System, program, and control method

Abstract

This application provides a system comprising: a resource management unit that manages the division of computing resources on an execution infrastructure into RAN occupancy domains that can only be used by RAN control functions controlling the RAN, AI occupancy domains that can only be used by AI processing functions (such as RAN control AI processing in RIC (RAN Intelligent Controller) and non-RAN control AI in MEC), and point domains that can be used by both the RAN control functions and the AI ​​processing functions; a resource allocation unit that allocates the computing resources on the execution infrastructure to the RAN control functions and the AI ​​processing functions, wherein if the amount of computing resources required by the RAN control functions is greater than the amount of computing resources required by the RAN occupancy domains, the resource allocation unit controls the allocation of the point domains to the RAN control functions preferentially; and an execution unit that executes the RAN control functions and the AI ​​processing functions on the execution infrastructure according to the amount of computing resources allocated by the resource allocation unit.

Description

Technical Field

[0001] This invention relates to systems, programs, and control methods. Background Technology

[0002] Patent Document 1 describes a wireless access network control device comprising a processor consisting of a Non-RT RIC (Non-Real Time RAN Intelligent Controller). The Non-RT RIC obtains the operating status of the multiple wireless access network nodes from a virtualized infrastructure that virtually manages a collection of multiple wireless access network nodes, obtains measured operating data from each wireless access network node, and, based on the operating status and operating data, issues operating guidelines related to the operation of each wireless access network node to at least one of the virtualized infrastructure and each wireless access network node.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: International Publication No. 2023-100385 Summary of the Invention

[0006] The technical problem that the invention aims to solve

[0007] Previously, technologies existed that allowed RAN service units and AI (Artificial Intelligence) application units to perform actions on the execution infrastructure. Currently, mobile networks are approaching the levels of water and electricity infrastructure; service interruptions would have a significant impact on people's lives. Therefore, traditional RAN base stations needed to implement very high SLAs (Service Level Agreements) and create redundancy mechanisms to ensure uninterrupted service. Consequently, only a fraction of the computing resources on the execution infrastructure could be used, resulting in a significant amount of unused computing resources. Furthermore, at night, although service demand decreases, maintaining the same service level as during the day would waste computing resources, resulting in a design that mirrors peak demand. While some improvements have been made by shutting down power to some computing resources at night to conserve power, this still constitutes a mechanism that wastes the power of advanced machines.

[0008] Solutions for solving technical problems

[0009] In the system described in this embodiment, for example, by running the RAN function on a high-performance GPU (Graphics Processing Unit) server instead of a general-purpose server, the remaining computing resources can be flexibly utilized in AI processing. Types of AI processing include AI processing associated with RAN control (sometimes referred to as RAN-controlled AI processing) and AI processing not associated with RAN control (sometimes referred to as non-RAN-controlled AI processing).

[0010] As an example of AI processing in RAN control, the RIC (Radio Access Network Intelligent Controller) can be cited. RICs are technologies that use AI to optimize RAN radio resources and automate RAN operations. RICs include Non-RT RICs and Near-RT RICs (Near-Real Time RICs). Non-RT RICs are sometimes also called Centralized RICs. Non-RT RICs reside within the SMO (Service Management and Orchestration) that manages and orchestrates the RAN. Non-RT RICs generate and notify RAN control-related policies and send information to Near-RT RICs. For example, a Non-RT RIC generates a learned model of RAN control by performing machine learning using data collected from the RAN and sends it to a Near-RT RIC. Near-RT RICs are sometimes also called Distributed RICs. Compared to Non-RT RICs, Near-RT RICs are located near RAN nodes (RUs (Radio Units), DUs (Distributed Units), and CUs (Central Units)) and perform RAN node control and resource control. Near-RT RICs perform more real-time processing compared to Non-RT RICs. For example, Near-RT RICs perform inference processing related to RAN control using learned models obtained from Non-RT RICs. RAN control AI processing is not limited to RICs.

[0011] Non-RAN controlled AI processing can be equated to so-called MEC (Multi-access Edge Computing) applications. Examples of non-RAN controlled AI processing include surveillance AI processing that determines the situation within the shooting range of an input captured image, and response AI processing that outputs answers to user queries, but it is not limited to these.

[0012] The concept of Service Level Agreements (SLAs), as described above, continues to exist, even as a means to flexibly utilize remaining computing resources in AI processing. For example, mobile network services have very high SLAs, such as 99.9999. However, considering user costs and service improvement cycles, MEC applications prioritize low execution costs and fast release cycles compared to high SLAs, unlike mobile network services. In the system described in this embodiment, for example, mechanisms can be provided to efficiently operate services with different SLAs and utilization strategies on the same infrastructure.

[0013] According to one embodiment of the present invention, a system is provided. The system may include: a resource management unit that manages the division of computing resources on an execution infrastructure into RAN occupancy domains that can only be used by RAN control functions controlling the RAN, AI occupancy domains that can only be used by AI processing functions performing AI processing, and point domains that can be used by both the RAN control functions and the AI ​​processing functions. The system may also include: a resource allocation unit that allocates computing resources on the execution infrastructure to the RAN control functions and the AI ​​processing functions; if the amount of computing resources required by the RAN control functions is greater than the amount of computing resources required by the RAN occupancy domains, the resource allocation unit controls the allocation of the point domains to the RAN control functions preferentially. The system may also include: an execution unit that executes the RAN control functions and the AI ​​processing functions on the execution infrastructure according to the amount of computing resources allocated by the resource allocation unit.

[0014] In the system, if the resource allocation unit has a first residual portion in the point domain that has not been allocated to the RAN control function, the resource allocation unit may allocate the first residual portion to the AI ​​processing function. The AI ​​processing function may include AI execution processing for performing AI and AI learning processing for learning AI. The resource allocation unit may allocate the AI ​​occupancy domain and the first residual portion to the AI ​​execution processing. If a second residual portion exists in the first residual portion that has not been allocated to the AI ​​execution processing, the second residual portion may be allocated to the AI ​​learning processing.

[0015] In any of the systems, the AI ​​execution processing may include RAN control AI execution processing associated with RAN control, and non-RAN control AI execution processing not associated with RAN control. In both the RAN control AI execution processing and the non-RAN control AI execution processing, the resource allocation unit may preferentially allocate the AI ​​occupancy domain and the first residual portion to the RAN control AI execution processing. The non-RAN control AI execution processing may include at least one of monitoring AI execution processing that determines the situation within the shooting range of the input captured image, and response AI execution processing that outputs an answer to an input user query.

[0016] In any of the systems, when at least a portion of the point domain is not utilized by the RAN control function, the resource allocation unit can determine, based on the needs of the AI ​​processing, to allocate the unused portion of the point domain to the AI ​​processing function, or to convert the hardware corresponding to the unused portion of the point domain into a power-saving state. The system may further include a state control unit that, when determining to convert the hardware corresponding to the unused portion of the point domain into a power-saving state, converts the hardware corresponding to the unused portion of the point domain into different power-saving states based on the utilization rate of the RAN occupied domain. If the utilization rate of the RAN occupied domain is lower than a predetermined first threshold, the state control unit can control the power supply to the hardware corresponding to the unused portion of the point domain to be disconnected; if the utilization rate of the RAN occupied domain is higher than the first threshold but lower than a second threshold higher than the first threshold, the state control unit controls the hardware corresponding to the unused portion of the point domain to be put into a sleep state.

