Execution base, control method, program, and system

Abstract

Provided is a system comprising a plurality of execution bases for controlling a RAN, the plurality of execution bases each including an inference server for executing AI inference processing, and a scheduler for managing the schedule of the inference server. In response to receiving an execution request for an inference task corresponding to a network slice, the scheduler determines whether or not to execute the inference task with the inference server managed by the scheduler, on the basis of a delay tolerance requirement of the network slice of the inference task, and if it was determined that the inference task is to be executed by the managed inference server, performs control so as to cause the managed inference server to execute the inference task . The execution base may include a RAN control function and an AI processing (RAN Intelligent Controller (RIC) or the like) function.

Description

Execution platform, control method, program, and system

[0001] The present invention relates to an execution platform, a control method, a program, and a system.

[0002] Patent Document 1 describes a resource management method for network slicing, which is adapted to manage radio resources of a terminal device under a radio access network. [Prior art documents] [Patent document 1] JP 2022-077481 A

[0003] In 5GC (5th Generation Core), there is a function called network slicing that creates VNFs (Virtualized Network Functions) suitable for each requirement of high speed, large capacity, ultra-large number of connections, and ultra-low latency and provides services. If you want to perform processing using AI (Artificial Intelligence) in MEC (Multi-access Edge Computing) in order to achieve low latency, the relevant slice creates a UPF (User Plane Function) in the same distributed DC (Data Center) as the MEC's ​​AI and connects it to the DN (Data Network) of the MEC's ​​AI to achieve low latency processing. Multiple such slices may exist. On the other hand, from a resource perspective, it is desirable for AI inference servers to accept tasks regardless of slice, without unnecessarily duplicating the server itself. Conventionally, when connecting to the same inference server from different slices, it was not possible to prioritize inference tasks. Furthermore, even if a slice did not require low latency, it was not possible to determine whether it was possible to send an inference task execution request from a UPF in a central DC to an inference server in a distributed DC that was available.

[0004] The system 10 according to this embodiment includes a scheduler that, for example, when sending an execution request for an inference task to an inference server, can send the execution request to the most appropriate inference server. Furthermore, for example, metadata that serves as the basis for the scheduler's decision based on slicing information is added in the UPF. By passing the scheduler before the inference server, the inference server can be used efficiently without compromising the slicing-based SLA (Service Level Agreement).

[0005] As a specific example, the system 10 according to this embodiment includes a distributed infrastructure having a RAN control function for controlling a RAN (Radio Access Network) and an AI processing function for performing AI processing, and a management infrastructure for managing multiple distributed infrastructures, and a scheduler is introduced into the multiple distributed infrastructures and the management infrastructure.

[0006] Types of AI processing include AI processing related to RAN control (sometimes referred to as RAN-controlled AI processing) and AI processing not related to RAN control (sometimes referred to as non-RAN-controlled AI processing).

[0007] An example of RAN control AI processing is RIC (RAN Intelligent Controller). RIC is a technology that uses AI to optimize RAN radio resources and automate RAN operations. RIC includes Non-RT RIC (Non-Real Time RIC) and Near-RT RIC (Near-Real Time RIC). Non-RT RIC is sometimes called Centralized RIC. Non-RT RIC is located inside SMO (Service Management and Orchestration), which manages and orchestrates the RAN. Non-RT RIC generates and notifies policies related to RAN control and sends information to Near-RT RIC. For example, the Non-RT RIC performs machine learning using data collected from the RAN to generate a trained model for RAN control and transmits it to the Near-RT RIC. The Near-RT RIC is sometimes called a Distributed RIC. Compared to the Non-RT RIC, the Near-RT RIC is located closer to the RAN nodes (RU (Radio Unit), DU (Distributed Unit), CU (Central Unit)) and controls the RAN nodes, resources, etc. The Near-RT RIC performs processing with higher real-time performance than the Non-RT RIC. The Near-RT RIC performs inference processing related to RAN control using, for example, a trained model acquired from the Non-RT RIC. RAN control AI processing is not limited to the RIC.

