subsystem 140 is used to monitor the operating status of the computing subsystem 110, the storage subsystem 120, and the communication subsystem 130. If there is at least one subsystem in a fault state, then it is necessary to analyze the cause of the fault based on the rule engine and determine the task instruction to solve the fault as each task instruction corresponding to the fault state. Furthermore, it judges the agent for executing each task instruction as the corresponding agent, and through each agent, executes each task instruction to automatically solve the fault and change the operating state from the fault state to the normal state.

[0061] The global management subsystem 140 is also used for:

[0062] Regularly collect monitoring data from each subsystem, and based on the rule engine, analyze the monitoring data to determine the target processing instruction;

[0063] Based on the commander agent, receive the target processing instruction pushed by the rule engine, analyze and disassemble the target processing instruction to obtain the task instruction, and send each task instruction to the corresponding functional agent respectively, so that each functional agent executes the corresponding task instruction.

[0064] Among them, the monitoring data is data collected from each subsystem for evaluating the operating state. The target processing instruction is a processing instruction for solving the fault existing in the monitoring data. The commander agent is an agent for overall planning, and the functional agent is an agent that completes different functions according to the requirements of the commander agent.

[0065] Specifically, the global management subsystem 140 can regularly collect monitoring data from each subsystem. When the operating state of at least one subsystem is in a fault state, based on the rule engine, it analyzes the monitoring data to analyze and solve the target processing instructions for the corresponding fault state. Based on the commander agent, it receives the target processing instructions pushed by the rule engine, analyzes and disassembles the target processing instructions, so as to disassemble the target processing instructions into task instructions that different functional agents can execute. Furthermore, it sends each task instruction to the corresponding functional agent respectively, so that each functional agent executes the corresponding task instruction to eliminate the fault.

[0066] Optionally, when determining the target processing instructions and disassembling the target processing instructions to obtain task instructions, a large language model and a knowledge retrieval enhancement strategy can also be used for determination.

[0067] The global management subsystem 140 is responsible for monitoring and scheduling, and is the core hub to ensure the stable and efficient operation of the space super-intelligent fusion computing power system. A system intelligent platform is deployed in the space super-intelligent fusion computing power system to monitor the status of each component (monitoring data) in real time, and use intelligent diagnosis technology to detect and solve faults in time to ensure the reliability of the system. This system intelligent platform can regularly collect data from the monitoring blades of each computing module of each computing subsystem 110, and deploy probes at each node of the modules of other subsystems to obtain data regularly from the probes.

[0068] The global management subsystem 140 can adopt a multi-agent system with embodied intelligence functions for intelligent system management. Set multiple agents, and set a commander agent in the system intelligent platform to receive the target processing instructions pushed by the system intelligent platform through the rule engine. The commander agent analyzes and disassembles the instructions to obtain each task instruction, and distributes each task instruction to the system perception intelligent agent, the strategy planning intelligent agent, the command operation execution intelligent agent, and the hardware replacement orbital robot, etc. according to the task requirements, so as to ensure the normal operation of the system.

[0069] Build a network security management system through software-defined means. Set up a hierarchical intelligent control system, and use system intelligent algorithms to realize real-time monitoring, fault prediction and automatic fault handling of the status of each component of the system to ensure the stable and efficient operation of the system. By setting up on-orbit robots, after receiving event notifications, they perform knowledge retrieval enhancement strategy reasoning based on a large language model to find the most suitable processing plan, automatically handle online software and hardware faults, avoid manual operation and maintenance, and at the same time reduce the dependence on high-price radiation-hardened and other hardware space reliability methods.

[0070] To enable the space super-intelligent fusion computing power system to adapt to being deployed in different structural types of space stations or satellites, three different composition methods of the system are set, which can be respectively called: integral type, ground-based type and uniform type.

[0071] The integrated construction method does not use space module segments in a space station or satellite. Instead, the servers are assembled into a whole using a ground-customized rack and then placed as a whole in a space facility (deployment space). This method has a small cost for transforming the ground computing power system and relatively low costs. It is suitable for being placed in a module-type space station with a good environment and sufficient space. However, due to being placed as a whole, it is not suitable for satellites with deployable structures, etc.