[0017] According to one embodiment of the present invention, a program is provided for performing: a resource management step, which manages the computing resources on an execution infrastructure by dividing them into a RAN occupancy domain that can only be used by a RAN control function that controls the RAN, an AI occupancy domain that can only be used by an AI processing function that performs AI processing, and a point domain that can be used by both the RAN control function and the AI ​​processing function; a resource allocation step, which allocates the computing resources on the execution infrastructure to the RAN control function and the AI ​​processing function, and, if the amount of computing resources required by the RAN control function is greater than the amount of computing resources required by the RAN occupancy domain, controls the allocation of the point domain to the RAN control function; and an execution step, which executes the RAN control function and the AI ​​processing function on the execution infrastructure according to the amount of computing resources allocated in the resource allocation step.

[0018] According to one embodiment of the present invention, a control method executed by a computer is provided. The control method may include: a resource management step, which manages the computing resources on an execution infrastructure by dividing them into a RAN occupancy domain that can only be used by a RAN control function that controls the RAN, an AI occupancy domain that can only be used by an AI processing function that performs AI processing, and a point domain that can be used by both the RAN control function and the AI ​​processing function. The control method may also include: a resource allocation step, which allocates the computing resources on the execution infrastructure to the RAN control function and the AI ​​processing function; if the amount of computing resources required by the RAN control function is greater than the amount of computing resources required by the RAN occupancy domain, control is performed to preferentially allocate the point domain to the RAN control function. The control method may also include: an execution step, which executes the RAN control function and the AI ​​processing function on the execution infrastructure according to the amount of computing resources allocated in the resource allocation step.

[0019] It should be noted that the above summary of the invention does not list all the essential features of the invention. Furthermore, sub-combinations of these feature groups can also constitute an invention. Attached Figure Description

[0020] Figure 1 An example of system 10 is illustrated schematically.

[0021] Figure 2 This schematically illustrates an example of the configuration of a distributed infrastructure 200.

[0022] Figure 3 This is an explanatory diagram illustrating the computing resources 210 utilized by the distributed infrastructure 200 in vRAN 203 and AI processing 204.

[0023] Figure 4 This is an explanatory diagram used to illustrate the generation example of the pedestrian flow estimation model 330.

[0024] Figure 5 This is an explanatory diagram used to illustrate the generation example of the pedestrian flow estimation model 330.

[0025] Figure 6 This is an illustrative diagram used to explain the management of computing resources 210 in a distributed infrastructure 200.

[0026] Figure 7 This is an illustrative diagram used to explain the allocation of computing resources 210 in the distributed infrastructure 200.

[0027] Figure 8 This is an illustrative diagram used to explain the allocation of computing resources 210 in the distributed infrastructure 200.

[0028] Figure 9 This is an illustrative diagram used to explain the allocation of computing resources 210 in the distributed infrastructure 200.

[0029] Figure 10 This is an illustrative diagram used to explain the allocation of computing resources 210 in the distributed infrastructure 200.

[0030] Figure 11 This is an illustration diagram used to explain the power-saving features of the distributed infrastructure 200.

[0031] Figure 12 This is an illustration diagram used to explain the power-saving features of the distributed infrastructure 200.

[0032] Figure 13 This schematically illustrates an example of the functional configuration of a distributed infrastructure 200.

[0033] Figure 14 This is an illustrative diagram used to illustrate the architecture 400 in system 10.

[0034] Figure 15 This illustration shows an example of the hardware configuration of a computer 1200 that functions as a management infrastructure 100 or a distributed infrastructure 200. Detailed Implementation

[0035] The present invention will now be described through embodiments thereof; however, these embodiments are not intended to limit the invention as defined in the claims. Furthermore, the combinations of features described in the embodiments are not necessarily all necessary for the solution of the invention.

[0036] Figure 1An example of system 10 is illustrated schematically. System 10 includes a distributed infrastructure 200. System 10 may also include multiple distributed infrastructures 200. System 10 may also include a management infrastructure 100 that manages the multiple distributed infrastructures 200. In system 10 according to this embodiment, for example, the management infrastructure 100 and the multiple distributed infrastructures 200 can coordinate to perform control of RAN 310 and AI processing.

[0037] RAN310 can be a virtualized vRAN (Virtual RAN), and system 10 can perform vRAN control. RAN310 can also be a physical RAN, and system 10 can perform physical RAN control. In this embodiment, the case where RAN310 is a vRAN will be used as an example for explanation.

[0038] The AI ​​processing performed by System 10 may include RAN control AI processing (sometimes referred to as RAN_AI). The AI ​​processing performed by System 10 may include non-RAN control AI processing (sometimes referred to as non-RAN_AI).

[0039] The distributed infrastructure 200 can be a data center configured in various locations. The distributed infrastructure 200 can be constructed from multiple devices. The distributed infrastructure 200 can be implemented on a virtualized infrastructure based on multiple devices. The distributed infrastructure 200 can also be implemented by a single device. That is, the distributed infrastructure 200 can also be a distributed system.

[0040] Management infrastructure 100 can be a data center that manages multiple distributed infrastructures 200. Management infrastructure 100 can be configured across multiple devices. Management infrastructure 100 can also be implemented on a virtualized infrastructure based on multiple devices. Management infrastructure 100 can also be implemented by a single device. That is, management infrastructure 100 can also be a management device.

[0041] Sometimes, the management infrastructure 100 is referred to as the Core Brain, and the distributed infrastructure 200 as the Regional Brain. It should be noted that... Figure 1The example illustrates a scenario where a single layer of management infrastructure 100 is configured below management infrastructure 100, but this is not a limitation. Distributed infrastructure 200 can also be multi-layered. For instance, in a scenario where two layers of distributed infrastructure 200 are configured below management infrastructure 100, management infrastructure 100 is sometimes referred to as the Core Brain, the lower-level distributed infrastructure 200 as the Regional Brain, and the even lower-level distributed infrastructure 200 as the Sub-Regional Brain.

[0042] In the distributed infrastructure 200, one or more CPUs (Central Processing Units) can be configured. One or more GPUs can be configured in the distributed infrastructure 200. Multiple superchips connecting the CPUs and GPUs via interconnects can be configured in the distributed infrastructure 200. This interconnect can be memory-coherent, enabling high-frequency and low-latency interconnects. Thus, the distributed infrastructure 200 can have both CPU and GPU resources as computing resources.