[0008] Non-RAN controlled AI processes may correspond to so-called MEC applications, and include any AI learning and inference processes that are not related to RAN control.

[0009] According to one embodiment of the present invention, there is provided a system. The system may include a plurality of execution platforms that control a RAN. Each of the plurality of execution platforms may have an inference server that executes AI inference processing. Each of the plurality of execution platforms may have a scheduler that manages the schedule of the inference server. In response to receiving an execution request for an inference task corresponding to a network slice, the scheduler may determine whether to execute the inference task on the inference server under its management based on the delay tolerance requirement of the network slice for the inference task, and if it is determined that the inference task should be executed on the inference server under its management, the scheduler may cause the inference server under its management to execute the inference task.

[0010] In the system, the scheduler may determine that the inference task is executed on the inference server under its own management when the delay tolerance requirement of the inference task is equal to or less than a predetermined threshold, and may transmit an execution request for the inference task to a scheduler of another execution platform when the delay tolerance requirement of the inference task is not equal to or less than the predetermined threshold.

[0011] In any of the above systems, when the delay tolerance requirement of the inference task is not below the predetermined threshold, the scheduler may decide, based on the utilization status of the computing resources of the inference server under its own management, whether to have the inference task executed by the inference server under its own management, or to send a request to execute the inference task to a scheduler of another execution platform.

[0012] In any of the above systems, each of the multiple execution platforms may have a schedule acquisition unit that acquires schedules of the inference servers of the other execution platforms, and when the scheduler determines not to execute the inference task on the inference server managed by the scheduler, the scheduler may determine the inference server of the other execution platform to request execution of the inference task based on the schedule of the inference server of the other execution platform.When the scheduler determines not to execute the inference task on the inference server managed by the scheduler, and there are multiple inference servers of the other execution platforms that can execute the inference task, the scheduler may preferentially select, from the multiple other execution platforms, the inference server of the other execution platform that is closest to the scheduler's own execution platform as the inference server to request execution of the inference task.

[0013] Any of the systems may further include a UPF that receives data of the inference task and sends metadata indicating the delay tolerance requirements of the network slice to the scheduler in addition to the data of the inference task, and the scheduler may decide whether to execute the inference task on the inference server under its management based on the delay tolerance requirements of the network slice of the inference task indicated by the metadata.

[0014] In any of the above systems, the multiple execution platforms may include multiple distributed platforms and a management platform that manages the multiple distributed platforms.

[0015] According to one embodiment of the present invention, there is provided an execution platform. The execution platform may include a RAN control unit that executes a RAN control function to control a RAN. The execution platform may include an inference server that executes AI inference processing. The execution platform may include a scheduler that manages a schedule of the inference server. In response to receiving an execution request for an inference task corresponding to a network slice, the scheduler may determine whether to execute the inference task on the inference server under its own management based on the delay tolerance requirement of the network slice for the inference task, and if it is determined that the inference task should be executed on the inference server under its own management, the scheduler may control the inference server under its own management to execute the inference task.

[0016] According to one embodiment of the present invention, a program is provided for causing a computer having a RAN control function for controlling a RAN and an inference server for executing AI inference processing to execute a decision step in response to receiving a request to execute an inference task corresponding to a network slice, in which the computer decides whether to execute the inference task on the inference server based on the delay tolerance requirements of the network slice for the inference task, and a control step in which the computer controls the inference server under its own management to execute the inference task if it is decided in the decision step that the inference task should be executed on the inference server under its own management.

[0017] According to one embodiment of the present invention, there is provided a control method executed by a computer including a RAN control function for controlling a RAN and an inference server for executing AI inference processing. The control method may include a decision step for determining whether to execute an inference task corresponding to a network slice in response to receiving an execution request for the inference task based on a delay tolerance requirement of the network slice for the inference task. If it is determined in the decision step that the inference task should be executed by the inference server under the control of the control method, the control method may include a control step for controlling the inference server under the control of the control method to execute the inference task.

[0018] The above summary of the invention does not list all of the necessary features of the present invention, and subcombinations of these features may also constitute inventions.

