Intelligent computing offloading method and system based on 6g network

Abstract

The application discloses an intelligent computing offloading method and system based on a 6G network, and belongs to the technical field of the 6G network.The technical problem to be solved by the application is how to slightly improve the existing network architecture, process and interface so that a terminal with limited computing power can support high-computing-power-demand services such as rendering and reasoning, and realize the landing of general computing services.The technical scheme adopted is as follows: the structure comprises the following steps: a new type of CM message and a new type of load container are defined, the AMF uses the load container type and the EPD field expansion value of the NAS message header to identify and route the CM message; the UE determines the computing task demand through a local or cloud LLM, and sends a computing offloading request to the AMF through a CM container of an uplink NAS transmission message; after identifying the CM message, the AMF sends the CM message to the CCF through a service interface; the CCF intelligently selects a computing power node according to the position and computing power demand of the UE, and returns the address of the computing power node to the UE through a downlink NAS transmission message; and the CCF is independently arranged or is arranged together with the SMF.

Description

Technical Field

[0001] This invention relates to the field of 6G network technology, specifically to an intelligent computing offloading method and system based on 6G networks. Background Technology

[0002] 6G networks are characterized by the convergence of communication, sensing, computing, and intelligence, providing capabilities in communication + AI, communication + sensing, and communication + computing. For converged communication and computing services, the computing power of the UE is limited by factors such as edge-side chips, storage, power consumption, and heat dissipation, making it unable to adequately support services with high computing power requirements such as rendering and inference. Therefore, a computing-network collaborative architecture has been proposed, introducing a Computing Control Function (CCF) to the core network. The UE sends a computing task offload request to the CCF, which then selects a suitable computing node based on the computing power requirement. The terminal establishes a computing session with the computing node to obtain the data processing results. Currently, converged communication and computing only has a basic framework and functional definitions, lacking specific implementation methods.

[0003] Given the current situation where 6G networks only have the framework and functional definition of general computing services and there is no implementation plan, how to slightly improve the existing network architecture, processes and interfaces so that terminals with limited computing power can support high computing power demand services such as rendering and inference, and realize the implementation of general computing services, is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] The technical objective of this invention is to provide an intelligent computing offloading method and system based on 6G networks to address the problem of how to slightly improve the existing network architecture, processes, and interfaces so that terminals with limited computing power can support high-computing-power-demand services such as rendering and inference, thereby realizing the implementation of general computing services.

[0005] The technical objective of this invention is achieved as follows: an intelligent computing offloading method based on a 6G network, the specific method of which is as follows:

[0006] Define a new type of CM (computation management) message and a new payload container type. AMF uses the payload container type and the extended protocol discriminator (EPD) field in the NAS message header to expand the values ​​for identification and routing of CM messages.

[0007] The UE determines the computing task requirements through local or cloud LLM, and sends the computing offload request to the AMF through the CM container that transmits the message via uplink NAS.

[0008] After AMF identifies the CM message, it sends the CM message to CCF through the service interface;

[0009] The CCF intelligently selects computing nodes based on the UE's location and computing power requirements, and returns the address of the computing node to the UE through downlink NAS transmission messages. The CCF can be set up independently or jointly with the SMF.

[0010] A computing session is established between the UE and the computing node. The management and mobility mechanisms of the computing session are consistent with those of the PDU session. The computing session identifier and the PDU session identifier are distinguished by their values.

[0011] As a preferred embodiment, the structure of NAS transmission messages is as follows:

[0012] Without changing the NAS transmission message structure, the possible values ​​of the EPD field are expanded, and one of the reserved values ​​is selected as the CM message.

[0013] Without changing the NAS transmission message structure, the possible values ​​for payload container type are expanded, and one of the reserved values ​​is selected to indicate that the payload container type is CM information.

[0014] More preferably, for NAS messages related to computing control, the EPD field value is specifically set as follows: the outer EPD value is still filled with "5GS MM message", and the inner plaintext NAS message EPD value is filled with "CM message", which is used by AMF to distinguish the NAS message type. At the same time, the second byte of the plaintext NAS message indicates "computing session identifier".

[0015] Among them, the computing control-related messages include computing unload requests, computing unload responses, and computing session management messages. The computing session management messages include request and response messages corresponding to the computing session establishment, modification, and release processes.

