Computing power dynamic expansion method and system

Abstract

The application provides a computing power dynamic expansion method and system. The method comprises: establishing a communication association channel between a computing power host and a computing power module, the computing power module determining a computing unit and obtaining a computing power parameter, encapsulating a virtual resource descriptor and sending it to the computing power host; the computing power host maps the computing unit to a remotely schedulable virtual resource node based on the descriptor, and integrates with the local computing power resource to generate a global computing power resource pool. Through the dynamic discovery and virtualization mapping mechanism of the master-slave architecture, the application solves the technical defects of poor scalability of traditional fixed computing power architecture and difficulty in effectively integrating external computing power, realizes unified scheduling and flexible expansion of heterogeneous computing power resources, and improves the overall utilization rate of computing power resources.

Description

Technical Field

[0001] This application relates to the field of computer computing power allocation technology, and more specifically, to a method and system for dynamic expansion of computing power. Background Technology

[0002] With the proliferation of artificial intelligence, big data analytics, and massively multi-scale computing tasks, the local computing power of a single electronic device often falls short of the ever-increasing demands for real-time processing. To improve computing performance, it is typically necessary to expand existing computing resources, enabling multiple computing devices to collaborate on complex tasks.

[0003] Existing computing power expansion solutions typically employ a clustered deployment strategy, connecting multiple servers together through a pre-configured fixed local area network environment. This approach first requires manual registration of the hardware configuration and network address allocation for each server; then, cluster management software establishes long-lived connections between the servers to maintain heartbeat monitoring; finally, a central scheduler divides the computing tasks into multiple subtasks and distributes them to the fixed nodes for execution according to a preset static load balancing algorithm.

[0004] However, this traditional computing power expansion scheme has significant technical drawbacks. Because it relies on a pre-configured static network environment and manual node registration, the system lacks the ability to dynamically perceive external idle computing power, making it unable to discover and utilize temporarily appearing computing resources in the environment in real time. Furthermore, due to the lack of a unified abstraction and virtualization encapsulation mechanism for heterogeneous computing power units, it is difficult to achieve deep topological integration of external computing power resources with local computing power, resulting in low flexibility in global resource scheduling and an inability to meet the complex and ever-changing dynamic expansion needs of computing power. Summary of the Invention

[0005] This application provides a method and system for dynamically expanding computing power to at least alleviate the aforementioned technical problems.

[0006] A method for dynamically scaling computing power includes: Step 1: Establish a communication channel between the computing power host and the computing power module. The computing power module determines its own computing unit and obtains the corresponding computing power parameters. It encapsulates the computing power parameters to generate a virtual resource descriptor and sends it to the computing power host through the communication channel. Step 2: The computing host maps the computing units in the computing module to virtual resource nodes that can be remotely scheduled based on the virtual resource descriptor and integrates them with its own local computing resources to generate a global computing resource pool that can be called by upper-layer applications.

[0007] Optionally, step 1 further includes: identifying all electronic devices participating in the dynamic expansion of computing power, configuring the electronic devices to perform dynamic expansion of computing power roles, dividing the electronic devices into computing power hosts that act as master control devices and computing power modules that act as slave devices, and establishing a communication association channel between the computing power hosts that act as master control devices and the computing power modules that act as slave devices.

[0008] Optionally, step 1 further includes: the computing power host broadcasts a computing power module detection request to the outside world according to a preset period, and receives a computing power idle status beacon frame returned by the computing power module, so as to identify the computing power module in the idle state, so that the computing power module in the idle state can determine its own computing unit and obtain the corresponding computing power parameters.

[0009] Optionally, when the computing power host determines that a computing power module is in an idle state, it parses the computing power idle state beacon frame to obtain the module computing power ready identifier and the idle time slot sequence, and determines whether the computing power module is in an idle state based on the module computing power ready identifier and the idle time slot sequence.

[0010] Optionally, when it is determined to be in an idle state, the computing power host generates a master-slave collaborative link establishment instruction and sends it to the computing power module. In response to receiving a link response confirmation frame returned by the computing power module, a communication association channel is established between the computing power host and the computing power module.

[0011] Optionally, before the computing power module determines its own computing unit and obtains the corresponding computing power parameters, the following steps are taken: based on the communication association channel, the computing power host and the computing power module of the return link response confirmation frame perform two-way authentication to verify the credibility of the communication between the computing power host and the computing power module of the return link response confirmation frame, obtain a two-way authentication pass credential, and establish a computing power interaction channel based on the two-way authentication pass credential to send the virtual resource descriptor to the computing power host through the computing power interaction channel.

[0012] Optionally, in step 2, the computing host maps the computing units in the computing module to remotely schedulable virtual resource nodes based on virtual resource descriptors and integrates them with its own local computing resources to generate a global computing resource pool that can be invoked by upper-layer applications, including: The virtual resource parsing engine configured on the computing power host parses the virtual resource descriptor to obtain the computing power unit topology attributes, converts the computing power unit topology attributes into a remote computing power scheduling adaptation instruction set and sends it to the computing power module. The resource attribute calibration unit in the computing power module performs computing power topology normalization calibration processing on the remote computing power scheduling adaptation instruction set to obtain standardized computing power topology mapping metadata. The computing power virtualization mapping unit in the computing power module maps its computing unit into a virtual resource node that can be remotely scheduled according to the standardized computing power topology mapping metadata.

[0013] Optionally, the computing host is configured with a global computing power orchestration engine and a resource identifier allocator. Correspondingly, in step 2, the computing host maps the computing units in the computing power module to remotely schedulable virtual resource nodes based on virtual resource descriptors and integrates them with its own local computing power resources to generate a global computing power resource pool that can be invoked by upper-layer applications, including: The global computing power orchestration engine assigns globally unique computing power resource identifiers to virtual resource nodes and determines the local computing power topology addressing index of the computing power host's own local computing power resources. Based on globally unique computing resource identifiers and local computing topology addressing indexes, the resource identifier allocator generates a cross-source computing resource collaborative addressing table. This table is then used to encapsulate and address the local computing resources of virtual resource nodes and the computing host itself, thereby generating a global computing resource pool that can be invoked by upper-layer applications.

[0014] Optionally, a cross-source computing resource collaborative addressing table is generated by the resource identifier allocator based on globally unique computing resource identifiers and local computing topology addressing indexes, including: The globally unique computing resource identifier is subjected to identifier feature entropy encoding to obtain a unique addressing feature code for the computing resource. Based on the unique addressing feature code for the computing resource, dual-end computing topology addressing benchmarks are generated for computing units in the computing host and computing module, respectively. Based on the dual-end computing topology addressing benchmarks corresponding to computing units in the computing host and computing module, a local computing topology addressing index is generated by same-source topology anchoring matching to generate a cross-source computing resource collaborative addressing table.

[0015] A dynamic computing power expansion system includes: a computing power host and computing power modules. A communication channel is established between the computing power host and the computing power modules. The computing power modules determine their own computing units and obtain corresponding computing power parameters. The computing power parameters are encapsulated to generate virtual resource descriptors and sent to the computing power host through the communication channel. The computing power host maps the computing units in the computing power modules to remotely scheduled virtual resource nodes based on the virtual resource descriptors and integrates them with its own local computing power resources to generate a global computing power resource pool that can be invoked by upper-layer applications.