[0019] FIG. 1 is an explanatory diagram for explaining slicing. FIG. 1 shows an example of a system 10. FIG. 1 is an explanatory diagram for explaining processing of inference tasks in the system 10 when the distributed infrastructure 200 does not include a scheduler 220 and the management infrastructure 100 does not include a scheduler 120. FIG. 1 is an explanatory diagram for explaining scheduling of inference tasks by the schedulers 120 and 220. FIG. 1 is an explanatory diagram for explaining scheduling of inference tasks by the schedulers 120 and 220. FIG. 1 is an explanatory diagram for explaining assignment of metadata by the UPF 204 in the system 10. FIG. 1 shows an example of the functional configuration of the distributed infrastructure 200. FIG. 1 shows an example of the functional configuration of the management infrastructure 100. FIG. 1 shows an example of the hardware configuration of a computer 1200 that functions as the management infrastructure 100 or the distributed infrastructure 200.

[0020] The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the scope of the invention as claimed. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0021] Fig. 1 is an explanatory diagram for explaining slicing. In Fig. 1, RU is a radio device of a base station, vDU (virtualized DU) is a radio signal processing unit of a virtualized base station, vCU (virtualized CU) is a data processing unit of a virtualized base station, UPF is a transmission / reception processing unit of user data, and App is an application server.

[0022] FIG. 1 illustrates three slices. Slice A is a slice for video distribution, which is a slice for realizing high-capacity mobile broadband through antenna cooperation. Slice B is a slice for IoT, which is a slice for efficiently accommodating a large amount of data. Slice C is a slice for remote control, which is a slice for realizing ultra-low latency through edge processing.

[0023] As shown in FIG. 1, depending on the slice requirements, the vDU and vCU are located at the antenna site, the local storage station, or the central storage station. This means that the vDU and vCU can be located at any hierarchical level depending on the slicing. In this way, it is desirable to select the node (distribution platform / management platform) running the pod that performs the inference task depending on which slice is used.

[0024] 2 schematically illustrates an example of a system 10 according to the present embodiment. The system 10 includes a distributed infrastructure 200. The distributed infrastructure 200 may be an example of an execution infrastructure. The system 10 may include a plurality of distributed infrastructures 200. The system 10 may include a management infrastructure 100 that manages the plurality of distributed infrastructures 200. The management infrastructure 100 may be an example of an execution infrastructure. In the system 10 according to the present embodiment, the management infrastructure 100 and the plurality of distributed infrastructures 200 may cooperate to control the RAN 310 and perform AI processing.

[0025] The RAN 310 may be a virtualized vRAN (Virtual RAN), and the system 10 may control the vRAN. The RAN 310 may be a physical RAN, and the system 10 may control the physical RAN. In this embodiment, a case where the RAN 310 is a vRAN will be mainly described as an example.

[0026] The AI ​​processing performed by the system 10 may include RAN-controlled AI processing (sometimes referred to as RAN_AI). The AI ​​processing performed by the system 10 may include non-RAN-controlled AI processing (sometimes referred to as non-RAN_AI).

[0027] The distributed infrastructure 200 may be a data center located in various locations. The distributed infrastructure 200 may be configured with multiple devices. The distributed infrastructure 200 may be realized on a virtualization platform made up of multiple devices. The distributed infrastructure 200 may also be realized by a single device. In other words, the distributed infrastructure 200 may be a distributed device.

[0028] The management infrastructure 100 may be a data center that manages multiple distributed infrastructures 200. The management infrastructure 100 may be configured with multiple devices. The management infrastructure 100 may be realized on a virtualization infrastructure made up of multiple devices. The management infrastructure 100 may also be realized by a single device. In other words, the management infrastructure 100 may be a management device.

[0029] The management infrastructure 100 may be referred to as a Core Brain, and the distributed infrastructure 200 may be referred to as a Regional Brain. While FIG. 1 illustrates an example in which a single-level management infrastructure 100 is arranged below the management infrastructure 100, this is not limiting. The distributed infrastructure 200 may have multiple levels. For example, when a two-level distributed infrastructure 200 is arranged below the management infrastructure 100, the management infrastructure 100 may be referred to as a Core Brain, the distributed infrastructure 200 at the level below that may be referred to as a Regional Brain, and the distributed infrastructure 200 at the level further below that may be referred to as a Sub-Regional Brain.