[0043] Figure 2 This schematically illustrates an example of the configuration of distributed infrastructure 200. Distributed infrastructure 200 executes control of vRAN 203 and AI processing 204 on a virtualized infrastructure 202 implemented on HW (Hardware) 201. HW 201 may include CPU, GPU, memory, hard disk, and SSD, etc. Control of vRAN 203 includes control of the multiple radio base stations 300 constituting RAN 310, coordination with other distributed infrastructure components 200, etc. AI processing 204 may include RAN_AI 205 and non-RAN_AI 206. Distributed infrastructure 200 uses the computing resources available in vRAN 203 and AI processing 204 to execute vRAN 203 and AI processing 204.

[0044] Figure 3This is an explanatory diagram illustrating the computing resources 210 utilized by the distributed infrastructure 200 in vRAN 203 and AI processing 204. The distributed infrastructure 200 of this embodiment can predict the demand of the RAN 310 it manages and determine the amount of computing resources 210 allocated to vRAN 203 (sometimes referred to as vRAN-use computing resources). After determining the vRAN-use computing resources, the distributed infrastructure 200 can allocate the remaining computing resources 210 to AI processing 204 if there are any surplus resources. Therefore, while prioritizing services for the RAN 310 with higher SLAs, AI processing can be further performed even after RAN 310 control has been completed and there is still capacity remaining.

[0045] The distributed infrastructure 200 can predict the demand for RAN310 based on the actual utilization results of RAN310. For example, if the utilization of RAN310 is on an increasing trend, the distributed infrastructure 200 predicts that the demand for RAN310 will increase, and if the utilization of RAN310 is on a decreasing trend, the distributed infrastructure 200 predicts that the demand for RAN310 will decrease.

[0046] The distributed infrastructure 200 can predict the demand of RAN310 by integrating the predicted demands of each of the multiple radio base stations 300 that make up RAN310. The changes in the demand of RAN310 are strongly correlated with moving objects such as people and vehicles they are moving; therefore, by understanding human movement, the demand of RAN can be predicted.

[0047] Human movement exhibits periodicity. For example, in dormitory towns, people leave home in the morning for work and school, returning in the evening and at night. Outside of dormitory towns, each region also exhibits its own movement patterns. Regarding mobility, for instance, in Japan, especially as an island nation, fluctuations in movement by plane and ship, if observed as minor effects, are to some extent the same population movement, repeatedly decreasing in one area and increasing in another. Therefore, by combining day of the week, season, and human trends, predictive models of population flow can be created. However, irregular population flows are possible due to events, disasters, etc. Distributed infrastructure 200, for example, generates mechanisms that can predict the future demand of highly correlated RANs by constructing population flow models. These models utilize periodic population flow data based on time-series data of day of the week, time periods, and population trends, population flow data during events such as surrounding venues and fireworks displays, and change data within short-term processing windows under disaster conditions.

[0048] Figure 4This is an explanatory diagram illustrating an example of generating the pedestrian flow estimation model 330. The distributed infrastructure 200 can acquire pedestrian flow data 320 in the coverage area 302 for each of multiple wireless base stations 300, and use the pedestrian flow data 320 to generate the pedestrian flow estimation model 330.

[0049] The pedestrian flow data 320 may include pedestrian flow data 322 representing pedestrian flow at each date and time during normal times within the coverage area 302, excluding periods of special activity. The normal time period pedestrian flow data 322 may include the number of people at each location within the coverage area 302 for each time unit. The time unit may be seconds, minutes, or hours, etc. The normal time period pedestrian flow data 322 may also further include pedestrian movement direction, movement speed, etc.

[0050] The pedestrian flow data 320 may include activity-time pedestrian flow data 324 representing pedestrian flow during events within the coverage area 302. Activity-time pedestrian flow data 324 may include the number of people at each location within the coverage area 302 for each time unit during events such as concerts and fireworks displays. The time unit may be seconds, minutes, or other time units. Activity-time pedestrian flow data 324 may also include pedestrian movement direction and speed. The pedestrian flow data 320 may also include activity-time pedestrian flow data 324 for each type of event.

[0051] The pedestrian flow data 320 may include disaster-period pedestrian flow data 326 representing the flow of people within the coverage area 302 during a disaster. Disaster-period pedestrian flow data 326 may include the number of people at each location within the coverage area 302 for each time unit during the disaster. The time unit may be seconds, minutes, or other time units. Disaster-period pedestrian flow data 326 may also include the direction and speed of people's movement. The pedestrian flow data 320 may also include disaster-period pedestrian flow data 326 for each type of disaster.

[0052] The distributed infrastructure 200 can use pedestrian flow data 320, including pedestrian flow data 322 during normal times, to generate a pedestrian flow estimation model 330 that takes date and time as input and pedestrian flow as output. The distributed infrastructure 200 can use pedestrian flow data 320, including pedestrian flow data 322 during normal times and pedestrian flow data 324 during event times, to generate a pedestrian flow estimation model 330 that takes date and time and event times as input and pedestrian flow as output. The distributed infrastructure 200 can use pedestrian flow data 320, including pedestrian flow data 322 during normal times and pedestrian flow data 326 during disaster times, to generate a pedestrian flow estimation model 330 that takes date and time and disaster times as input and pedestrian flow as output. The distributed infrastructure 200 can use pedestrian flow data 320, including pedestrian flow data 322 during normal times, pedestrian flow data 324 during event times, and pedestrian flow data 326 during disaster times, to generate a pedestrian flow estimation model 330 that takes date and time, event times, and disaster times as input and pedestrian flow as output. Through these methods, in addition to assessing pedestrian flow in the absence of activity or disaster, it is possible to estimate pedestrian flow in the presence of activity and in the presence of disaster with relatively high accuracy.

[0053] For each of the multiple coverage areas 302 constituting RAN 310, the distributed infrastructure 200 can use a pedestrian flow estimation model 330 to estimate pedestrian flow and predict the demand for RAN 310 based on the estimation results. The distributed infrastructure 200 can predict the demand for RAN 310 based on the premise that areas with higher pedestrian flow have higher wireless communication loads. Alternatively, the distributed infrastructure 200 can use a pedestrian flow demand estimation model 330 to estimate the demand for RAN 310 based on the pedestrian flow estimated by the pedestrian flow estimation model 330. This pedestrian flow demand estimation model is generated through machine learning using actual results data on pedestrian flow and RAN demand within RAN 310, taking pedestrian flow within RAN 310 as input and RAN demand as output. The distributed infrastructure 200 can pre-store the pedestrian flow demand estimation model.

[0054] As mentioned above, the population flow estimation model 330 needs to pay close attention to the characteristics of the region. Therefore, by optimizing the region, a more efficient model can be created. However, in this case, to optimize the target region, it is necessary to collect a large amount of actual results data per target wireless base station 300 and update the model. Since such processing may result in a very high load, it is desirable to reduce the number of models by appropriately clustering the wireless base stations 300.

[0055] Figure 5 This is an explanatory diagram used to illustrate the generation example of the pedestrian flow estimation model 330. Figure 5In the example shown, the distributed infrastructure 200 classifies the multiple radio base stations 300 constituting RAN 310 into multiple clusters 340, and generates a pedestrian flow estimation model 330 for each of the multiple clusters 340. The distributed infrastructure 200 can generate the pedestrian flow estimation model 330 based on pedestrian flow data 320 within the coverage area 302 of the multiple radio base stations 300 included in each of the multiple clusters 340.