[0072] Generally, satellites with deployable structures have lower launch vehicle costs than space stations and are a better choice for low-cost launch vehicles of large-scale computing power systems. However, for a space ultra-intelligent fusion computing power system, the integrated construction method cannot be used for satellites with deployable structures due to its integrity. The ground-based construction method is relatively more suitable and has a moderate transformation cost. The ground-based construction method involves functionally partitioning the roles of each space module, placing devices with different functions in different space modules. Before launch, the devices are pre-installed on available installation planes (for convenience of description, space modules will be used instead later), and then folded and placed in the main structure cabin of the satellite. After entering orbit and completing attitude capture, the satellite unfolds to the working state.

[0073] Suppose there are M space modules, and each space module is used to place the component parts of each subsystem of the space ultra-intelligent fusion computing power system. According to the preset computing power demand C, the computing power (computing volume) C of a single computing blade server s and the number Q of computing blade servers that can be placed in a single space module, calculate the number N of modules required for the computing subsystem 110 c : .

[0074] According to the preset storage demand S (including the capacity provided by all types of storage servers such as SSD servers and HDD servers), the storage capacity list L of storage blade servers (the capacity that different servers can provide varies according to different storage media) L = [s1, s2... s n and the number R of storage blade servers that can be placed in a single space module, calculate the number N of space modules required for the storage subsystem 120 p : . Among them, the total number of servers with different storage media can be determined.

[0075] According to the preset computing power demand C, the computing power C of a single computing blade server s , the preset storage demand S, the storage capacity list L, and the number T of network blade servers that can be placed in a single space module, determine the number N of space modules required for the network core switch (communication subsystem 130) s : . Among them, g(a, b) represents the number of required network core switches calculated according to the network structure and the number of ports, where a represents the number of computing servers and b represents the number of storage servers.

[0076] To ensure that the space super-intelligent fusion computing power system has sufficient robustness as much as possible, N m (N m ≥2) space modules are set as global management modules to place the hardware devices required by the global management subsystem. The number of servers that can be placed in each of the above-mentioned single space modules is based on the basis that each space module also needs to be equipped with a switching blade server, which is used for communication and interconnection with the core switch, and the network structure is a fat tree. The number of space modules required by each subsystem also needs to meet the following restrictions: .

[0077] Based on the above example, taking the ground-based type as an example, there are system racks in the currently deployed space, and the system racks evenly distribute multiple space modules, and each subsystem is placed on at least one space module; the first number of space modules corresponding to the computing subsystem 110 is determined according to the preset computing power demand, the computing volume corresponding to a single computing blade server, and the maximum number of computing blade servers corresponding to each space module; the second number of space modules corresponding to the storage subsystem 120 is determined according to the preset storage demand, the storage capacity of the storage blade servers corresponding to each storage medium, and the maximum number of storage blade servers corresponding to each space module; the third number of space modules corresponding to the communication subsystem 130 is determined according to the preset computing power demand, the computing volume corresponding to a single computing blade server, the preset storage demand, the storage capacity of the storage blade servers corresponding to each storage medium, and the maximum number of network blade servers corresponding to each space module; the fourth number of space modules corresponding to the global management subsystem 140 is at least two; the sum of the first number, the second number, the third number, and the fourth number is less than or equal to the total number of space modules on the system rack.

[0078] Among them, the first number is the in the above example, the second number is the in the above example, the third number is the in the above example, and the fourth number is the .

[0079] Based on the above example, the subsystems placed in each space module in the space super-intelligent fusion computing power system can be reasonably planned. Specifically, it can be:

[0080] The space module in the middle area of the system rack is used to place the communication subsystem 130, the space module in the outermost area of the system rack is used to place the global management subsystem 140, the space module in the second outermost area of the system rack is used to place the storage subsystem 120, and the space module in other areas of the system rack is used to place the computing subsystem 110.