[0030] In the case where a Core Brain, a Regional Brain, and a Sub Regional Brain are present, the scheduling of inference tasks according to this embodiment may be performed by a plurality of Sub Regional Brains, by a plurality of Sub Regional Brains and one or more Regional Brains, by a plurality of Regional Brains and one or more Core Brains, or by a plurality of Sub Regional Brains, a plurality of Regional Brains, and one or more Core Brains.

[0031] The distributed infrastructure 200 may be arranged with one or more central processing units (CPUs). The distributed infrastructure 200 may be arranged with one or more graphics processing units (GPUs). The distributed infrastructure 200 may be arranged with multiple super chips, each of which has a CPU and a GPU connected via an interconnect. The interconnect may have memory consistency and may be capable of achieving high bandwidth and low latency. In this way, the distributed infrastructure 200 may have CPU resources and GPU resources as computational resources.

[0032] FIG. 3 is an explanatory diagram for explaining the processing of inference tasks in the system 10 when the distribution infrastructure 200 does not include the scheduler 220 and the management infrastructure 100 does not include the scheduler 120. In FIG.

[0033] Here, a vDU / vCU 202, a UPF 204, and an App 206 (which may be referred to as a first vDU / vCU, a first UPF, or a first App) corresponding to a low-latency slice 410, and a vDU / vCU 202, a UPF 204, and an App 206 (which may be referred to as a second vDU / vCU, a second UPF, or a second App) corresponding to a low-latency slice 420 are configured in one distributed infrastructure 200 (which may be referred to as a first distributed infrastructure). The first vDU / vCU, the first UPF, the first App, the second vDU / vCU, the second UPF, and the second App may be configured and managed by a management unit 210 of the first distributed infrastructure. In this example, the delay tolerance requirement of the low-delay slice 410 is SLA 10 ms (milliseconds), and the delay tolerance requirement of the low-delay slice 420 is SLA 50 ms.

[0034] Furthermore, a vDU / vCU 202, a UPF 204, and an App 206 (which may be referred to as a third vDU / vCU, a third UPF, or a third App) corresponding to the low-latency slice 430, and a vDU / vCU 202 (which may be referred to as a fourth vDU / vCU) corresponding to the non-low-latency slice 440 are configured in another distributed infrastructure 200 (which may be referred to as a second distributed infrastructure), and a UPF 104 and an App 106 (which may be referred to as a fourth UPF or a fourth App) corresponding to the non-low-latency slice 440 are configured in the management infrastructure 100. The third vDU / vCU, the third UPF, the third App, and the fourth vDU / vCU may be configured and managed by the management unit 210 of the second distributed infrastructure. The fourth UPF and the fourth App may be configured and managed by the management unit 110 of the management infrastructure 100. The delay tolerance requirement of the low delay slice 430 is SLA 50 ms, and the delay tolerance requirement of the non-low delay slice 440 is SLA 1 s.

[0035] Conventionally, it has not been possible to prioritize inference tasks from multiple slices in the inference server 230. As shown in Figure 3, when the first distribution platform receives inference tasks from two slices, the inference tasks cannot be prioritized, and both are processed in the inference server 230.

[0036] In addition, in the past, the resources of the inference server 230 and the inference server 130 were not necessarily utilized effectively, and regardless of the status of the second inference server 230 and the first inference server 130, inference tasks from the low-latency slice 430 were processed by the inference server 230 of the second distributed infrastructure, and inference tasks from the non-low-latency slice 440 were processed by the inference server 130 of the management infrastructure 100.

[0037] In contrast, in the system 10 according to this embodiment, the distribution infrastructure 200 includes the scheduler 220. Alternatively, the management infrastructure 100 may include the scheduler 120.

[0038] 4, 5, and 6 are explanatory diagrams for explaining the scheduling of inference tasks by the scheduler 120 and the scheduler 220. FIG.