[0016] Technical advantages of the technical solution provided in this application This application's method for dynamically expanding computing power addresses the technical shortcomings of traditional computing power expansion schemes, such as a lack of dynamic awareness, difficulty in deeply integrating heterogeneous resources, and low scheduling flexibility. It establishes a communication channel between the computing power host and the computing power modules, and the computing power modules proactively encapsulate virtual resource descriptors and send them to the host. This solves the problems of manual registration and static configuration in traditional schemes. Compared to the fixed server cluster connections in traditional schemes, this application, through the decoupling and dynamic association between the computing power host and the computing power modules, enables the system to automatically sense and acquire the computing power parameters of external computing units, improving the automation level of resource acquisition.

[0017] By mapping computing modules to remotely schedulable virtual resource nodes using virtual resource descriptors and integrating them with local resources to generate a global computing resource pool, this approach solves the problem of deep integration between external and local resources in traditional solutions. Traditional solutions typically only achieve task-level distribution, while this application constructs a unified addressing and scheduling logic through virtualization mapping of heterogeneous computing units. The generated global computing resource pool can shield the underlying hardware differences. Compared to the static load balancing of traditional solutions, this integration method based on virtual resource nodes makes it as smooth as calling local resources when upper-layer applications access computing power, significantly improving the flexibility of computing resource scheduling and the overall utilization efficiency of global resources. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for dynamically expanding computing power according to an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of a computing power dynamic expansion system according to an embodiment of this application. Detailed Implementation

[0020] like Figure 1 The image shows an embodiment of a method for dynamically expanding computing power according to this application, which includes: Step 1: Establish a communication channel between the computing power host and the computing power module. The computing power module determines its own computing unit and obtains the corresponding computing power parameters. It encapsulates the computing power parameters to generate a virtual resource descriptor and sends it to the computing power host through the communication channel. Step 2: The computing host maps the computing units in the computing module to virtual resource nodes that can be remotely scheduled based on the virtual resource descriptor and integrates them with its own local computing resources to generate a global computing resource pool that can be called by upper-layer applications.

[0021] Optionally, step 1 further includes: identifying all electronic devices participating in the dynamic expansion of computing power, configuring the electronic devices to perform dynamic expansion of computing power roles, dividing the electronic devices into computing power hosts as master control devices and computing power modules as slave devices, and establishing a communication association channel between the computing power hosts as master control devices and the computing power modules as slave devices.

[0022] Preferably, in determining all electronic devices participating in the dynamic expansion of computing power, a network discovery protocol is used to periodically send online status polling signals to nodes within the local area network to obtain a set of candidate electronic devices currently in an active state. Hardware attribute sniffing is performed on each candidate electronic device in the set to obtain basic hardware parameters, including physical address, core clock speed, free memory capacity, and cache bandwidth. Based on these basic hardware parameters, the candidate electronic devices are scored according to their computing performance weights. The candidate electronic device with the highest score is defined as a computing power host with management privileges, while other candidate electronic devices with scores below a threshold (e.g., devices with scores below 60) are defined as computing power modules with controlled execution privileges, thereby completing the dynamic expansion of computing power role configuration.

[0023] Preferably, when configuring dynamic computing power expansion roles, the computing power host pushes role confirmation signals to each computing power module in the set of candidate electronic devices to activate the role response mechanism within the computing power module. When the computing power module receives the role confirmation signal, its internal role switching logic performs a secondary verification of the current operating load. After confirming that it is not in a high-load state, it sends a role acceptance response signal back to the computing power host. Through this two-way confirmation handshake process, the computing power host and the computing power module logically establish master-slave relationships, providing a clear control flow for subsequent cross-device computing power resource integration and avoiding potential instruction conflicts during concurrent control of multiple devices.

[0024] Preferably, in the specific technical implementation of establishing the communication association channel, the computing power host assigns a unique logical link identifier (e.g., a 32-bit binary string) to each confirmed computing power module based on the underlying transmission control protocol. The computing power host uses the logical link identifier to establish a logical link based on asynchronous sockets with each computing power module, thereby constructing a communication association channel for carrying control commands and virtual resource descriptors. The communication association channel has a heartbeat detection mechanism, which evaluates the communication quality of the communication association channel by monitoring the round-trip latency of the logical link in real time, ensuring that the packet loss rate of the communication association channel remains at a low level during the transmission of the virtual resource descriptors.

[0025] Preferably, to improve the granularity of the computing power host's management of the computing power modules, during the establishment of the communication association channel, the computing power host further divides the communication association channel into a control plane sub-channel and a data plane sub-channel for different communication service types. The control plane sub-channel is specifically responsible for transmitting lightweight control messages, including role confirmation signaling and connection maintenance instructions, while the data plane sub-channel is responsible for transmitting subsequently generated virtual resource descriptors. Through this sub-channel isolation mechanism, even when the data volume of the virtual resource descriptors is large, the computing power host can still achieve real-time control of the computing power modules, ensuring that the scheduling instructions in the entire dynamic expansion process of computing power are not blocked.

[0026] Preferably, in one scenario, to address the complexity of the local area network environment, the computing host will activate adaptive multipath transmission logic when establishing a communication association channel. This logic simultaneously detects multiple physical links, including Wireless Fidelity and Ultra Wideband, and selects the optimal physical bearer path for the communication association channel based on the signal-to-noise ratio and bandwidth stability of each physical link. Through this multipath redundancy mapping method, a highly stable communication association channel is established, providing a highly reliable transmission medium for the subsequent identification of the computing units within the computing module and the encapsulation and transmission of virtual resource descriptors.

[0027] Preferably, after forming the computing power host as the master control device and the computing power modules as slave devices, the computing power host performs unified indexing management of all communication channels, generating a master-slave device topology association table. The master-slave device topology association table records in detail the physical address, allocated logical link identifier, and available bandwidth of the current physical bearer path for each computing power module. By maintaining the master-slave device topology association table, the computing power host can monitor the online status of the computing power modules in real time. Once a computing power module goes offline, causing its corresponding communication channel to disconnect, the computing power host will immediately update the master-slave device topology association table and trigger the computing power dynamic expansion role reconfiguration logic, thereby ensuring the real-time performance of the global computing power resource pool and the robustness of the computing power dynamic expansion scheme.

[0028] Optionally, step 1 further includes: the computing power host broadcasts a computing power module detection request to the outside world according to a preset period, and receives a computing power idle status beacon frame returned by the computing power module to determine the computing power module in the idle state, so that the computing power module in the idle state can determine its own computing unit and obtain the corresponding computing power parameters.

[0029] Preferably, in the specific technical implementation where the computing power host broadcasts computing power module detection requests according to a preset period, the scheduling and management logic inside the computing power host activates a high-precision timer trigger. The timer trigger generates a detection activation pulse based on a preset detection frequency (e.g., triggered once every 500 to 2000 milliseconds). In response to the detection activation pulse, the computing power host encapsulates and generates a computing power module detection request containing the host identity identifier, the current system timestamp, and the expected computing power level requirement, and multicasts the computing power module detection request to each computing power module within the local area network environment through the networking protocol layer. Through this periodic broadcast mechanism, the computing power host can continuously sense newly added or changing computing power modules in the network topology.