[0081] To achieve high performance in a small space, a high-density assembly method is adopted overall. All servers involved in the modules are installed in the form of server blades, and each module provides a backplane, which provides communication and power access.

[0082] For the homogenized construction method, each space module has exactly the same computing power, storage capacity, and network capacity. It is necessary to deploy these three types of blade servers in one space module at the same time. Since it is difficult to maintain and update equipment in space, to improve maintainability, a modular design is usually adopted, which requires that the machine groups in each space module need to be structurally independent and the interfaces be standardized, that is, each machine group needs to include equivalent computing units, storage units, network units, management units, etc. In this way, if a certain module can no longer provide services, it can be replaced with a new module. If there are M space modules, such as the preset computing power demand C, preset storage demand S, and network interaction ability N of the entire space super-intelligent fusion computing power system, then each module should have a computing power C s = C / M, storage capacity S s = S / M, network interaction ability N s = N / M.

[0083] When using the integral and ground-based construction, the global management node is placed in the global management module. However, for the normalization method, in each space module, the global management module hardware is deployed, and the resource scheduling service is deployed in the global management module. Each service manages the computing resources in all modules, and the status information required by the service is guaranteed based on the RAFT protocol.

[0084] Specifically, there are three optional hardware composition modes: one is to directly use the entire space super-intelligent integrated computing power system as a module, which can be deployed in a module-based space station; another is to build different functional space modules on the available installation planes according to the operating requirements of the space super-intelligent integrated computing power system, and then combine them into a complete machine system, which can be deployed in a module-based space station or in a satellite with an expandable structure; finally, build a unique space module and install it on all available installation planes. All space modules have the same function and are suitable for deployment in satellites with expandable structures such as triangular prisms and stacked configurations. Logically, various subsystem modules are classified by function, and a system is constructed including hardware subsystems such as computing subsystem 110, storage subsystem 120, communication subsystem 130, global management subsystem 140, energy and thermal control subsystem, and support infrastructure, and software subsystems such as application software subsystem, system software subsystem, and global management subsystem 140. Each subsystem cooperates with each other and is closely related, jointly providing guarantee for the stable operation and high efficiency of the space computing power system, filling the gap in the current method of building a centralized large-scale space super-intelligent integrated computing power system.

[0085] Based on the above example, the space super-intelligent integrated computing power system also includes: an energy and thermal control management subsystem; the energy and thermal control management subsystem is connected to the energy system and the refrigeration system of the currently deployed space; the energy and thermal control management subsystem is used to connect the inlet and outlet of the liquid cooling plate on each blade server in each subsystem to the refrigeration system of the currently deployed space; the energy and thermal control management subsystem also includes a peripheral air cooling device for thermal management of each subsystem.

[0086] Specifically, the energy and thermal control subsystem provides energy and thermal control guarantees for the stable operation of the space super-intelligent fusion computing power system, which is the key to the continuous operation of the system. For space infrastructure such as space stations or satellites (currently deployed in space), there are energy systems, such as: setting up solar wings to collect solar energy and setting up battery energy storage, which can continuously supply power to the space super-intelligent fusion computing power system, giving priority to guaranteeing the power supply of key components. The space super-intelligent fusion computing power system as a whole adopts multi-way power supply access to ensure energy reliability. For thermal control, the space infrastructure (currently deployed in space) adopts a multi-level thermal control closed loop. For the space infrastructure as a whole, the principle of radiation heat dissipation is used to quickly dissipate the heat generated by the equipment into space. A two-phase capillary pump circulation heat pipe array is integrated on the bottom plate of each module to ensure uniform heating of the bottom plate. However, due to the particularity of the space super-intelligent fusion computing power system, it has the characteristics of local high heat, so it is necessary to further perform thermal control on high-heat components such as chips. At the chip, for the integrated construction method, the computing system is connected to the cooling system (liquid cooling loop) provided by the current deployment space. The liquid cooling plate on each blade server provides a liquid inlet and outlet, which is interconnected with the cooling system provided by the current deployment space, and the local high-heat components are covered with liquid cooling plates and memory clips. For the uniform and foundation types, it is necessary to apply a directional high-thermal conductivity graphene composite film (in-plane thermal conductivity coefficient>1500W / (m·K)) on the chip surface, establish a three-dimensional heat flow channel through the optimization design of the heat conduction path, conduct the local heat flux density>50W / cm² hot spots to the secondary heat sink system, and realize heat diffusion in combination with the isothermal characteristics of the heat pipe array.