[0016] As a preferred method, the following methods can be used to distinguish between the computation session identifier and the PDU session identifier:

[0017] ① Reuse the PDU session identifier as the computing session identifier, and identify the computing session by EPD or / and payload container type + PDU session identifier;

[0018] ② By segmenting the PDU session identifier, different ranges of values ​​are used to distinguish them.

[0019] An intelligent computing offloading system based on a 6G network includes a UE, AMF, CCF, UPF, PCF, UDM, computing nodes, and LLM. The UE interacts with an app or large model to determine its on-device computing task requirements. The UE sends a UL NASTransport message to the AMF, which contains a computing offloading request NAS message. The UL NASTransport message has an EPD of "5G MM message" and a payload container type of "N1 CM information." The computing offloading request message is carried in the payload container, and its EPD is also "5G CM message." The AMF determines the information carried in the N1 CM container based on the payload container type and sends it to the CCF. After verification using the message EPD, the AMF calls the CCF's computing demand service operation through a service interface, sending a computing offloading request message to the CCF. This message also includes the UE's location information. The CCF intelligently determines the computing node based on the UE's computing task requirements and location information. The CCF returns a computing demand response message to the AMF, which includes the address of the computing node. The AMF then sends a DL NAS message to the UE. The Transport message carries the compute offloading response through the payload container, in which the compute offloading N1 CM message returns the compute node address; a compute session is established between the UE and the compute node, and the process of establishing the compute session is similar to that of establishing a PDU session.

[0020] As a preferred option, before sending the N1 CM container information to the CCF, the AMF uses the EPD of the CM message to verify the message type (CM).

[0021] Preferably, the computation unloading request message includes computation requirements, and the computation task requirements include at least computing power (TOPS) and latency (ms);

[0022] For inference tasks, computational requirements also include model requirements. The UE determines the computational requirements through local App, on-device large model, or cloud large model.

[0023] Preferably, the second byte of the compute unloading N1 CM message indicates the "compute session identifier", and the compute session identifier and PDU session identifier are distinguished by different ranges of values;

[0024] Preferably, the computing nodes are on-net UPF, MEC, and cloud nodes.

[0025] Even better, the CCF adds intelligent selection of computing nodes on the basis of SMF, and the computing session management and mobility mechanism are consistent with the PDU session. The CCF can be set up separately or together with SMF; among them, the computing session management includes the establishment, modification and release of computing sessions.

[0026] The intelligent computing offloading method and system based on 6G network of the present invention has the following advantages:

[0027] (i) This invention is based on the evolution of the 5G-A architecture and the adaptation modification of the NAS protocol to realize the implementation of general computing services;

[0028] (II) In order to reduce the impact on the existing network architecture, processes and interfaces and simplify the design, the management and mobility mechanism of the computing session and the PDU session are consistent. The main function of CCF is to add intelligent selection of computing nodes on the basis of SMF. CCF can be set up independently or jointly with SMF. A simpler implementation is to be set up jointly with SMF and reuse the existing N4 interface to interact with UPF. The structure of CM message is similar to that of SM message to support the management of computing session. Attached Figure Description

[0029] The invention will be further described below with reference to the accompanying drawings.

[0030] Appendix Figure 1 This is a schematic diagram of an intelligent computing offloading system based on a 6G network. Detailed Implementation

[0031] The intelligent computing offloading method and system based on 6G network of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] Example 1:

[0033] This embodiment provides an intelligent computing offloading method based on a 6G network, as detailed below:

[0034] S1. Define a new type of CM (computation management) message and a new payload container type. AMF uses the payload container type and the extended protocol discriminator (EPD) field of the NAS message header to expand the values ​​for identification and routing of CM messages.

[0035] S2, UE determines computing task requirements through local or cloud LLM, and sends computing offload requests to AMF through the CM container that transmits messages via uplink NAS;

[0036] After S3 and AMF identify the CM message, they send the CM message to CCF through the service interface;

[0037] S4, CCF intelligently selects computing power nodes based on the UE's location and computing power requirements, and returns the address of the computing power node to the UE through downlink NAS transmission messages. CCF can be set up independently or together with SMF.

[0038] A computing session is established between S5, UE and computing node. The management and mobility mechanism of the computing session is consistent with that of the PDU session. The computing session identifier and the PDU session identifier are distinguished by their values.