[0039] 4 , the scheduler 220 of the first distributed infrastructure may cause the inference server 230 of the first distributed infrastructure to execute both the inference task of the low-latency slice 410 and the inference task of the low-latency slice 420 if both are executable by the inference server 230 of the first distributed infrastructure. The scheduler 220 of the first distributed infrastructure may determine that the inference task of the low-latency slice 410, which has strict delay tolerance requirements, is to be executed by the inference server 230 of the first distributed infrastructure, and may determine that both the inference task of the low-latency slice 410 and the inference task of the low-latency slice 420 are executable by the inference server 230 of the first distributed infrastructure if the scheduler 220 of the first distributed infrastructure determines that the inference server 230 of the first distributed infrastructure can satisfy the delay tolerance requirements of the inference task of the low-latency slice 420 even when executing the inference task of the low-latency slice 410. In this case, the scheduler 220 of the first distribution infrastructure may control the execution of inference tasks of the low-delay slice 410, which has stricter delay tolerance requirements, prior to the execution of inference tasks of the low-delay slice 420. This allows the inference tasks of slices with different SLAs to be executed using the computational resources of a single inference server 230 if the delay tolerance requirements can be met. It also makes it possible to prioritize the processing of inference tasks with stricter delay tolerance requirements.

[0040] When the scheduler 220 of the first distributed infrastructure determines that the delay tolerance requirements of the low-latency slice 420 cannot be satisfied when the inference server 230 of the first distributed infrastructure executes the inference task of the low-latency slice 410, the scheduler 220 may request the scheduler 220 of another distributed infrastructure 200 to execute the inference task of the low-latency slice 420. The scheduler 220 of the first distributed infrastructure may, for example, request the scheduler 220 of the second distributed infrastructure to execute the inference task of the low-latency slice 420. When the scheduler 220 of the second distributed infrastructure determines that the delay tolerance requirements of the low-latency slice 420 can be satisfied by the inference server 230 of the second distributed infrastructure, the scheduler 220 may cause the inference server 230 of the second distributed infrastructure to execute the inference task of the low-latency slice 420. When the scheduler 220 of the second distributed infrastructure determines that the delay tolerance requirements of the low-latency slice 420 cannot be satisfied by the inference server 230 of the second distributed infrastructure, the scheduler 220 may request the execution of the inference task of the low-latency slice 420 from the scheduler 220 of another distributed infrastructure 200. In this way, in a situation where the inference server 230 managed by the scheduler 220 cannot satisfy the delay tolerance requirements, the inference server 230 of another distributed infrastructure 200 may execute the inference task, thereby making it possible to satisfy the delay tolerance requirements.

[0041] The scheduler 220 of the first distributed infrastructure may cause another distributed infrastructure 200 to execute multiple inference tasks, regardless of whether the inference tasks can be executed by the inference server 230 it manages, as long as the delay tolerance requirements are met. For example, the schedules of the inference servers 230 are shared among the schedulers 220. As a specific example, the scheduler 220 of the first distributed infrastructure provides the schedule of the inference server 230 of the first distributed infrastructure to the scheduler 220 of the second distributed infrastructure. The scheduler 220 of the second distributed infrastructure provides the schedule of the inference server 230 of the second distributed infrastructure to the scheduler 220 of the second distributed infrastructure. The scheduler 220 of the first distributed platform may determine, based on the schedule of the inference server 230 of the second distributed platform, whether the inference server 230 of the second distributed platform can meet the inference task-on-delay tolerance requirements of the low-latency slice 420, and if it determines that the requirements can be met, send a request to execute the inference task of the low-latency slice 420 to the scheduler 220 of the second distributed platform.