[0087] Based on the above example, the energy and thermal management subsystem can also perform computing power consumption control, which can be used for:

[0088] According to the moon shadow status, the preset minimum main frequency and the preset maximum main frequency are determined;

[0089] According to the target job, the estimated number of computing nodes is determined, and the number of retained nodes is determined according to the estimated number of computing nodes and the idle node retention threshold; the switch state of each computing node in the computing subsystem is controlled according to the retained node amount; for each computing node whose switch state is turned on, when the computing node does not receive a job signal notification, the computing node is controlled to operate at a preset minimum main frequency, and when the computing node receives a job signal notification, the computing node is controlled to operate at a preset maximum main frequency.

[0090] Among them, the computing nodes correspond to computing blade servers. The formation of the lunar shadow is due to the fact that when the moon is between the sun and the earth, the moon blocks some of the sunlight, causing certain areas on the earth to be in shadow. This shadow is called the lunar shadow, and the lunar shadow state is used to indicate whether there is a lunar shadow. The idle node retention threshold is the part of the nodes reserved to ensure the normal use of the system. The estimated number of computing nodes is the number of computing nodes expected to be used when the target job is executed. The retained node quantity is the sum of the estimated number of computing nodes and the idle node retention threshold, and it is the number of computing nodes that the control system retains and can use. The switch state includes on and off. The job signal notification is the notification for executing the target job. The preset minimum main frequency and the preset maximum main frequency are the minimum and maximum main frequencies that can be adopted when the computing nodes are running, which are preset.

[0091] Specifically, observe the lunar shadow state, and according to the different lunar shadow states, determine the corresponding preset minimum main frequency and preset maximum main frequency. Analyze the target job to predict the number of computing nodes required, which is the estimated number of computing nodes. Determine the sum of the estimated number of computing nodes and the idle node retention threshold as the retained node quantity. In the control computing subsystem, select the computing nodes with the retained node quantity to be turned on, that is, the switch state is on, and the switch states of other computing nodes are off, so as to minimize the energy consumption generated by turning on redundant computing nodes. For each computing node with the switch state on, when the computing node does not receive the job signal notification, in order to reduce energy consumption, the computing node can be controlled to operate at the preset minimum main frequency. When the computing node receives the job signal notification, control the computing node to operate at the preset maximum main frequency to execute the target job.

[0092] Exemplarily, in terms of computing power consumption control, for computing nodes, that is, computing blade servers, the computing volume (estimated number of computing nodes) can be estimated according to the target job, the idle node retention threshold can be set, the retained node quantity can be determined, and all computing nodes exceeding the retained node quantity are turned off (the on state is off). For processors that do not participate in computing and the computing nodes are in the idle state, the preset minimum main frequency is set to maintain basic operation. After the computing node receives the job signal notification, the preset maximum main frequency is set to ensure job computing, thereby saving energy consumption. Since the heat dissipation capacity per unit area in the current deployment space is still limited, and too high a temperature will cause the chip working efficiency to decline, a power consumption - performance model is established to ensure a limited reduction in performance while controlling power consumption. Before the lunar shadow state appears, the power consumption requirement is lowered, and after the lunar shadow state appears, it is restored to the maximum load.

[0093] Collect solar energy through the solar wings and set up energy storage batteries to maintain power supply during the influence of the lunar shadow. Overall, a multi-level thermal control closed loop is adopted, and the thermal control closed loop of high-density servers in multiple space modules is completed through high thermal conductivity interface materials, heat pipe arrays, loop heat pipes, and radiation heat dissipation, etc., to keep them with high computing capabilities. In terms of details, refined energy-saving control is carried out on the chips, and the influence period of the lunar shadow is automatically identified and estimated, and the load is reduced in advance. When the external temperature is low, the equipment is kept at a constant temperature.