[0039] In the existing 3GPP standard (TS24.501), the structures of plaintext NAS messages and security-protected NAS messages are shown in Tables 1 and 2, respectively. Except for a few NAS messages before the security mode is activated, all NAS message transmissions between the UE and the core network are security-protected. Plaintext NAS messages are filled into the "Plain 5GS NAS message" field of the security-protected NAS message. For the 1-byte EPD (Extended Protocol Discriminator) field, the outer EPD is set to "5GS MM message" ("5GS mobility management messages"), while the inner EPD is filled with "5GS MM message" or "5GS SM message" ("5GS session management messages") according to the actual situation. For SM messages, the second byte of the plaintext NAS message is the PDU session identifier.

[0040] Table 1 Plaintext NAS Message Structure

[0041]

[0042] Table 2 NAS Message Structure for Security Protection

[0043]

[0044] The specific structure of the NAS transmission message in this embodiment is as follows:

[0045] Without changing the NAS transmission message structure, the values ​​of the EPD field are expanded, and one of the reserved values ​​is selected as CM message, as shown in Table 3. For NAS messages related to computing control, the specific values ​​of the EPD field are as follows: the outer EPD value is still filled with "5GS MM message", and the EPD value of the inner plaintext NAS message is filled with "CM message", which is used by AMF to distinguish the NAS message type. At the same time, the second byte of the plaintext NAS message indicates "computing session identifier". Among them, computing control related messages include computing offload requests, computing offload responses and computing session management messages. Computing session management messages include request and response messages corresponding to the establishment, modification and release of computing sessions. Computing session management includes the establishment, modification and release of computing sessions.

[0046] Table 3 EPD Values

[0047]

[0048] Without changing the NAS transmission message structure, the possible values ​​for payload container type are expanded. One of the reserved values ​​is selected to indicate that the payload container type is CM information, as shown in Table 4. The value 1010 is represented as N1 CM information.

[0049] Table 4 Payload container type values

[0050]

[0051] In this embodiment, the methods for distinguishing between the calculated session identifier and the PDU session identifier include the following:

[0052] ① Reuse the PDU session identifier as the computing session identifier, and identify the computing session by EPD or / and payload container type + PDU session identifier;

[0053] ② By segmenting the PDU session identifier and using different ranges of values ​​to distinguish them, for example, the first 4 bits are used as the calculation session and the last 4 bits are used as the PDU session identifier.

[0054] In order to maintain the mapping method of PDU session identifier in the air interface and enable the base station to recognize computing sessions and PDU sessions, it is recommended to use scheme ②.

[0055] Example 2:

[0056] As attached Figure 1As shown, this embodiment provides an intelligent computing offloading system based on a 6G network. The system includes a UE, AMF, CCF, UPF, PCF, UDM, computing nodes, and LLM. The UE interacts with an app or large model to determine the edge computing task requirements. The UE sends a UL NAS Transport message to the AMF, which contains a computing offloading request NAS message. The UL NAS Transport message has an EPD of "5G MM message" and a payload container type of "N1 CM information". The computing offloading request message is carried by the payload container, and its EPD is "5G CM message". The AMF determines the information carried in the N1 CM container based on the payload container type value and sends it to the CCF. After verification using the message EPD, the AMF calls the CCF's computing power requirement service operation through a service interface, sending a computing offloading request message to the CCF. This computing offloading request message also contains the UE's location information. The CCF intelligently determines the computing node based on the UE's computing task requirements and location information. The CCF returns a computing power requirement response message to the AMF, which contains the address of the computing node. The AMF sends a DL NAS message to the UE. The Transport message carries the compute offloading response through the payload container, in which the compute offloading N1 CM message returns the compute node address; a compute session is established between the UE and the compute node, and the process of establishing the compute session is similar to that of establishing a PDU session.

[0057] In this embodiment, before AMF sends N1 CM container information to CCF, it uses the EPD of the CM message to verify the message type (CM).

[0058] The computation unloading request message in this embodiment contains computation requirements, and the computation task requirements include at least computing power (TOPS) and latency (ms).

[0059] For inference tasks, computational requirements also include model requirements. The UE determines the computational requirements through local App, on-device large model, or cloud large model.

[0060] In this embodiment, the second byte of the compute offload N1 CM message indicates the "compute session identifier", and the compute session identifier and PDU session identifier are distinguished by different ranges of values;

[0061] In this embodiment, the computing nodes are on-network UPF, MEC, and cloud nodes.