[0056] The distributed infrastructure 200 can use known clustering methods such as k-means to classify multiple wireless base stations 300. The clustering method used by the distributed infrastructure 200 is not limited to k-means; any method can be used.

[0057] The distributed infrastructure 200, for example, collects pedestrian flow data 320 from multiple wireless base stations 300 and uses the collected pedestrian flow data 320 to classify the multiple wireless base stations 300. The distributed infrastructure 200, for example, classifies the multiple wireless base stations 300 by aggregating wireless base stations 300 with high similarity in their pedestrian flow data 320. Thus, it is possible to appropriately classify multiple wireless base stations 300 with similar pedestrian flow trends.

[0058] The distributed infrastructure 200 classifies the multiple wireless base stations 300, for example, based on their locations. The distributed infrastructure 200 classifies the multiple wireless base stations 300, for example, by aggregating those that are close together. People in the coverage area 302 of wireless base stations 300 that are close together are more likely to show similar trends, thus enabling the appropriate classification of the multiple wireless base stations 300.

[0059] The distributed infrastructure 200 classifies multiple wireless base stations 300 based on their surrounding environment, for example. The distributed infrastructure 200 classifies the multiple wireless base stations 300 by aggregating wireless base stations 300 with similar environments such as roads and buildings surrounding them. Since the flow of people in the coverage area 302 of wireless base stations 300 with similar surrounding environments is highly likely to show similar trends, the multiple wireless base stations 300 can be appropriately classified as such.

[0060] The distributed infrastructure 200, for example, collects publishing data published by UEs 300 residing in each of multiple wireless base stations 300. It then performs embedding processing on the collected publishing data to generate base station vector data. Based on the base station vector data of the multiple wireless base stations 300, it classifies the multiple wireless base stations 300. The publishing data may be, for example, publishing data from SNS (Social Networking Service). The distributed infrastructure 200 classifies the multiple wireless base stations 300 by aggregating wireless base stations 300 with high similarity in their base station vector data. Similarity in the publishing data of the wireless base stations 300 refers to a high probability of similar pedestrian traffic or similarity at some scale; therefore, it is possible to appropriately classify multiple wireless base stations 300 with high correlation.

[0061] As described above, the distributed infrastructure 200 performs RAN control functions and AI processing functions, and therefore performs functions with different SLAs and priorities. Thus, the distributed infrastructure 200 can manage these functions based on their different SLAs and priorities, for example: high priority functions with very high SLAs (such as RAN control); medium priority functions with high SLAs (such as real-time processing like AI execution); and low priority functions with low SLAs (such as batch processing that can queue tasks, like AI learning processing).

[0062] Figure 6 This is an illustrative diagram used to explain the management of computing resources 210 in a distributed infrastructure 200. Figure 6 In the example shown, the distributed infrastructure 200 manages the computing resources 210 by dividing them into a RAN occupancy domain 212, which is only available for RAN control functions, an AI occupancy domain 214, which is only available for AI processing functions, and a point domain 216, which is available for both RAN control functions and AI processing functions.

[0063] Figure 7 , Figure 8 , Figure 9 , Figure 10 This is an illustrative diagram illustrating the allocation of computing resources 210 in the distributed infrastructure 200. Based on the predicted demand of RAN310, the distributed infrastructure 200 determines the amount of computing resources to allocate to vRAN203. If the computing resources required by vRAN203 exceed the RAN occupied domain 212, the distributed infrastructure 200 preferentially allocates point domain 216 to vRAN203. The distributed infrastructure 200 does not allocate AI occupied domain 214 to vRAN203.

[0064] like Figure 7As illustrated, the distributed infrastructure 200 may allocate up to the entirety of the RAN occupied domain 212 and point domain 216 to vRAN 203.

[0065] like Figure 8 As illustrated, if there is a residual portion 217 in point domain 216 that has not been allocated to vRAN 203, the distributed infrastructure 200 can allocate the residual portion 217 to AI processing 204. Thus, while maintaining the execution of AI processing 204 to a minimum, the amount of computing resources allocated to vRAN 203 can be increased according to the increased demand of RAN 310, and the allocation to AI processing 204 can be increased when the demand of RAN 310 is not high. After this allocation, if an increase in RAN demand is predicted, the distributed infrastructure 200 can allocate the residual portion 217 allocated to AI processing 204 to vRAN 203.

[0066] The distributed infrastructure 200 can also prioritize resource allocation to AI execution processing in both AI execution processing and AI learning processing. For example, the distributed infrastructure 200 allocates the AI ​​occupancy domain 214 and the residual portion 217 to AI execution processing. If there are residual portions 217 that have not been allocated to AI execution processing, those residual portions are allocated to AI learning processing. This allows for the priority execution of AI execution processing and AI learning processing that require more real-time processing.

[0067] In both RAN control AI execution processing associated with RAN control and non-RAN control AI execution processing not associated with RAN control, the distributed infrastructure 200 can prioritize resource allocation to RAN control AI execution processing. For example, the distributed infrastructure 200 allocates AI occupancy domain 214 and residual portion 217 to RAN control AI execution processing. If there are residual portions 217 that have not been allocated to RAN control AI execution processing, those residual portions are allocated to non-RAN control AI execution processing. This makes it possible to prioritize RAN control AI execution processing, which is considered to have a higher SLA, among both RAN control AI execution processing and non-RAN control AI execution processing.

[0068] like Figure 9As illustrated, if point domain 216 is not entirely allocated to vRAN 203, distributed infrastructure 200 can allocate point domain 216 entirely to AI processing 204. If AI occupancy domain 214 and point domain 216 are allocated to AI execution processing and there is a residual portion of point domain 216 not allocated to AI execution processing, distributed infrastructure 200 can allocate that residual portion to AI learning processing. If AI occupancy domain 214 and point domain 216 are allocated to RAN-controlled AI execution processing and there is a residual portion of point domain 216 not allocated to RAN-controlled AI execution processing, distributed infrastructure 200 can allocate that residual portion to non-RAN-controlled AI execution processing.

[0069] like Figure 10 As illustrated, even when vRAN203 requires less computing resources than RAN occupied domain 212, the distributed infrastructure 200 does not allocate RAN occupied domain 212 to AI processing 204. Therefore, if the demand for RAN310 increases, the remaining portion of RAN occupied domain 212 can be immediately allocated to vRAN203. Furthermore, assuming RAN occupied domain 212 could be allocated to AI processing 204, if AI processing 204 utilizes a portion of RAN occupied domain 212, an increase in the demand for RAN310 would necessitate immediately stopping AI processing 204. By managing the process by not allocating RAN occupied domain 212 to AI processing 204, such a situation can be prevented.

[0070] Electricity is a concern when advocating for the SDGs (Sustainable Development Goals). For example, there is a possibility of wasting significant output by unnecessarily starting and executing high-performance GPUs. Therefore, while ensuring efficient use of high-performance GPUs, it is important to establish mechanisms to stop unnecessary GPU use. In particular, demand for RAN310 decreases during nighttime hours, so reducing power waste there is crucial. However, scenarios involving a sharp increase in demand for RAN310 need to be considered.