[0094] Based on the above examples, the space super-intelligent fusion computing power system further includes:

[0095] The system software subsystem is used to store and start the system software required by each subsystem;

[0096] The application software subsystem is used to store and integrate the application software corresponding to each technical field.

[0097] Specifically, the system software subsystem includes the operating system used by the host, resource scheduling tools, compilation environment, file system software, parallel environment, etc., to support the high-speed and efficient operation of applications. The application software subsystem (packaged, pre-set software source, cold storage of Blu-ray) integrates the computing requirements in multiple fields and provides key support for each field by virtue of the unique environmental advantages of space. In terms of intelligent training and reasoning, the space environment is used to achieve efficient model training, which helps tasks such as autonomous navigation and target recognition of aircraft, and improves the intelligent level of space missions. In the fields of astronomy and cosmology, it quickly processes the massive data obtained by astronomical equipment, helps scientists explore the mysteries of the universe, and promotes the research on the evolution of the universe, dark matter, etc. In earth observation and environmental monitoring, satellite data is analyzed in real time to achieve rapid early warning of natural disasters and climate change, and provides data support for earth ecological protection and disaster response. Space science experiments and simulations provide data and theoretical support for actual space experiments by simulating experiments in the space environment, reducing the experimental cost and risk. Low-altitude economic space management and aircraft scheduling achieve precise scheduling and airspace management through space supercomputing, promoting the orderly development of the low-altitude economy. It provides reference for the national defense and military industry. In the fields of high-energy physics and particle research, space mission planning and autonomous navigation research, biology and life science, and deep space exploration science, etc., powerful computing capabilities can be obtained through space supercomputing, promoting the in-depth development of research in each field. The diverse requirements at the application layer are the source driving force for the development of the entire space supercomputing technology system architecture, driving the continuous optimization and innovation of other levels.

[0098] The distributed system software technology that supports the operation of the space super-intelligent fusion computing power system is proposed, which can support the space computing power system with different hardware composition modes. The software includes a distributed scheduling center, a multi-level storage medium control system, etc., which can support cross-space module computing resource scheduling and data sharing, and can be extended to multi-satellite computing resource scheduling and data sharing according to requirements.

[0099] Different from previous space computing power systems that focused on small computing power single satellites and single computing architectures, the technology described in this embodiment proposes new collaborative service technologies for space-based and ground-based computing power systems, scheduling technologies for multiple space-based computing power systems based on federated scheduling, and collaborative scheduling and management technologies for computing resources with different architectures such as ultra-high-precision computing and intelligent computing. Combining parallel and distributed computing frameworks, the resources of the space-ground integrated super-intelligent fusion computing power system are managed and scheduled at different granularities. While being able to support applications with traditional weak computing requirements, it realizes the efficient processing of complex computing tasks on the space-ground integrated super-intelligent fusion computing power system, breaking through the limitations of the traditional situation of space computing power systems, such as insufficient computing power integration ability between small computing power single satellites and multiple satellites and single type of computing resources.

[0100] The space super-intelligent fusion computing power system aims to build a comprehensive space computing power system integrating powerful computing capabilities, efficient storage, stable communication, reliable energy supply and heat dissipation, and a solid hardware foundation. By integrating the innovative technologies of each subsystem, it realizes a leap in space computing capabilities, achieves collaborative services with ground supercomputers, and meets the needs of complex computing tasks in multiple fields.