[0030] Preferably, when each computing module in the network receives a computing module probe request sent by the computing host, the resource monitoring module inside each computing module extracts the expected computing power level requirement. Each computing module then calls the system kernel's performance monitoring interface in real time to obtain the current processor utilization, memory usage, and task queue depth. Each computing module calculates the matching degree between the obtained performance indicators and the expected computing power level requirement to obtain a resource availability score reflecting the current device workload. If the resource availability score reaches a preset threshold, each computing module sets its status to standby and prepares to construct a computing power idle state beacon frame, providing a decision-making basis for subsequent feedback of the computing power idle state beacon frame.

[0031] Preferably, when each computing module is in the standby state, its internal beacon generation module performs a data encapsulation operation to generate the computing power idle state beacon frame. The computing power idle state beacon frame contains an 8-bit or 16-bit module computing power readiness identifier and an idle time slot sequence reflecting the future schedulable time window of the device. The idle time slot sequence consists of multiple discrete time segments, each recording its start time and duration. Each computing module selects a transmission time using a random delay backoff algorithm to feed back the computing power idle state beacon frame to the computing power host. By introducing the idle time slot sequence, the computing power host can not only know which computing modules are idle, but also predict the computing power supply stability of each computing module over time.

[0032] Preferably, after receiving the idle status beacon frames returned by each computing module, the computing power host initiates a parallel parsing process through its internal resource evaluation logic. The computing power host first extracts the module's computing power readiness identifier from the idle status beacon frame and verifies its validity, discarding invalid frames with abnormal identifiers or verification errors. Subsequently, the computing power host performs time overlap analysis on the verified idle time slot sequences, calculating the computing power contribution weight of each computing module in a predetermined future period. Based on the verification results of the module's computing power readiness identifier and the ranking of the computing power contribution weights, the computing power host identifies the computing power modules in an idle state and adds them to the active computing power supply list, thereby achieving dynamic and accurate locking of distributed idle computing power.

[0033] Preferably, after identifying the idle computing modules, the computing host sends a computing unit identification command to the target idle computing modules. Upon receiving the computing unit identification command, the idle computing module invokes the underlying hardware abstraction layer interface to enumerate various physical computing units integrated within it, such as general-purpose processors, graphics processors, neural network processors, and digital signal processors. For each enumerated physical computing unit, the idle computing module obtains corresponding computing power parameters through performance evaluation operators. These parameters include, but are not limited to, peak floating-point operation value, memory access bandwidth, and instruction set compatibility version. Through this deep hardware sniffing, the idle computing module can fully grasp the baseline of its own available raw computing power resources, forming a multi-dimensional set of computing power parameters.

[0034] Preferably, after acquiring the computing power parameters, the idle computing power module initiates virtualization encapsulation logic to perform structured processing on the computing power parameters. Based on the affinity and topology of the physical computing units, the idle computing power module converts fragmented physical hardware information into a logically continuous computing resource view. The virtualization encapsulation logic associates the computing resource view with the corresponding computing power parameters and encapsulates it into a virtual resource descriptor according to a predefined description protocol. Subsequently, the idle computing power module sends the virtual resource descriptor to the computing host through an established communication channel. By converting the original physical parameters into the virtual resource descriptor, the hardware heterogeneity of the underlying computing power modules is shielded, enabling the computing host to perform resource mapping and scheduling integration of remote computing units based on a unified logical standard.

[0035] Optionally, when the computing power host determines that a computing power module is in an idle state, it parses the computing power idle state beacon frame to obtain the module computing power ready identifier and the idle time slot sequence, and determines whether the computing power module is in an idle state based on the module computing power ready identifier and the idle time slot sequence.

[0036] Preferably, during the process of the computing power host determining that a computing power module is in an idle state, the computing power host receives a computing power idle status beacon frame fed back from the computing power module. The computing power host performs streaming demultiplexing processing on the binary bit stream of the computing power idle status beacon frame using a preset frame destructuring operator to extract the payload feature data located after the frame header check field. The computing power host then uses the frame destructuring operator to perform offset-based field segmentation on the payload feature data to separate the module computing power readiness identifier representing the real-time readiness level of the hardware and the idle time slot sequence reflecting the resource distribution in the time dimension, thereby providing raw data support for subsequent accurate status determination.

[0037] Preferably, in the technical processing step of parsing the module computing power readiness identifier, the computing power host inputs the separated module computing power readiness identifier into a pre-built state mapping table. The state mapping table records the logical relationship between different values ​​and the internal hardware self-test status of the computing power module (for example, a value of one represents that the core logic unit is normal, and a value of zero represents a hardware fault). The computing power host determines whether the underlying hardware of the computing power module meets the physical prerequisites for receiving external scheduling instructions by comparing the current module computing power readiness identifier with the target health threshold, and thus produces a readiness verification result. This step can eliminate abnormal nodes that are online but in a hardware error state, improving the success rate of subsequent computing power resource integration.

[0038] Preferably, in the specific technical implementation of parsing the idle time slot sequence, the computing power host performs time-domain reconstruction of the idle time slot sequence using a time-series decoder. The idle time slot sequence is technically represented as a discrete time-state matrix, where the rows of the discrete time-state matrix represent preset monitoring period time steps (e.g., every ten milliseconds as a step size), and the columns of the discrete time-state matrix represent the indices of computing units within the computing power module. The element located at the intersection of the row and column represents the predicted load occupancy of the corresponding computing unit at the corresponding time step. The computing power host traverses the discrete time-state matrix, counts the consecutive idle time steps, and thus generates an available time assessment value.

[0039] Preferably, after determining the available duration assessment value, the computing power host performs a logical matching process based on resource weights. The computing power host defines a target computing power duration threshold based on the scale of the computing tasks issued by the current upper-layer application. The computing power host compares the available duration assessment value with the target computing power duration threshold to determine whether the computing power module's resource reserves have high stability over a time scale. If the available duration assessment value is higher than the target computing power duration threshold, it is determined that the computing power module has a time window to execute a complete computing task, thereby generating a time-domain availability determination result.

[0040] Preferably, in one scenario, the computing power host inputs the readiness verification result and the time-domain availability determination result to the multi-dimensional collaborative logic determination module. The multi-dimensional collaborative logic determination module performs a logical AND operation, that is: the multi-dimensional collaborative logic determination module officially confirms that the computing power module is in an idle state only when the readiness verification result indicates that the module's computing power readiness identifier is within a legal range defined by the target health threshold, and the time-domain availability determination result indicates that the idle time slot sequence meets the continuity requirements of task scheduling. Through this dual verification mechanism combining hardware status and time distribution, scheduling failures caused by the computing power module being momentarily idle but about to enter a high-load state are avoided, thus ensuring the reliability of each node in the global computing power resource pool.