[0042] 5 , if the inference server 230 of the second distributed platform can satisfy the delay tolerance requirements of the low-latency slice 430 for the inference task of the low-latency slice 430, the scheduler 220 of the second distributed platform may have the inference server 230 of the second distributed platform execute the inference task of the low-latency slice 430. If the scheduler 220 of the second distributed platform determines that the delay tolerance requirements of the low-latency slice 430 cannot be satisfied, for example, because the second inference server 230 is already executing the inference task of the low-latency slice 420 in response to an execution request from the scheduler 220 of the first distributed platform, the scheduler 220 of the second distributed platform may have the inference server 230 of the first distributed platform execute the inference task of the low-latency slice 430, or may request the scheduler 220 of another distributed platform 200 to execute the inference task of the low-latency slice 430.

[0043] 6, the scheduler 120 of the management infrastructure 100 may cause the inference server 130 to execute the inference task of the non-low-latency slice 440 if it determines that the inference server 130 can satisfy the delay tolerance requirements. The scheduler 120 may also cause the inference server 230 of the first distributed infrastructure or the inference server 230 of the second distributed infrastructure to execute the inference task of the non-low-latency slice 440.

[0044] For example, scheduler 120, scheduler 220 of the first distributed infrastructure, and scheduler 220 of the second distributed infrastructure share schedules for inference server 130, inference server 230 of the first distributed infrastructure, and inference server 230 of the second distributed infrastructure. Scheduler 120 causes inference task 440 to be executed by inference server 230 of the first distributed infrastructure or inference server 230 of the second distributed infrastructure, depending on whether inference server 230 of the first distributed infrastructure has available resources and whether the inference server 230 of the first distributed infrastructure can satisfy the delay tolerance requirements of non-low-latency slice 440, and whether inference server 230 of the second distributed infrastructure has available resources and whether the inference server 230 of the second distributed infrastructure can satisfy the delay tolerance requirements of non-low-latency slice 440.

[0045] 7 is an explanatory diagram for explaining the assignment of metadata by the UPF in the system 10. Here, the assignment of metadata by the UPF 204 in the distributed infrastructure 200 will be described as an example, but the UPF 104 in the management infrastructure 100 may also assign metadata in a similar manner.

[0046] The UPF 204 receives data corresponding to a slice via a GTP (GPRS (General Packet Radio Switching) Tunneling Protocol) connection from the vDU / vCU 202. In the example shown in Figure 7, the UPF 204 receives data 510 and data 520.

[0047] The UPF 204 assigns metadata to the data based on the slicing information of the received data. The UPF 204 may assign metadata indicating delay tolerance requirements included in the slicing information to the data. In the example shown in FIG. 7 , the UPF 204 assigns metadata 512 to the data 510 that satisfies the delay tolerance requirements included in the slicing information of the data 510, and assigns metadata 522 to the data 520 that satisfies the delay tolerance requirements included in the slicing information of the data 520. The UPF 204 transmits the data with the assigned metadata to the scheduler 220.

[0048] The scheduler 220 acquires delay tolerance requirements from the metadata attached to the data. In the example shown in Fig. 7, the scheduler 220 acquires delay tolerance requirements for the data 510 from the metadata 512, and acquires delay tolerance requirements for the data 520 from the metadata 522. Based on the delay tolerance requirements, the scheduler 220 may determine whether to have the inference server 230 it manages execute the inference task, or whether to have the inference server 230 of another distributed infrastructure 200 or the inference server 130 of the management infrastructure 100 execute the inference task.

[0049] 7, the scheduler 220 determines that both data 510 and data 520 should be executed by the inference server 230 under its management. The scheduler 220 may determine the priority of the data 510 and data 520 based on delay tolerance requirements, and manage the data so that the data with the higher priority is processed first. In the example shown in FIG. 7, the scheduler 220 manages the data 510 and data 520 so that the data 510 is processed first.

[0050] 8 shows an example of the functional configuration of the distributed infrastructure 200. The distributed infrastructure 200 includes a storage unit 208, a management unit 210, a RAN control unit 212, an AI management unit 214, a schedule acquisition unit 216, a scheduler 220, and an inference server 230. Note that it is not essential for the distributed infrastructure 200 to include all of these units.

[0051] The storage unit 208 stores various data. The management unit 210 performs various management operations. The management unit 210 may configure the vDU / vCU 202, the UPF 204, or the App 206.