[0101] The operation mode can refer to Figure 2 As shown, according to different actual needs, multiple space super-intelligent fusion computing power systems can be deployed in low-Earth orbits such as low-Earth orbits and sun-synchronous radiation orbits, or even in deep space. Every certain number of space super-intelligent fusion computing power systems in the same orbit form a computing power system cluster to provide computing services externally. These space super-intelligent fusion computing power systems can interact with super, intelligent, and data centers (supercomputing centers, intelligent computing centers, data centers) built on the ground in cities, oceans, and even deserts, collaborate to provide computing services, meet the needs of vehicle networking, artificial intelligence, and low-altitude economic space management, and can bring the services to the ground through mobile phones by virtue of space communication advantages. The space super-intelligent fusion computing power systems in low-Earth orbits can also directly observe remote sensing data and perform remote sensing data processing, avoiding the low efficiency caused by remotely transmitting remote sensing data. Due to the low ground coverage rate of satellites in low-Earth orbits, communication dedicated satellites are deployed in high-Earth orbits to build a GEO (geostationary orbit satellite) communication network for fast data communication between low-Earth orbit computing power systems in different regions and also for data transmission relay with ground communication. In addition, high-Earth orbit satellites can collect deep space data and transmit it back to low-Earth orbit computing power system satellites for computing and processing.

[0102] The present invention has the following technical effects:

[0103] The computing subsystem determines the computing resources within the current deployment space, receives the target job, and when the computing resources meet the computing requirements of the target job, processes the target job based on the computing resources. The storage subsystem determines the read-write characteristics corresponding to the target job during the execution of the target job, and based on the read-write characteristics, determines the read-write path, and stores the data corresponding to the target job based on the read-write path. The communication subsystem establishes a first communication channel between the subsystems and a second communication channel between the space super-intelligent fusion computing system and the external system. The global management subsystem monitors the operating states of the subsystems, and when the operating state of at least one subsystem is a fault state, based on the rule engine, determines the task instructions corresponding to the fault state, and according to the agents corresponding to the task instructions, executes the task instructions to change the operating state to a normal state, achieving the effect of integrating powerful computing capabilities, efficient storage, stable communication, reliable energy supply and heat dissipation, and a solid hardware foundation in space, effectively improving the performance of space intelligent computing.

[0104] It should be noted that the terms used in the present invention are only for describing specific embodiments and do not limit the scope of the present application. As shown in the specification of the present invention, unless the context clearly indicates an exception, words such as "a", "one", "kind", and / or "the" do not specifically refer to the singular and may also include the plural. The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method or device including the said element.

[0105] It should also be noted that the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation of the present invention. Unless otherwise clearly specified and limited, terms such as "installed", "connected", "connected to" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A space super-intelligent fusion computing power system, characterized in that, Including: A computing subsystem for determining computing resources within the current deployment space, receiving a target job, and processing the target job based on the computing resources when the computing resources meet the computing requirements of the target job; A storage subsystem for determining read-write characteristics corresponding to the target job when executing the target job, determining a read-write path according to the read-write characteristics, and storing data corresponding to the target job based on the read-write path; wherein, the read-write characteristics are determined according to the read-write throughput, read-write burstiness, and source data operation volume of the target job; A communication subsystem for establishing a first communication channel between each subsystem and a second communication channel between the space super-intelligent fusion computing power system and an external system; A global management subsystem for monitoring the operating states of each subsystem, and when there is at least one subsystem with a fault state, determining respective task instructions corresponding to the fault state based on a rule engine, and executing the respective task instructions according to respective agents corresponding to the respective task instructions to change the operating state to a normal state.

2. The system according to claim 1, wherein There is a system rack within the current deployment space, the system rack evenly distributes a plurality of space modules, and each subsystem is placed on at least one space module; the first quantity of space modules corresponding to the computing subsystem is determined according to a preset computing requirement, the computing capacity corresponding to a single computing blade server, and the maximum quantity of computing blade servers corresponding to each space module; the second quantity of space modules corresponding to the storage subsystem is determined according to a preset storage requirement, the storage capacity of storage blade servers corresponding to each storage medium, and the maximum quantity of storage blade servers corresponding to each space module; the third quantity of space modules corresponding to the communication subsystem is determined according to the preset computing requirement, the computing capacity corresponding to a single computing blade server, the preset storage requirement, the storage capacity of storage blade servers corresponding to each storage medium, and the maximum quantity of network blade servers corresponding to each space module; the fourth quantity of space modules corresponding to the global management subsystem is at least two; the sum of the first quantity, the second quantity, the third quantity, and the fourth quantity is less than or equal to the total quantity of space modules on the system rack.