[0071] Figure 11 , Figure 12 This is an illustrative diagram illustrating the power-saving function of the distributed infrastructure 200. The distributed infrastructure 200 can convert the hardware corresponding to the unused portion of point domain 216 into a power-saving state. For example, if point domain 216 is implemented using multiple GPUs, the distributed infrastructure 200 can convert one or more GPUs corresponding to the unused portion of point domain 216 into a power-saving state.

[0072] Power saving mode can be a state where the power is off. Power saving mode can be a hibernation state. Power saving mode can be a standby state. Power saving mode can also be a state where normally running applications are not in operation.

[0073] like Figure 11 As shown, when a portion of point domain 216 is unused, the distributed infrastructure 200 can switch the hardware corresponding to the unused portion of point domain 216 to a power-saving state. For example... Figure 12 As shown, when point domain 216 is completely unused, the distributed infrastructure 200 can switch the hardware corresponding to point domain 216 to a power-saving state. This appropriately reduces power consumption.

[0074] It should be noted that the distributed infrastructure 200 can also switch the hardware corresponding to the unused portion of point domain 216 to different power-saving states based on the allocation rate of vRAN 203 to RAN occupancy domain 212. For example, if the allocation rate of RAN occupancy domain 212 is lower than a first threshold, the distributed infrastructure 200 controls the power supply to the hardware corresponding to the unused portion of point domain 216. If the allocation rate of RAN occupancy domain 212 is higher than the first threshold but lower than a second threshold (which is higher than the first threshold), the distributed infrastructure 200 controls the hardware corresponding to the unused portion of point domain 216 to enter a sleep state. Thus, when the load on vRAN 203 is low and RAN occupancy domain 212 is sufficient, disconnecting the power supply can improve power saving. Furthermore, when the load on vRAN 203 increases to some extent, and the possibility of exceeding RAN occupancy domain 212 increases, switching to a sleep state allows for rapid allocation of point domain 216 when the load on vRAN 203 becomes high.

[0075] Figure 13 The diagram illustrates an example of the functional configuration of a distributed infrastructure 200. The distributed infrastructure 200 includes a storage unit 222, a data acquisition unit 224, a classification unit 226, a model generation unit 228, a demand forecasting unit 230, a forecast result acquisition unit 232, an AI management unit 234, a resource allocation unit 236, an execution unit 238, a resource management unit 240, and a state control unit 242. It should be noted that the distributed infrastructure 200 does not necessarily need to possess all of these components.

[0076] The data acquisition unit 224 acquires various types of data. The data acquisition unit 224 stores the acquired data in the storage unit 222.

[0077] The data acquisition unit 224 acquires, for example, data about its subordinate wireless base station 300. The data acquisition unit 224 can acquire location data of the wireless base station 300. The location data of the wireless base station 300 may, for example, represent the latitude and longitude of the wireless base station 300. The data acquisition unit 224 can acquire surrounding environment data representing the surrounding environment of the wireless base station 300. The surrounding environment data may, for example, be three-dimensional map data of the area surrounding the wireless base station 300.

[0078] The data acquisition unit 224 can acquire actual result data of the wireless communication service provided by the wireless base station 300. The actual result data may include data from the UE 30 residing in the wireless base station 300. The actual result data may include the communication traffic of the UE 30 residing in the wireless base station 300. The data acquisition unit 224 can acquire publication data published by the UE 30 residing in the wireless base station 300.

[0079] The data acquisition unit 224 can acquire pedestrian flow data 320 within the coverage area 302 of the wireless base station 300. The data acquisition unit 224 can receive pedestrian flow data 320 from external sources. The data acquisition unit 224 can acquire pedestrian flow data 320 generated in the distributed infrastructure 200. The pedestrian flow data 320 may include pedestrian flow data 322 during normal times. The pedestrian flow data 320 may include pedestrian flow data 324 during activity periods. The pedestrian flow data 320 may include pedestrian flow data 326 during disaster periods.

[0080] Classification unit 226 classifies multiple wireless base stations 300 into multiple clusters. Classification unit 226 can use known clustering methods such as k-means to classify the multiple wireless base stations 300. The clustering method used by distributed infrastructure 200 is not limited to k-means; any method can be used.

[0081] The classification unit 226, for example, uses pedestrian flow data 320 from multiple wireless base stations 300 to classify the multiple wireless base stations 300. The classification unit 226, for example, classifies the multiple wireless base stations 300 by aggregating wireless base stations 300 with high similarity to the pedestrian flow data 320.

[0082] Classification unit 226, for example, classifies multiple wireless base stations 300 based on location data. Distributed infrastructure 200, for example, classifies multiple wireless base stations 300 by aggregating nearby wireless base stations 300.

[0083] The classification unit 226, for example, classifies multiple wireless base stations 300 based on surrounding environmental data. The classification unit 226, for example, classifies multiple wireless base stations 300 by aggregating information on wireless base stations 300 with similar surrounding environments such as roads and buildings.

[0084] For example, for each of the multiple wireless base stations 300, the classification unit 226 generates base station vector data by embedding the publication data published by the UE 30 located within the range of the wireless base station 300, and classifies the multiple wireless base stations 300 based on the base station vector data of the multiple wireless base stations 300. For example, the classification unit 226 classifies the multiple wireless base stations 300 by aggregating wireless base stations 300 with high similarity of base station vector data.

[0085] The model generation unit 228 generates a pedestrian flow estimation model 330. The model generation unit 228 stores the generated pedestrian flow estimation model 330 in the storage unit 222. The model generation unit 228 can generate a pedestrian flow estimation model 330 for each of the multiple wireless base stations 300 by performing machine learning using pedestrian flow data 320.

[0086] The model generation unit 228 can use pedestrian flow data 320, including pedestrian flow data 322 during normal times, to generate a pedestrian flow estimation model 330 that takes date and time as input and pedestrian flow as output. The model generation unit 228 can also use pedestrian flow data 320, including pedestrian flow data 322 during normal times and pedestrian flow data 324 during event times, to generate a pedestrian flow estimation model 330 that takes date, time, and event as input and pedestrian flow as output. The model generation unit 228 can also use pedestrian flow data 320, including pedestrian flow data 322 during normal times and pedestrian flow data 326 during disaster times, to generate a pedestrian flow estimation model 330 that takes date, time, and disaster as input and pedestrian flow as output. Finally, the model generation unit 228 can use pedestrian flow data 320, including pedestrian flow data 322 during normal times, pedestrian flow data 324 during event times, and pedestrian flow data 326 during disaster times, to generate a pedestrian flow estimation model 330 that takes date, time, event, and disaster as input and pedestrian flow as output.

[0087] The model generation unit 228 can also perform machine learning on each of the multiple clusters classified by the classification unit 226, using the pedestrian flow data 320 within the coverage area 302 of the multiple wireless base stations 300 included in the cluster, to generate a pedestrian flow estimation model 330.