[0041] Preferably, after confirming that the computing module is in an idle state, the computing host synchronizes the physical address of the idle computing module and the corresponding available duration assessment value to the dynamic resource entry in the master-slave device topology association table. The computing host locks the physical address of the idle computing module by performing hash addressing on the master-slave device topology association table and triggers subsequent computing unit identification instructions to it. This process enables the computing host to monitor the distribution heatmap of idle resources within the local area network environment in real time, providing a deterministic execution target for subsequent topology mapping and computing power integration based on virtual resource descriptors.

[0042] Optionally, when it is determined to be in an idle state, the computing power host generates a master-slave collaborative link establishment instruction and sends it to the computing power module. In response to receiving a link response confirmation frame returned by the computing power module, a communication association channel is established between the computing power host and the computing power module.

[0043] Preferably, after the computing power host determines that the computing power module is in an idle state, the connection management logic of the computing power host performs pre-handshake preparation actions. The connection management logic extracts the physical address of the idle computing power module from the master-slave device topology association table and, combined with the current local area network load, generates a master-slave collaborative link establishment instruction using the security algorithm. The master-slave collaborative link establishment instruction contains an encrypted hash authentication code generated by the security algorithm and a dynamic port number pre-assigned to this connection task. By generating the master-slave collaborative link establishment instruction containing the aforementioned encrypted hash authentication code and the dynamic port number, a logical foundation is provided for subsequently constructing a communication association channel with unique addressing capabilities.

[0044] Preferably, the computing host encapsulates the generated master-slave collaborative link establishment command into a transport layer protocol data frame payload, and sends the transport layer protocol data frame payload to the idle computing module using a network interface circuit. Simultaneously with sending the master-slave collaborative link establishment command, the retransmission control operator inside the computing host starts a link response timer. If no return signal is detected within a preset timeout threshold, the retransmission control operator triggers a retransmission process for the master-slave collaborative link establishment command. This signaling interaction method based on acknowledgment and retransmission can cope with sudden fluctuations in the local area network environment, improving the reliability of the delivery of the master-slave collaborative link establishment command.

[0045] Preferably, after receiving the master-slave collaborative link establishment instruction sent by the computing host, the idle computing module's internal instruction parsing logic unpacks the data frame payload of the transport layer protocol, extracts and verifies the encrypted hash authentication code within it using the security algorithm. When the verification result indicates that the master-slave collaborative link establishment instruction is legitimate, the idle computing module starts its local service port listening logic based on the dynamic port number carried in the master-slave collaborative link establishment instruction. Simultaneously, the idle computing module performs parameter configuration on its internal controlled hardware interface, pre-allocates a memory buffer for storing data to be transmitted, and thus produces a resource pre-locking result. This process, by pre-locking physical resources, prevents resource race conflicts during link establishment.

[0046] Preferably, based on the resource pre-occupancy locking result, the signaling generation module of the idle computing module constructs a link response confirmation frame. The link response confirmation frame carries a status code reflecting the local resource readiness state and a monotonically increasing synchronization sequence number. The idle computing module transmits the link response confirmation frame back to the computing host via the physical bearer path. This feedback loop constitutes a bidirectional peer-to-peer confirmation between the master and slave devices, enabling the computing host to know in real time that the idle computing module has entered the communication-ready state and has completed clock synchronization and state alignment based on the synchronization sequence number.

[0047] Preferably, in response to receiving the link response confirmation frame from the idle computing module, the computing host performs the final activation process of the logical link. The computing host verifies the status code and synchronization sequence number in the link response confirmation frame. After confirming that the idle computing module has completed port opening and resource locking, the computing host formally establishes a communication association channel between the computing host and the idle computing module. This communication association channel is technically abstracted as a virtual bidirectional pipe with deterministic latency, responsible for shielding the underlying physical link differences between different computing modules, thereby achieving logical coupling between master and slave devices at the resource scheduling level.

[0048] Preferably, after the communication association channel is successfully established and operating stably, the computing power host sends a channel activation signal to the idle computing power module. In response to the channel activation signal, the idle computing power module formally activates its internal virtualization encapsulation logic. At this time, the idle computing power module encapsulates the computing power parameters according to an agreed-upon description format, thereby generating a virtual resource descriptor. Through the communication association channel, the idle computing power module can accurately push the virtual resource descriptor to the virtual resource resolution engine of the computing power host, providing a high-bandwidth and low-error-rate communication implementation for the subsequent unified addressing and dynamic expansion of the global computing power resource pool.

[0049] Optionally, before the computing power module determines its own computing unit and obtains the corresponding computing power parameters, the following steps are taken: based on the communication association channel, the computing power host and the computing power module of the return link response confirmation frame perform two-way authentication to verify the credibility of the communication between the computing power host and the computing power module of the return link response confirmation frame, obtain a two-way authentication pass credential, and establish a computing power interaction channel based on the two-way authentication pass credential to send the virtual resource descriptor to the computing power host through the computing power interaction channel.

[0050] Preferably, in the pre-processing logic where the computing power module determines its own computing unit and obtains the corresponding computing power parameters, the computing power host utilizes the established communication association channel to initiate a two-way authentication process for the idle computing power module that has returned a link response confirmation frame. The security hardening engine inside the computing power host first invokes a preset asymmetric encryption algorithm to generate an original challenge message containing a high-precision timestamp of the current system and random noise perturbation codes. The computing power host pushes the original challenge message to the idle computing power module through the communication association channel as the first probe signal to trigger the two-way authentication process. This dynamically generated challenge mechanism ensures the uniqueness of each authentication process and prevents replay attacks in the local area network environment.

[0051] Preferably, upon receiving the original challenge message, the idle computing module's integrated hardware security module extracts the random noise perturbation code. The hardware security module then uses the computing module's private key to digitally sign the random noise perturbation code, generating device identity verification data. Simultaneously, to achieve two-way verification, the random number generator within the idle computing module generates a reverse verification random number. The idle computing module encapsulates the device identity verification data and the reverse verification random number into an authentication response message and sends it back to the computing host. This step, through hardware-level private key signing, ensures the authenticity and non-repudiation of the idle computing module's identity source.

[0052] Preferably, after receiving the authentication response message, the computing power host uses the public key of the computing power module corresponding to the idle computing power module, stored in the local device list, to verify the device identity data. If the verification passes, the computing power host initially confirms that the idle computing power module is a legitimate modular component. Next, the computing power host uses its private key to perform encryption and transformation on the reverse verification random number in the authentication response message, generating a host identity feature credential. The computing power host then sends the host identity feature credential back to the idle computing power module. This process establishes a basic trust relationship between the master and slave devices through the interactive verification of the public and private key pairs, laying a secure environment for the subsequent transmission of sensitive computing power parameters.

[0053] Preferably, after receiving the host identity feature credential, the idle computing power module calls the pre-stored public key of the computing power host to perform decryption and consistency comparison processing. When the comparison result indicates that the feature data returned by the computing power host completely matches the original reverse verification random number, the idle computing power module confirms that the computing power host has management authority. At this time, the authentication logic units of both the master and slave parties jointly sign the two-way identity authentication pass instruction, generating a two-way identity authentication pass credential with time constraints. The two-way identity authentication pass credential contains a negotiated encryption algorithm identifier and a seed key for subsequent communication. The generation of this credential signifies that the master and slave parties have completed deep verification of communication trustworthiness, effectively intercepting malicious intrusion of unauthorized devices into the global computing power resource pool.