3. The system according to claim 1, characterized in that, The second communication channel includes an inter-satellite communication channel and a satellite-ground communication channel, and the communication subsystem is further configured to: Transmit data corresponding to the target job based on an encrypted time-sensitive network; Wherein, the encrypted time-sensitive network extends encrypted information in the network identification number field of the data frame at the Ethernet link layer, and adds a security label field and an integrity check value field to the data frame.

4. The system according to claim 1, wherein The computing subsystem is further configured to: when the computing resources do not meet the computing requirements of the target job, send the computing resources and the computing requirements to the communication subsystem; The communication subsystem is further configured to: receive the computing resources and the computing requirements, and send the computing resources and the computing requirements to the ground scheduling server, so that the ground scheduling server determines a space cooperation system according to the computing resources and the computing requirements; receive the space cooperation system corresponding to the computing requirements fed back by the ground scheduling server, establish a second communication channel with the space cooperation system, and cooperate with the space super-intelligent fusion computing power system in the current deployment space and the space cooperation systems in other deployment spaces to process the target job based on the second communication channel; 5. The system according to claim 1, wherein The storage subsystem includes: an acceleration storage layer, a persistent storage layer, and a cold data storage layer; the number of storage blade servers in the persistent storage layer is less than the number of storage blade servers in the acceleration storage layer; wherein, the read-write path includes a first path and a second path; the acceleration storage layer and the persistent storage layer together serve as the first path, and the persistent storage layer serves as the second path; when the read-write path is the first path, a parallel file system is constructed based on the acceleration storage layer, data is stored based on the parallel file system, and after the target job is completed, the data in the parallel file system is copied to the persistent storage layer, and the parallel file system is destroyed; The cold data storage layer is configured to store the packaged data corresponding to the generated data in the persistent storage layer when the stable time of the stored data in the persistent storage layer reaches a first time threshold; 6. The system according to claim 5, wherein The storage subsystem further includes: a software image library; wherein, the software image library is configured to store software images corresponding to each application program, so as to pull the corresponding software image from the software image library when the application program cannot run; when the update completion duration reaches a preset update period, the software image library obtains a new software image from the ground data server through the communication subsystem to update the software image library; 7. The system according to claim 1, characterized in that, It further includes: an energy and thermal control management subsystem; the energy and thermal control management subsystem is connected to the energy system and the refrigeration system in the current deployment space; The energy and thermal control management subsystem is configured to connect the liquid cooling inlets and outlets of each blade server in each subsystem to the refrigeration system in the current deployment space; the energy and thermal control management subsystem further includes a peripheral air cooling device for thermal management of each subsystem; 8. The system according to claim 7, wherein The energy and thermal control management subsystem is further configured to: determine an idle node retention threshold according to the lunar phase state; determine an estimated number of computing nodes according to the target job, and determine the number of retained nodes according to the estimated number of computing nodes and the idle node retention threshold; Control the switch states of the computing nodes in the computing subsystem according to the remaining node quantity; for each computing node with the switch state being on, when the computing node does not receive a job signal notification, control the computing node to operate at a preset minimum main frequency, and when the computing node receives a job signal notification, control the computing node to operate at a preset maximum main frequency; wherein, the computing node corresponds to a computing blade server.

9. The system according to claim 1, wherein The global management subsystem is further configured to: Regularly collect monitoring data from each subsystem, analyze the monitoring data based on a rule engine, and determine a target processing instruction; Based on a commander agent, receive the target processing instruction pushed by the rule engine, analyze and disassemble the target processing instruction to obtain task instructions, and send each task instruction to the corresponding function agent respectively, so that each function agent executes the corresponding task instruction.

10. The system according to claim 1, characterized in that, It further includes: A system software subsystem for storing and starting the system software required by each subsystem; An application software subsystem for storing and integrating the application software corresponding to each technical field.

Images