[0088] The demand forecasting unit 230 can use a pedestrian flow estimation model 330 to estimate pedestrian flow for each of the multiple coverage areas 302 constituting RAN 310, and predict the demand for RAN 310 based on the estimation results. The demand forecasting unit 230 can predict the demand for RAN 310 based on the premise that wireless communication load is higher in areas with more pedestrian flow. The demand forecasting unit 230 can use a pedestrian flow demand estimation model 330 to estimate the demand for RAN 310 based on the pedestrian flow estimated by the pedestrian flow estimation model 330. This pedestrian flow demand estimation model is generated through machine learning using actual results data on pedestrian flow and RAN demand within RAN 310, taking pedestrian flow within RAN 310 as input and RAN demand as output. The pedestrian flow demand estimation model can be stored in a pre-storage unit 222.

[0089] The forecast result acquisition unit 232 acquires the forecast result after forecasting the demand of RAN310. The forecast result acquisition unit 232 can acquire the forecast result of the demand of RAN310 predicted by the demand forecast unit 230. The forecast result acquisition unit 232 can also receive the forecast result of the demand of RAN310 from an external source.

[0090] AI Management Department 234 manages AI processing. For example, AI Management Department 234 acquires and manages AI processing execution requests that request the execution of AI processing.

[0091] The resource allocation unit 236 determines the amount of computing resources allocated to the distributed infrastructure 200 for RAN control functions and the amount of computing resources allocated to the distributed infrastructure 200 for AI processing functions, and performs allocation for the RAN control functions and AI processing functions. The resource allocation unit 236 can determine the amount of computing resources allocated to the distributed infrastructure 200 for RAN control functions and the amount of computing resources allocated to the distributed infrastructure 200 for AI processing functions based on the predicted demand of RAN 310 obtained by the prediction result acquisition unit 232, and perform allocation for the RAN control functions and AI processing functions.

[0092] The resource allocation unit 236 can prioritize allocating computing resources to the RAN control function among the RAN control function and the AI ​​processing function. For example, after determining the amount of computing resources to be allocated to the RAN control function based on the predicted demand of the system 10, the resource allocation unit 236 may allocate the remaining computing resources to the AI ​​processing function if there are any surplus resources.

[0093] The execution unit 238 performs RAN control functions and AI processing functions on the distributed infrastructure 200 based on the amount of computing resources allocated by the resource allocation unit 236.

[0094] Resource Management Unit 240 manages the computing resources on the distributed infrastructure 200 by dividing them into RAN occupancy domains that are only available for RAN control functions, AI occupancy domains that are only available for AI processing functions, and point domains that are available for both RAN control functions and AI processing functions.

[0095] Resource allocation unit 236 can allocate RAN occupied domains to RAN control functions. Resource allocation unit 236 can allocate AI occupied domains to AI processing functions. Resource allocation unit 236 can prioritize allocating point domains to RAN control functions over AI processing functions. For example, if the RAN control function requires more computing resources than the RAN occupied domains, resource allocation unit 236 will prioritize allocating point domains to the RAN control function.

[0096] If there is a residual portion (sometimes referred to as the first residual portion) in a point domain that has not been allocated to RAN control functions, the resource allocation unit 236 may allocate the first residual portion to the AI ​​processing function. If the entire point domain has not been allocated to RAN control functions, the entire point domain may be allocated to the AI ​​processing function as the first residual portion. After allocating part or all of a point domain to the AI ​​processing function, if an increase in demand for RAN310 is predicted, the resource allocation unit 236 may reassign part or all of the point domain allocated to the AI ​​processing function to the RAN control function.

[0097] As described above, the AI ​​processing function may include AI execution processing and AI learning processing. When the resource allocation unit 236 allocates the AI ​​occupied domain and the first residual portion to the AI ​​execution processing, and there is a residual portion (sometimes referred to as a second residual portion) in the first residual portion that has not been allocated to the AI ​​execution processing, it may allocate the second residual portion to the AI ​​learning processing.

[0098] As described above, AI execution processing can include RAN-controlled AI execution processing and non-RAN-controlled AI execution processing. In both RAN-controlled and non-RAN-controlled AI execution processing, the resource allocation unit 236 can preferentially allocate the AI ​​occupied domain and the first residual portion to the RAN-controlled AI execution processing.

[0099] If at least a portion of the point domain is not allocated to the RAN control function, the resource allocation unit 236 may, based on the needs of AI processing, determine whether to allocate the portion of the point domain not allocated to the RAN control function to the AI ​​processing function, or to convert the hardware corresponding to the portion of the point domain not allocated to the RAN control function to a power-saving state.

[0100] Resource allocation unit 236 determines the AI ​​processing requirement, for example, based on the status of AI processing execution requests managed by AI management unit 234. Resource allocation unit 236, for example, if at least a portion of the point domain is not allocated to the RAN control function, determines the amount of computing resources allocated to the AI ​​processing function in the point domain based on the AI ​​processing requirement, and if there is a surplus, determines to switch the hardware corresponding to that surplus to a power-saving state.

[0101] When the status control unit 242 determines that the hardware corresponding to the unused portion of the point domain will be converted to a power-saving state by the resource allocation unit 236, it can convert the hardware corresponding to the unused portion of the point domain to different power-saving states according to the allocation rate of RAN control functions to the RAN occupancy domain. As described above, the power-saving state can be a state where the power is turned off. The power-saving state can be a sleep state. The power-saving state can be a standby state. The power-saving state can also be a state where an application that is normally started is not running.

[0102] When the allocation rate of the RAN occupancy domain is lower than a predetermined first threshold, the status control unit 242 controls to disconnect the power supply of the hardware corresponding to the unused portion of the point domain. When the allocation rate of the RAN occupancy domain is higher than the first threshold but lower than a second threshold higher than the first threshold, the status control unit 242 controls to put the hardware corresponding to the unused portion of the point domain into a sleep state.

[0103] Figure 14 This is an illustrative diagram illustrating the architecture 400 in system 10. The distributed infrastructure 200 clusters its multiple wireless base stations 300 based on actual utilization data of RAN310, population flow data, activity data, disaster data, and data from other regions in the area corresponding to RAN310. It then selects a model creation region and generates an AI model (such as a population flow estimation model) for that region. The distributed infrastructure 200 uses the AI ​​model to generate predicted demand data for RAN310. Furthermore, the distributed infrastructure 200 collects AI demands and assigns priorities to AI processing execution requests. The distributed infrastructure 200 registers low-priority AI processing execution requests to a queue.

[0104] Based on these, the distributed infrastructure 200 allocates resources such as RAN control functions from the RAN occupancy domain and point domain. According to the priority of AI processing execution requests, resources are allocated from the AI ​​occupancy domain, and if any remainder exists in the point domain, from the remainder, to AI processes with higher priority. After this allocation, if any remainder still exists in the point domain, that remainder is allocated to AI processes registered in the queue.