[0054] Preferably, after obtaining the two-way authentication pass credential, the computing power host and the idle computing power module initiate the construction process of the computing power interaction channel. Based on the seed key in the two-way authentication pass credential, the computing power host calls a key derivation operator to generate a set of symmetric encryption session keys. On the logical link of the communication association channel, the computing power host performs streaming encryption processing on the transmitted data packets using the symmetric encryption session keys, thereby establishing an independent computing power interaction channel. Technically, the computing power interaction channel is a high-security encrypted tunnel nested within the communication association channel, specifically responsible for carrying virtual resource descriptors related to the underlying hardware topology.

[0055] Preferably, after the computing power interaction channel is operating stably, the idle computing power module responds to the computing unit identification command issued by the computing power host and formally initiates the identification action of its own computing unit. The idle computing power module inputs the collected computing power parameters into the virtualization encapsulation logic, producing a virtual resource descriptor that reflects the actual processing capability of the hardware. Since the computing power interaction channel protected by the two-way authentication process has been established at this time, the idle computing power module can send the virtual resource descriptor, which contains core performance data, memory topology, and addressing index, to the virtual resource parsing engine of the computing power host in an encrypted and encapsulated form. This interaction mechanism based on the two-way authentication and credential construction ensures the integrity and confidentiality of the metadata of heterogeneous computing power resources during cross-device transmission when they are mapped to virtual resource nodes.

[0056] Optionally, in step 2, the computing host maps the computing units in the computing module to remotely schedulable virtual resource nodes based on virtual resource descriptors and integrates them with its own local computing resources to generate a global computing resource pool that can be invoked by upper-layer applications, including: The virtual resource parsing engine configured on the computing power host parses the virtual resource descriptor to obtain the computing power unit topology attributes, converts the computing power unit topology attributes into a remote computing power scheduling adaptation instruction set and sends it to the computing power module. The resource attribute calibration unit in the computing power module performs computing power topology normalization calibration processing on the remote computing power scheduling adaptation instruction set to obtain standardized computing power topology mapping metadata. The computing power virtualization mapping unit in the computing power module maps its computing unit into a virtual resource node that can be remotely scheduled according to the standardized computing power topology mapping metadata.

[0057] Preferably, during the virtual resource descriptor (VRD) resolution process by the computing host, the VRD resolution engine configured within the computing host invokes a streaming scan operator to perform a structured deconstruction operation on the VRD received through the communication association channel. The streaming scan operator identifies the hardware description tags and resource quantization fields in the VRD, and the VRD resolution engine uses the structured deconstruction operation to transform the originally discrete binary data stream into a raw feature stream of computing units with semantic information. This process establishes the computing host's initial understanding of the physical hardware foundation of idle computing modules through the hardware description tags and resource quantization fields, providing standardized data input for subsequent extraction of computing unit topology attributes.

[0058] Preferably, in the technical step of extracting the topological attributes of the computing power unit, the virtual resource parsing engine uses graph analysis operators to perform correlation mining on the original feature flow of the computing power unit. The graph analysis operators generate the topological attributes of the computing power unit by performing correlation analysis on each hardware dimension in the original feature flow of the computing power unit. Technically, the topological attributes of the computing power unit are represented as a heterogeneous resource topology matrix reflecting the physical connection relationships between computing units, cache levels, and interconnect bandwidth. In this heterogeneous resource topology matrix, rows represent indices of physical computing cores, columns represent corresponding level-three cache paths and memory access channels, and elements at the intersection of rows and columns represent weight values ​​for data interactions. Through this matrix-based representation, the computing power host can clearly identify the affinity characteristics between different computing units within the idle computing power module and generate a topology mapping benchmark based on the heterogeneous resource topology matrix.

[0059] Preferably, during the generation of the remote computing power scheduling adaptation instruction set, the computing power host retrieves preset protocol conversion logic based on the topology mapping benchmark. This protocol conversion logic maps the globally unified scheduling primitives issued by the upper-layer application into low-level configuration instructions that can be recognized by the local execution environment of the idle computing power module, thereby generating the remote computing power scheduling adaptation instruction set. The remote computing power scheduling adaptation instruction set includes a series of parameter configuration items for direct access to remote memory and register context prefetching. The computing power host sends the remote computing power scheduling adaptation instruction set to the idle computing power module through the established communication connection channel, thereby utilizing the scheduling primitives to achieve the distribution and synchronization of cross-device scheduling logic, providing a control basis for the subsequent computing power topology normalization calibration processing executed on the computing power module side.

[0060] Preferably, during the process of performing computing topology normalization calibration on the idle computing module, the resource attribute calibration unit in the idle computing module performs semantic verification on the received remote computing scheduling adaptation instruction set. The resource attribute calibration unit, combined with locally detected real-time operating parameters such as clock frequency deviation, power consumption constraints, and temperature thresholds, dynamically corrects the scheduling weights in the remote computing scheduling adaptation instruction set to eliminate the physical dimension differences between the computing host and the idle computing module through semantic verification. This calibration action ultimately produces standardized computing topology mapping metadata. The standardized computing topology mapping metadata records the logical mapping bias of the physical computing unit in the global spatiotemporal context, ensuring high response consistency when the remote computing power is invoked by upper-layer applications.

[0061] Preferably, in the process of mapping computing units to remotely schedulable virtual resource nodes, the computing power virtualization mapping unit in the idle computing power module performs logical isolation and resource reorganization of computing units based on a hardware-assisted virtualization mechanism and utilizing the standardized computing power topology mapping metadata. The computing power virtualization mapping unit abstracts and maps complex physical entities into logically shaped virtual resource nodes by allocating logical handles to physical hardware and configuring shadow registers. The shadow registers provide real-time mirroring of the hardware state, ensuring that each virtual resource node possesses independent computing capability description information and operational status monitoring identifiers. This achieves the essential transformation of physical computing power resources into software-defined resources based on the logical handles, giving them the technical attributes of being remotely addressed and uniformly controlled by the computing power host.

[0062] Preferably, in the final integration stage of generating the global computing power resource pool, the computing host receives and mounts the logical handles of the virtual resource nodes reported by each of the idle computing power modules. The resource aggregation logic inside the computing host performs unified orchestration and addressing encapsulation of the virtual resource nodes corresponding to these logical handles with its own local computing power resources. By constructing a global resource addressing tree, the resource aggregation logic uses the global resource addressing tree to integrate heterogeneous resources distributed across different physical entities into a logically continuous global computing power resource pool. This process enables the upper-layer application to seamlessly call the virtual resource nodes distributed across different computing power modules when requesting computing resources, thereby achieving dynamic elastic expansion and efficient collaborative utilization of computing power resources at the logical level.