[0062] In this embodiment, the CCF adds intelligent selection of computing nodes on the basis of SMF. The computing session management and mobility mechanism are consistent with the PDU session. The CCF can be set up independently or jointly with SMF. The computing session management includes the establishment, modification and release of computing sessions.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent computing offloading based on a 6G network, characterized in that, The method is as follows: Define a new type of CM message and a new payload container type. AMF uses the payload container type and the extended values ​​of the EPD field in the NAS message header to identify and route CM messages; where CM stands for Compute Management. The UE determines the computing task requirements through local or cloud LLM, and sends the computing offload request to the AMF through the CM container that transmits the message via uplink NAS. After AMF identifies the CM message, it sends the CM message to CCF through the service interface; The CCF intelligently selects computing nodes based on the UE's location and computing power requirements, and returns the address of the computing node to the UE through downlink NAS transmission messages. The CCF can be set up independently or jointly with the SMF. A computing session is established between the UE and the computing node. The management and mobility mechanisms of the computing session are consistent with those of the PDU session. The computing session identifier and the PDU session identifier are distinguished by their values. Specifically, for NAS messages related to computing control, the EPD field value is as follows: the outer EPD value is still filled with "5GS MM message", and the inner plaintext NAS message EPD value is filled with "CM message", which is used by AMF to distinguish the NAS message type. At the same time, the second byte of the plaintext NAS message indicates "computing session identifier". Among them, the computing control-related messages include computing unload requests, computing unload responses, and computing session management messages. The computing session management messages include request and response messages corresponding to the computing session establishment, modification, and release processes.

2. The intelligent computing offloading method based on a 6G network according to claim 1, characterized in that, The specific structure of NAS message transmission is as follows: Without changing the NAS transmission message structure, the possible values ​​of the EPD field are expanded, and one of the reserved values ​​is selected as the CM message. Without changing the NAS transmission message structure, the possible values ​​for payload container type are expanded, and one of the reserved values ​​is selected to indicate that the payload container type is CM information.

3. The intelligent computing offloading method based on a 6G network according to claim 1, characterized in that, The methods for distinguishing between the computed session identifier and the PDU session identifier include the following: ① Reuse the PDU session identifier as the computing session identifier, and identify the computing session by EPD or / and payload container type + PDU session identifier; ② By segmenting the PDU session identifier, different ranges of values ​​are used to distinguish them.

4. An intelligent computing offloading system based on a 6G network, characterized in that, The system includes a UE, AMF, CCF, UPF, PCF, UDM, computing nodes, and LLM. The UE interacts with the APP or large model to determine the edge computing task requirements. The UE sends a UL NAS Transport message to the AMF, which contains a computing offload request NAS message. The EPD of the UL NAS Transport message is "5G MM message," and the payload container type is "N1 CMinformation." The computing offload request message is carried through the payload container, and its EPD is "5G CM message." The AMF determines the information carried in the N1 CM container based on the payload container type value and sends it to the CCF. After verification using the message EPD, the AMF calls the CCF's computing demand service operation through a service interface, sending a computing offload request message to the CCF. This computing offload request message also contains the UE's location information. The CCF intelligently determines the computing node based on the UE's computing task requirements and location information. The CCF returns a computing demand response message to the AMF, which contains the address of the computing node. The AMF sends a DL NAS Transport message to the UE, through the payload... The container carries the compute unloading response, in which the compute node address is returned in the compute unloading N1 CM message; a compute session is established between the UE and the compute node, and the establishment process of the compute session is similar to that of the PDU session; where CM stands for compute management.

5. The intelligent computing offloading system based on a 6G network according to claim 4, characterized in that, Before sending N1 CM container information to CCF, AMF uses the EPD of the CM message to verify the message type.

6. The intelligent computing offloading system based on a 6G network according to claim 4, characterized in that, The compute unloading request message contains compute requirements, which include at least computing power and latency. For inference tasks, computational requirements also include model requirements. The UE determines the computational requirements through local App, on-device large model, or cloud large model.

7. The intelligent computing offloading system based on a 6G network according to claim 4, characterized in that, The second byte of the compute unload N1CM message indicates the "compute session identifier", which distinguishes the compute session identifier from the PDU session identifier by taking different ranges of values.

8. The intelligent computing offloading system based on a 6G network according to claim 4, characterized in that, The computing nodes are on-network UPF, MEC, and cloud nodes.

9. The intelligent computing offloading system based on a 6G network according to any one of claims 4 to 8, characterized in that, The CCF adds intelligent selection of computing nodes on top of the SMF. The computing session management and mobility mechanism are consistent with the PDU session. The CCF can be set up independently or jointly with the SMF. The computing session management includes the establishment, modification and release of computing sessions.

Images