[0105] In the event of an abnormal situation such as a sharp increase in demand for RAN310, the distributed infrastructure 200 can also perform emergency allocation processing for abnormal periods. Such abnormal situations can be detected using pre-generated and stored abnormal period prediction models. For example, by using actual utilization data of abnormal periods, an abnormal period prediction model can be generated that can predict the occurrence of abnormal situations based on actual utilization data. Similarly, by using pedestrian flow data of abnormal periods, an abnormal period prediction model can be generated that can predict the occurrence of abnormal situations based on pedestrian flow data.

[0106] In the event of an abnormal situation, such as detecting the idle resources of the nationwide RAN, the distributed infrastructure 200 can control the use of idle resources corresponding to the abnormal situation if idle resources exist. If no idle resources exist, the distributed infrastructure 200 can control the use of its own or other distributed infrastructure 200's point domains as resources corresponding to the abnormal situation. During the execution of these actions, the distributed infrastructure 200 determines whether there are any unused resources in its own computing resources. If unused resources are determined to exist, it can control the hardware corresponding to the unused resources to switch to a power-saving state.

[0107] Figure 15 This schematically illustrates an example of the hardware configuration of a computer 1200 that functions as a management infrastructure 100 or a distributed infrastructure 200. A program installed on the computer 1200 enables the computer 1200 to function as one or more "parts" of the apparatus according to this embodiment, or enables the computer 1200 to perform operations associated with the apparatus according to this embodiment or the one or more "parts," and / or enables the computer 1200 to perform the processes according to this embodiment or the steps of those processes. To enable the computer 1200 to perform specified operations associated with several or all of the blocks in the flowcharts and block diagrams described in this specification, such a program may be executed by the CPU 1212.

[0108] The computer 1200 based on this embodiment includes a CPU 1212, a GPU 1213, a RAM 1214, and a graphics controller 1216, which are interconnected via a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive 1226, and an IC card driver, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive 1226 can be a DVD-ROM drive or a DVD-RAM drive, etc. The storage device 1224 can be a hard disk drive or a solid-state drive, etc. The computer 1200 also includes conventional input / output units such as a ROM 1230 and a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0109] CPU 1212 operates according to the program stored in ROM 1230 and RAM 1214, thereby controlling each unit. Graphics controller 1216 obtains image data generated by CPU 1212 from the frame buffer provided in RAM 1214 or from itself, and displays the image data on display device 1218.

[0110] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 within the computer 1200. The DVD drive 1226 reads programs or data from the DVD-ROM 1227, etc., and provides them to the storage device 1224. The IC card driver reads programs and data from the IC card and / or writes programs and data to the IC card.

[0111] ROM 1230 stores boot programs and / or programs that depend on the hardware of computer 1200, which are executed by computer 1200 at startup. Input / output chip 1240 can also connect various input / output units to input / output controller 1220 via USB port, parallel port, serial port, keyboard port, mouse port, etc.

[0112] The program is provided by a computer-readable storage medium such as a DVD-ROM 1227 or an IC card. The program is read from the computer-readable storage medium, installed in a storage device 1224, RAM 1214, or ROM 1230 (also examples of computer-readable storage media), and executed by the CPU 1212. The information processing described within this program is read by the computer 1200, resulting in cooperation between the program and the aforementioned types of hardware resources. The apparatus or method can be configured to perform information manipulation or processing according to the use of the computer 1200.

[0113] For example, when communication is performed between computer 1200 and external devices, CPU 1212 can execute a communication program loaded in RAM 1214, and perform communication processing on communication interface 1222 based on the processing described in the communication program. Under the control of CPU 1212, communication interface 1222 reads transmission data stored in the transmission buffer area provided in the recording medium such as RAM 1214, storage device 1224, DVD-ROM 1227, or IC card, and sends the read transmission data to the network, or writes received data received from the network into the receive buffer area provided on the recording medium, etc.

[0114] Furthermore, the CPU 1212 can read all or a portion of files or databases stored in external recording media such as storage devices 1224, DVD drives 1226 (DVD-ROM 1227), and IC cards from the RAM 1214, and perform various types of processing on the data in the RAM 1214. The CPU 1212 can then write the processed data back to the external recording medium.

[0115] Various types of information, such as programs, data, tables, and databases, can be stored in the recording medium and processed. The CPU 1212 can perform various types of processing on data read from the RAM 1214, including various types of operations specified by the program's instruction sequence as described throughout this disclosure, information processing, conditional judgment, conditional branching, unconditional branching, information retrieval / replacement, etc., and write the results back to the RAM 1214. Furthermore, the CPU 1212 can retrieve information from files, databases, etc., within the recording medium. For example, when multiple entries with attribute values ​​of a first attribute each associated with the attribute value of a second attribute are stored in the recording medium, the CPU 1212 can retrieve from these multiple entries an entry that matches the condition specified with the attribute value of the first attribute, read the attribute value of the second attribute stored in that entry, and obtain the attribute value of the second attribute associated with the first attribute that satisfies the predetermined condition thereon.

[0116] The programs or software modules described above can be stored on or near computer-readable storage media on computer 1200. Alternatively, a recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can be used as a computer-readable storage medium, thereby providing the program to computer 1200 via the network.

[0117] In this embodiment, the blocks in the flowcharts and block diagrams may represent steps in the process of performing an operation or "parts" of a device that performs the operation. The specified steps and "parts" may be installed by dedicated circuitry, programmable circuitry supplied along with computer-readable instructions stored on a computer-readable storage medium, and / or a processor supplied along with computer-readable instructions stored on a computer-readable storage medium. Dedicated circuitry may include digital and / or analog hardware circuitry, and may also include integrated circuits (ICs) and / or discrete circuitry. Programmable circuitry may include reconfigurable hardware circuitry, including, for example, AND gates, OR gates, XOR gates, NAND gates, NOR gates, and other logic operations, flip-flops, registers, and memory elements such as field-programmable gate arrays (FPGAs) and programmable logic arrays (PLAs).

[0118] Computer-readable storage media can include any tangible device capable of storing instructions executable by a suitable device. As a result, a computer-readable storage medium having instructions stored therein comprises an instruction executable for creating units for performing operations specified in a flowchart or block diagram. Examples of computer-readable storage media include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable storage media include floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), optical disc read-only memory (CD-ROM), digital multipurpose disc (DVD), Blu-ray disc, memory stick, integrated circuit card, etc.

[0119] Computer-readable instructions may include any of the source code or object code described in any combination of one or more programming languages, including assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or object-oriented programming languages ​​such as Smalltalk (registered trademark), JAVA (registered trademark), C++, and conventional procedural languages ​​such as the "C" programming language or similar programming languages.

[0120] For the processor of a programmable data processing device, such as a computer, or for a programmable circuit to execute computer-readable instructions in order to generate units for performing operations specified in a flowchart or block diagram, these computer-readable instructions are provided via a local area network (LAN), a wide area network (WAN) such as the Internet, to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device. Here, a computer can be a PC (personal computer), a tablet computer, a smartphone, a workstation, a server computer, a general-purpose computer, or a special-purpose computer, or it can be a computer system connecting multiple computers. Such a computer system connecting multiple computers is also called a distributed computing system, which is a broad definition of a computer. In a distributed computing system, each computer executes a part of a program, transferring data between computers as needed, thereby enabling multiple computers to collectively execute the program.