[0063] Optionally, the computing host is configured with a global computing power orchestration engine and a resource identifier allocator. Correspondingly, in step 2, the computing host maps the computing units in the computing power module to remotely schedulable virtual resource nodes based on virtual resource descriptors and integrates them with its own local computing power resources to generate a global computing power resource pool that can be invoked by upper-layer applications, including: The global computing power orchestration engine assigns globally unique computing power resource identifiers to virtual resource nodes and determines the local computing power topology addressing index of the computing power host's own local computing power resources. Based on globally unique computing resource identifiers and local computing topology addressing indexes, the resource identifier allocator generates a cross-source computing resource collaborative addressing table. This table is then used to encapsulate and address the local computing resources of virtual resource nodes and the computing host itself, thereby generating a global computing resource pool that can be invoked by upper-layer applications.

[0064] Preferably, during the process of integrating heterogeneous computing resources into a global computing resource pool by the computing host, the global computing orchestration engine configured within the computing host uses a resource scanning operator to locate the virtual resource node corresponding to each idle computing module by querying the master-slave device topology association table. The global computing orchestration engine extracts the resource affinity of the idle computing modules at the physical layer and, in conjunction with the logical link identifier assigned when establishing the communication association channel, performs global identity assignment processing. Through this global identity assignment processing, the global computing orchestration engine associates the physical layer resource affinity with the logical link identifier, assigning a globally unique computing resource identifier to each virtual resource node. This globally unique computing resource identifier, in essence, serves as a logical index of the virtual resource node within the global resource domain. It eliminates addressing conflicts that may arise during parallel scheduling of cross-source resources and provides a unique resource anchor point for the subsequent construction of a cross-source computing resource collaborative addressing table.

[0065] Preferably, while performing resource integration, the global computing power orchestration engine utilizes a topology-aware operator to perform deep topology attribute detection on the local computing power resources of the computing power host itself. The topology-aware operator maps the physical addresses of the local computing power resources of the computing power host to an index tree with a logical hierarchy by reading the memory management unit parameters of the computing power host kernel and the bus topology relationships in the hardware abstraction layer. The global computing power orchestration engine calls a spatiotemporal coordinate transformation operator to perform spatiotemporal coordinate transformation processing on the index tree, producing a local computing power topology addressing index. The local computing power topology addressing index records the topological distribution of the hardware resources inside the computing power host in the logical space and carries the hierarchical addressing logic transformed from the index tree. It serves as the base data corresponding to the globally unique computing power resource identifier, supporting subsequent homogenization addressing conversion.

[0066] Preferably, after receiving the globally unique computing resource identifier, the resource identifier allocator configured inside the computing power host performs feature encoding processing based on information entropy. The resource identifier allocator uses an identifier feature entropy encoding algorithm to analyze the distribution characteristics and redundancy of the globally unique computing resource identifier at the bit level, compresses it, and maps it into a fixed-length feature vector, thereby producing a unique addressing feature code for the computing power resource. By performing the identifier feature entropy encoding processing, complex resource description information is transformed into a unique addressing feature code for the computing power resource with high discriminative power. This not only reduces the space overhead of the cross-source computing resource collaborative addressing table during storage but also provides a standardized data carrier for fast convolutional alignment in the cross-device topology matching process.

[0067] Preferably, the resource identifier allocator utilizes the unique addressing feature code of the computing power resources to perform the construction of a dual-end computing power topology addressing benchmark. The resource identifier allocator uses a benchmark generation operator to extract the resource level attributes contained in the unique addressing feature code of the computing power resources. Based on the resource level attributes, it constructs a first spatial location description for the local computing power resources of the computing power host itself, and a second spatial location description for the computing units in the idle computing power modules. The resource identifier allocator associates and encapsulates the first and second spatial location descriptions to generate a dual-end computing power topology addressing benchmark. This dual-end computing power topology addressing benchmark, through the resource level attributes, logically establishes the relative mapping relationship between the master and slave computing power resources in a unified coordinate system, eliminating addressing logic mismatches caused by differences in the underlying hardware architecture.

[0068] Preferably, after obtaining the dual-end computing power topology addressing reference, the resource identifier allocator performs logical form synthesis processing based on homogeneous topology anchoring matching. The resource identifier allocator uses the local computing power topology addressing index as a background space and employs a convolutional matching algorithm to perform spatial alignment processing between the first and second spatial location descriptions in the dual-end computing power topology addressing reference and the background space. By identifying the homogeneity of the dual-end computing power topology addressing reference and the background space in terms of topological features, the resource identifier allocator associates resource handles physically distributed across different devices with the same addressing plane and generates a cross-source computing power resource collaborative addressing table based on the addressing plane. The cross-source computing power resource collaborative addressing table records the cross-source addressing mapping relationship between virtual resource nodes and physical hardware entities on the addressing plane globally, and is the core technology for realizing unified addressing and encapsulation of computing power resources.

[0069] Preferably, after generating the cross-source computing resource collaborative addressing table, the computing host utilizes a resource encapsulation operator to perform logical aggregation of global resources. The resource encapsulation operator injects the mapping relationships in the cross-source computing resource collaborative addressing table into the operating system's kernel resource scheduler by calling the underlying redirection logic of the Application Programming Interface (API). Through this unified addressing and encapsulation process, the computing host, based on the cross-source computing resource collaborative addressing table, merges the virtual resource nodes corresponding to each idle computing module with its own local computing resources, ultimately generating a global computing resource pool that can be invoked by upper-layer applications. This process, utilizing the cross-source computing resource collaborative addressing table, allows upper-layer applications to seamlessly access virtual resource nodes distributed on remote devices when requesting computing resources, much like accessing local memory space, thus truly realizing dynamic and elastic expansion of cross-source computing resources at the logical level.

[0070] Optionally, a cross-source computing resource collaborative addressing table is generated by the resource identifier allocator based on globally unique computing resource identifiers and local computing topology addressing indexes, including: The globally unique computing resource identifier is subjected to identifier feature entropy encoding to obtain a unique addressing feature code for the computing resource. Based on the unique addressing feature code for the computing resource, dual-end computing topology addressing benchmarks are generated for computing units in the computing host and computing module, respectively. Based on the dual-end computing topology addressing benchmarks corresponding to computing units in the computing host and computing module, a local computing topology addressing index is generated by same-source topology anchoring matching to generate a cross-source computing resource collaborative addressing table.

[0071] Preferably, during the processing of the resource identifier allocator based on the globally unique computing resource identifier, an identifier feature entropy encoding algorithm is used to analyze the bit distribution pattern of the globally unique computing resource identifier. The resource identifier allocator uses the identifier feature entropy encoding algorithm to calculate the probability of each bit appearing in the globally unique computing resource identifier, and calculates its information entropy feature based on the probability of each bit appearing. Based on the information entropy feature, the resource identifier allocator performs an asymmetric compression encoding operation to remove redundant data information from the globally unique computing resource identifier, thereby producing a unique addressing feature code that can uniquely represent the corresponding virtual resource node. By performing the identifier feature entropy encoding processing, the long-bit-width identifier is transformed into a unique addressing feature code with higher information density, which not only reduces the comparison bandwidth requirement during subsequent addressing but also improves the resource discrimination efficiency in a cross-device scheduling environment.

[0072] Preferably, after obtaining the unique addressing feature code of the computing power resource, the resource identifier allocator uses a reference generation operator to perform spatial coordinate mapping processing on the computing host and the computing units in the idle computing modules. The reference generation operator, based on the unique addressing feature code of the computing power resource, constructs a local addressing coordinate system for the computing host and a remote addressing coordinate system for the computing units in the idle computing modules, according to a preset topology dimension. The reference generation operator projects the addressing logic at both ends of the local and remote addressing coordinate systems into the logical space defined by the unique addressing feature code of the computing power resource, ultimately generating a dual-end computing power topology addressing reference. This dual-end computing power topology addressing reference, combining the local and remote addressing coordinate systems, provides a logical foundation for establishing a unified addressing reference system among heterogeneous devices.

[0073] Preferably, during the process of performing source-based topology anchoring matching, the resource identifier allocator uses an anchoring matching operator to perform a topology similarity comparison between the local computing power topology addressing index and the dual-end computing power topology addressing benchmark. The anchoring matching operator first identifies physical key nodes in the local computing power topology addressing index (e.g., setting the physical key nodes as cache mount points or memory channel access points) and defines these physical key nodes as local anchors. Subsequently, the anchoring matching operator locates logical anchors in the logical coordinate space that have functional consistency with the local anchors based on the topological relationships carried in the dual-end computing power topology addressing benchmark. The anchoring matching operator uses an anchoring processing method based on both physical and logical attributes to perform source alignment between the local anchors and the logical anchors, so that subsequent addressing translation no longer depends on fixed physical addresses, but is performed based on the relative association relationships of the topology structure.

[0074] Preferably, after acquiring the local anchor point and the logical anchor point, the resource identifier allocator uses the collaborative table construction logic to perform cross-source resource mapping item filling processing. The collaborative table construction logic embeds the addressing features associated with each globally unique computing power resource identifier into the addressing path covered by the local computing power topology addressing index, according to its corresponding dual-end computing power topology addressing benchmark. If addressing path overlap conflicts are detected during the filling process, the resource identifier allocator performs path redirection correction based on the priority weight associated with the unique addressing feature code of the computing power resource. The resource identifier allocator generates a cross-source computing power resource collaborative addressing table based on the priority weight after eliminating the addressing path overlap conflicts. The cross-source computing power resource collaborative addressing table clearly illustrates how each remote virtual resource node is transformed into a directly addressable logical entity through logical mapping relationships, realizing a logical closed loop in the master-slave device addressing space.

[0075] Preferably, after generating the cross-source computing resource collaborative addressing table, the computing host invokes the resource encapsulation operator to perform a unified encapsulation action on the virtual resource node and the computing host's own local computing resources. The resource encapsulation operator uses the cross-source computing resource collaborative addressing table as an index base to shield upper-layer applications from the heterogeneous attributes of the underlying physical links. In specific implementation, the resource encapsulation operator rewrites the access function interface of the underlying data link to convert resource requests for remote virtual resource nodes into routing addressing instructions pointing to the communication association channel in real time by retrieving the cross-source computing resource collaborative addressing table. The resource encapsulation operator uses the routing addressing instructions to place the virtual resource node and the computing host's own local computing resources at the same scheduling level, thereby generating a global computing resource pool that can be invoked by upper-layer applications.

[0076] Preferably, the generated global computing resource pool achieves atomic operations for computing resource addressing through the cross-source computing resource collaborative addressing table. When the upper-layer application issues a computing task request, the computing host responds to the task request. Based on the atomic operation, it does not need to perform underlying device probing again, but directly retrieves the cross-source computing resource collaborative addressing table to determine the physical execution entity corresponding to the unique addressing feature code of the computing resource. Through the topological homogeneity established by the dual-end computing topology addressing benchmark, the computing host uses the atomic operation to ensure the accuracy of topology awareness during task distribution, effectively reducing the additional addressing latency caused by cross-source access. This progressive processing logic, from identifier feature entropy encoding processing to the generation of the cross-source computing resource collaborative addressing table, ensures that heterogeneous computing modules can be integrated into the global computing environment in a logically native way.

[0077] like Figure 2 The diagram illustrates a dynamic computing power expansion system according to an embodiment of this application. The system includes a computing power host and a computing power module. A communication channel is established between the computing power host and the computing power module. The computing power module determines its own computing unit and obtains corresponding computing power parameters. These parameters are then encapsulated to generate a virtual resource descriptor, which is sent to the computing power host via the communication channel. The computing power host maps the computing units in the computing power module to remotely schedulable virtual resource nodes based on the virtual resource descriptors and integrates them with its own local computing power resources to generate a global computing power resource pool that can be invoked by upper-layer applications.

[0078] A computing power host is defined as a "data sovereignty device," serving as the sole physical carrier of a user's personal identity, core data, files, and operating system. At the hardware level, it typically manifests as a terminal device with a complete human-computer interaction unit (such as a screen, keyboard, and touchscreen), such as a thin and light laptop, a high-end tablet, or a smartphone. The computing power host has built-in basic computing power sufficient for everyday lightweight tasks; its core value lies in supporting users' data assets and operating habits, rather than extreme computing performance. In terms of technical logic, it acts as the master control terminal of the computing power aggregation protocol, receiving and authenticating external modules through communication channels. It maps the computing units of these modules into remotely schedulable virtual resource nodes and integrates them with its own local computing power resources into a unified global computing power resource pool, achieving completely transparent performance scheduling for applications.

[0079] In the architecture, the computing module is defined as an "on-demand expansion device." Its technical essence is contributing pure computational "muscle" (such as an NPU, GPU, or CPU) to the system without involving persistent storage of user's private data. The computing module does not need to run a full operating system; it only runs lightweight protocol slave firmware and does not have permanent storage units (the cache is cleared after the task is completed). Its hardware form factor is highly flexible, including but not limited to portable computing expansion docks, vehicle-mounted computing modules, wearable computing wristbands, or backpack-style computing modules. In system collaboration, it is responsible for detecting and determining the performance parameters of its own computing unit, encapsulating them as virtual resource descriptors, and actively declaring resources to the computing host through the computing power interaction channel. This allows users to achieve a gradual increase in system computing power by dynamically connecting more powerful modules without changing the host hardware.

[0080] For example, in a mobile office scenario, the computing power host, acting as a data sovereignty device, operates independently. This host houses the operating system, authentication information, and core database, utilizing its built-in computing power to perform routine logical operations and human-computer interaction tasks. Simultaneously, the host's resource monitoring module monitors the load rate of its local computing resources in real time. When an application initiates a high-load computing request (such as a generative AI image rendering task), and the host determines that the computing power required for the task exceeds its local computing power limit, the system triggers a dynamic computing power expansion mechanism.

[0081] The computing power module, acting as an on-demand expansion device, connects to the computing power host via a standardized computing power expansion interface. The connection management logic within the computing power host identifies this physical access and broadcasts a computing power module probe request at a preset interval. The computing power module responds to the probe request and sends back a computing power idle status beacon frame. Both parties then perform two-way authentication based on an asymmetric encryption algorithm. After successful authentication, a communication association channel for data exchange and a highly secure computing power interaction channel are established.