[0121] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, and microcontrollers. A computer can have one or more processors. In a multiprocessor system with multiple processors, each processor executes a portion of the program, transferring data between processors as needed, thus allowing multiple processors to collectively execute the program. For example, in multitasking, multiple processors execute a portion of each task by switching tasks according to time slices. In this case, which part of the program each processor executes is dynamically changing. Which part of the program each processor executes can also be determined by being aware of the multiprocessor programming.

[0122] The present invention has been described above using embodiments; however, the scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. It is obvious, as described in the claims, that forms with such modifications or improvements can be included within the scope of the present invention.

[0123] It is important to note that the execution order of actions, processes, steps, and stages in the apparatus, systems, programs, and methods shown in the claims, description, and drawings can be arbitrary unless explicitly stated as "before" or "in advance," and the output of a preceding process is not used in a subsequent process. Regarding the flow of actions in the claims, description, and drawings, even if terms such as "firstly," "next," etc., are used for ease of explanation, this does not imply that the actions must be performed in that specific order.

[0124] Explanation of reference numerals in the attached figures

[0125] 10: System; 30: UE; 100: Management Infrastructure; 200: Distributed Infrastructure; 201: HW; 202: Virtualization Infrastructure; 203: vRAN; 204: AI Processing; 205: RAN_AI; 206: Non-RAN_AI; 210: Computing Resources; 212: RAN Occupied Domain; 214: AI Occupied Domain; 216: Point Domain; 217: Residual Part; 222: Storage Department; 224: Data Acquisition Department; 226: Classification Department; 228: Model Generation Department; 230: Demand Forecasting Department; 232: Forecast Result Acquisition Department; 234: AI Management Department; 236: Resource Allocation Department; 238: Execution Department; 240: Resource Management Department; 242: State Control Department 300: Wireless base station; 302: Coverage area; 310: RAN; 320: People flow data; 322: Normal time people flow data; 324: Activity time people flow data; 326: Disaster time people flow data; 330: People flow estimation model; 340: Cluster; 400: Architecture; 1200: Computer; 1210: Host controller; 1212: CPU; 1213: GPU; 1214: RAM; 1216: Graphics controller; 1218: Display device; 1220: Input / output controller; 1222: Communication interface; 1224: Storage device; 1226: DVD drive; 1227: DVD-ROM; 1230: ROM; 1240: Input / output chip.

Claims

1. A system that possesses: The Resource Management Department manages the computing resources on an execution infrastructure by dividing them into RAN Occupancy Domains that can only be used by RAN Control Functions (which control the RAN (Radio Access Network), AI Occupancy Domains that can only be used by AI Processing Functions (which perform AI (Artificial Intelligence) processing), and Point Domains that can be used by both the RAN Control Functions and the AI ​​Processing Functions. The resource allocation unit allocates computing resources on the execution infrastructure to the RAN control function and the AI ​​processing function. When the amount of computing resources required by the RAN control function is greater than the amount of computing resources occupied by the RAN domain, the resource allocation unit controls the allocation of the point domain to the RAN control function with priority. as well as The execution unit, based on the amount of computing resources allocated by the resource allocation unit, executes the RAN control function and the AI ​​processing function on the execution infrastructure.

2. The system according to claim 1, wherein, If there is a first residual portion in the point domain that has not been allocated to the RAN control function, the resource allocation unit allocates the first residual portion to the AI ​​processing function.

3. The system according to claim 2, wherein, The AI ​​processing functions include AI execution processing for performing AI, and AI learning processing for learning AI. The resource allocation unit allocates the AI-occupied domain and the first residual portion to the AI ​​for execution processing. If there is a second residual portion in the first residual portion that has not been allocated to the AI ​​for execution processing, the second residual portion is allocated to the AI ​​for learning processing.

4. The system according to claim 3, wherein, The AI ​​execution processing includes RAN control AI execution processing associated with RAN control, and non-RAN control AI execution processing not associated with RAN control. In the RAN-controlled AI execution process and the non-RAN-controlled AI execution process, the resource allocation unit preferentially allocates the AI ​​occupied domain and the first residual portion to the RAN-controlled AI execution process.

5. The system according to claim 4, wherein, The non-RAN control AI execution process includes at least one of the following: monitoring AI execution process that determines the situation within the shooting range of the input captured image, and answering AI execution process that outputs an answer to the input user's query.

6. The system according to any one of claims 1 to 5, wherein, If at least a portion of the point domain is not utilized by the RAN control function, the resource allocation unit determines, based on the needs of the AI ​​processing, whether to allocate the unused portion of the point domain to the AI ​​processing function or to convert the hardware corresponding to the unused portion of the point domain to a power-saving state.

7. The system according to claim 6, wherein, The system further includes a state control unit that, when it is determined that the hardware corresponding to the unused portion of the point domain should be converted to a power-saving state, converts the hardware corresponding to the unused portion of the point domain to different power-saving states based on the utilization rate of the RAN occupied domain.

8. The system according to claim 7, wherein, If the utilization rate of the RAN occupancy domain is lower than a predetermined first threshold, the state control unit controls the power supply to the hardware corresponding to the unused portion of the point domain. If the utilization rate of the RAN occupancy domain is higher than the first threshold but lower than a second threshold higher than the first threshold, the state control unit controls the hardware corresponding to the unused portion of the point domain to enter a sleep state.

9. A program that causes a computer to execute: The resource management steps involve dividing the computing resources on an execution infrastructure into a RAN occupancy domain that can only be used by the RAN control function that controls the RAN, an AI occupancy domain that can only be used by the AI ​​processing function that performs AI processing, and a point domain that can be used by both the RAN control function and the AI ​​processing function. The resource allocation step involves allocating computing resources on the execution infrastructure to the RAN control function and the AI ​​processing function. If the amount of computing resources required by the RAN control function is greater than the amount of computing resources occupied by the RAN domain, control is performed to prioritize the allocation of the point domain to the RAN control function. as well as The execution step involves performing the RAN control function and the AI ​​processing function on the execution infrastructure, based on the amount of computing resources allocated in the resource allocation step.

10. A control method executed by a computer, the control method comprising: The resource management steps involve dividing the computing resources on an execution infrastructure into a RAN occupancy domain that can only be used by the RAN control function that controls the RAN, an AI occupancy domain that can only be used by the AI ​​processing function that performs AI processing, and a point domain that can be used by both the RAN control function and the AI ​​processing function. The resource allocation step involves allocating computing resources on the execution infrastructure to the RAN control function and the AI ​​processing function. If the amount of computing resources required by the RAN control function is greater than the amount of computing resources occupied by the RAN domain, control is performed to prioritize the allocation of the point domain to the RAN control function. as well as The execution step involves performing the RAN control function and the AI ​​processing function on the execution infrastructure, based on the amount of computing resources allocated in the resource allocation step.

Images