[0082] The firmware logic within the computing module performs hardware attribute sniffing on its integrated computing units (such as high-performance NPUs or GPUs) to obtain computing power parameters, including peak floating-point operation speed and memory bandwidth. The computing module then uses virtualization encapsulation logic to convert these computing power parameters into a structured data model, thereby generating a virtual resource descriptor. This virtual resource descriptor is accurately pushed to the virtual resource resolution engine of the computing host through the computing power interaction channel, serving as the metadata basis for heterogeneous resource mapping.

[0083] The virtual resource resolution engine of the computing host performs structured deconstruction on the received virtual resource descriptors and extracts the topological attributes of the computing units. The global computing orchestration engine then assigns a globally unique computing resource identifier to each external computing unit and uses the resource identifier allocator to generate a cross-source computing resource collaborative addressing table. Through same-source topology anchoring and matching, the computing host maps the physical computing units in the computing module to virtual resource nodes that can be remotely scheduled, and performs homogenization merging with the local computing resources to ultimately produce a global computing resource pool.

[0084] The system scheduler automatically allocates computing tasks exceeding the local capacity of the computing power host to the virtual resource nodes in the global computing power resource pool for execution. This scheduling process is transparent to upper-layer applications, which do not need to be aware of the differences in the distribution of underlying physical computing power. After the computing task is completed, the computing power module only sends the computing result data back to the computing power host. The computing power module does not have a permanent storage unit for user data, and immediately clears all intermediate data in the memory buffer after the task is completed, ensuring that data sovereignty always remains on the computing power host side.

[0085] When the system detects a further increase in external computing demand, it supports the access of new computing power modules with higher performance levels through the standardized computing power expansion interface. The system supports the dynamic addition and replacement of computing power modules, enabling a gradual improvement in the system's computing power capability by forming dynamic computing power aggregation relationships through the computing power aggregation protocol without replacing the computing power host hardware.

Claims

1. A method for dynamically expanding computing power, characterized in that, include: Step 1: Establish a communication channel between the computing power host and the computing power module. The computing power module determines its own computing unit and obtains the corresponding computing power parameters. It encapsulates the computing power parameters to generate a virtual resource descriptor and sends it to the computing power host through the communication channel. Step 2: The computing host maps the computing units in the computing module to virtual resource nodes that can be remotely scheduled based on the virtual resource descriptor and integrates them with its own local computing resources to generate a global computing resource pool that can be called by upper-layer applications.

2. The method according to claim 1, characterized in that, Step 1 also includes: identifying all electronic devices participating in the dynamic expansion of computing power, configuring the electronic devices to perform dynamic expansion of computing power roles, dividing the electronic devices into computing power hosts that act as master control devices and computing power modules that act as slave devices, and establishing a communication association channel between the computing power hosts that act as master control devices and the computing power modules that act as slave devices.

3. The method according to claim 1, characterized in that, Step 1 further includes: the computing power host broadcasts a computing power module detection request to the outside world according to a preset period, and receives the computing power idle status beacon frame returned by the computing power module to identify the computing power module in the idle state, so that the computing power module in the idle state can determine its own computing unit and obtain the corresponding computing power parameters.

4. The method according to claim 3, characterized in that, When the computing power host determines that a computing power module is in an idle state, it parses the computing power idle state beacon frame to obtain the module computing power ready identifier and idle time slot sequence, and determines whether the computing power module is in an idle state based on the module computing power ready identifier and idle time slot sequence.

5. The method according to claim 4, characterized in that, When the system is determined to be in an idle state, the computing power host generates a master-slave collaborative link establishment instruction and sends it to the computing power module. In response to receiving the link response confirmation frame returned by the computing power module, a communication association channel is established between the computing power host and the computing power module.

6. The method according to claim 4, characterized in that, Before the computing power module determines its own computing unit and obtains the corresponding computing power parameters, it includes: based on the communication association channel, the computing power host and the computing power module of the return link response confirmation frame perform two-way identity authentication to verify the trustworthiness of the communication between the computing power host and the computing power module of the return link response confirmation frame, obtain two-way identity authentication pass credentials, and establish a computing power interaction channel based on the two-way identity authentication pass credentials to send the virtual resource descriptor to the computing power host through the computing power interaction channel.

7. The method according to claim 4, characterized in that, Step 2: The computing host maps the computing units in the computing module to remotely schedulable virtual resource nodes based on virtual resource descriptors and integrates them with its own local computing resources to generate a global computing resource pool that can be invoked by upper-layer applications, including: The virtual resource parsing engine configured on the computing power host parses the virtual resource descriptor to obtain the computing power unit topology attributes, converts the computing power unit topology attributes into a remote computing power scheduling adaptation instruction set and sends it to the computing power module. The resource attribute calibration unit in the computing power module performs computing power topology normalization calibration processing on the remote computing power scheduling adaptation instruction set to obtain standardized computing power topology mapping metadata. The computing power virtualization mapping unit in the computing power module maps its computing unit into a virtual resource node that can be remotely scheduled according to the standardized computing power topology mapping metadata.

8. The method according to claim 4, characterized in that, The computing host is equipped with a global computing power orchestration engine and a resource identifier allocator. Correspondingly, in step 2, the computing host maps the computing units in the computing power module to remotely schedulable virtual resource nodes based on virtual resource descriptors and integrates them with its own local computing power resources to generate a global computing power resource pool that can be invoked by upper-layer applications, including: The global computing power orchestration engine assigns globally unique computing power resource identifiers to virtual resource nodes and determines the local computing power topology addressing index of the computing power host's own local computing power resources. Based on globally unique computing resource identifiers and local computing topology addressing indexes, the resource identifier allocator generates a cross-source computing resource collaborative addressing table. This table is then used to encapsulate and address the local computing resources of virtual resource nodes and the computing host itself, thereby generating a global computing resource pool that can be invoked by upper-layer applications.

9. The method according to claim 8, characterized in that, Based on globally unique computing resource identifiers and local computing topology addressing indexes, a cross-source computing resource collaborative addressing table is generated through the resource identifier allocator, including: The globally unique computing resource identifier is subjected to identifier feature entropy encoding to obtain a unique addressing feature code for the computing resource. Based on the unique addressing feature code for the computing resource, dual-end computing topology addressing benchmarks are generated for computing units in the computing host and computing module, respectively. Based on the dual-end computing topology addressing benchmarks corresponding to computing units in the computing host and computing module, a local computing topology addressing index is generated by same-source topology anchoring matching to generate a cross-source computing resource collaborative addressing table.

10. A computing power dynamic expansion system, characterized in that, include: The computing power host and computing power modules are connected by a communication channel. The computing power module determines its own computing unit and obtains the corresponding computing power parameters. It encapsulates the computing power parameters to generate a virtual resource descriptor and sends it to the computing power host through the communication channel. The computing power host maps the computing units in the computing power module to virtual resource nodes that can be remotely scheduled based on the virtual resource descriptor and integrates them with its own local computing power resources to generate a global computing power resource pool that can be called by upper-layer applications.

Images