[0052] The RAN control unit 212 executes a RAN control function to control the RAN 310 configured by a plurality of radio base stations 300. The RAN control unit 240 may execute a function of a so-called vRAN (Virtual RAN). The RAN control unit 240 controls communications in the RAN 310 by controlling the plurality of radio base stations 300 that configure the RAN 310 and coordinating with other distributed infrastructures 200.

[0053] The AI ​​management unit 214 manages AI processing. The AI ​​management unit 214 may execute RAN-controlled AI processing. The AI ​​management unit 214 may execute non-RAN-controlled AI processing. The AI ​​management unit 214 may cause the inference server 230 to execute AI inference processing.

[0054] The schedule acquisition unit 216 acquires the schedule of the inference server of another execution platform. For example, the schedule acquisition unit 216 acquires the schedule of the inference server 230 of another distributed platform 200 from the scheduler 220 of the other distributed platform 200. For example, the schedule acquisition unit 216 acquires the schedule of the inference server 130 of the management platform 100 from the scheduler 120 of the management platform 100.

[0055] The scheduler 220 manages the schedule of the inference server 230 of the distributed infrastructure 200 (sometimes referred to as the local machine) that has the scheduler 220. The inference server 230 executes AI inference processing.

[0056] In response to receiving a request to execute an inference task corresponding to a network slice, the scheduler 220 determines whether or not to execute the inference task on its own inference server 230 based on the delay tolerance requirements of the network slice of the inference task. When the scheduler 220 determines to execute the inference task on its own inference server 230, the scheduler 220 causes its own inference server 230 to execute the inference task.

[0057] The scheduler 220 may receive inference task data with the metadata attached from a UPF that receives the inference task data and adds metadata to the inference task data indicating the delay tolerance requirements of the network slice of the inference task. The scheduler 220 may determine whether to execute the inference task on its own inference server 230 based on the delay tolerance requirements of the network slice of the inference task indicated by the metadata.

[0058] For example, if the delay tolerance requirement of the inference task is equal to or less than a predetermined threshold, the scheduler 220 determines to execute the inference task on its own inference server 230. The threshold may be set arbitrarily and may be changeable after being set. For example, if the threshold is 10 ms, the scheduler 220 determines to execute the inference task on its own inference server 230 if the delay tolerance requirement of the inference task is equal to or less than 10 ms.

[0059] For example, if the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, the scheduler 220 transmits an execution request for the inference task to a scheduler of another execution platform. For example, if the threshold is 10 ms, and the delay tolerance requirement of the inference task is not equal to or less than 10 ms, the scheduler 220 transmits an execution request for the inference task to a scheduler of another execution platform. If the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, the scheduler 220 may transmit an execution request for the inference task to the scheduler 220 of another distributed platform 200. If the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, the scheduler 220 may transmit an execution request for the inference task to the scheduler 120 of the management platform 100.

[0060] When the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, the scheduler 220 may determine, based on the utilization status of the computational resources of its own inference server 230, whether to have its own inference server 230 execute the inference task or to send an execution request for the inference task to a scheduler of another execution platform. For example, when the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, if the remaining amount of computational resources of its own inference server 230 is greater than the predetermined threshold, the scheduler 220 has its own inference server 230 execute the inference task, whereas if the remaining amount of computational resources of its own inference server 230 is less than the predetermined threshold, the scheduler 220 sends an execution request for the inference task to a scheduler of another execution platform. The threshold may be set arbitrarily and may be changeable after being set.

[0061] When the scheduler 220 receives execution requests for multiple inference tasks, the scheduler 220 may determine whether to have the inference server 230 of its own device execute the inference tasks in order of strictest delay tolerance requirements. For example, the scheduler 220 may first determine that the inference server 230 of its own device executes an inference task (sometimes referred to as a first inference task) that has the strictest delay tolerance requirements among the multiple inference tasks. Next, the scheduler 220 may determine whether the inference server 230 of its own device can execute the first inference task and then satisfy the delay tolerance requirements of the second inference task (sometimes referred to as a second inference task). If the scheduler 220 determines that the delay tolerance requirements can be satisfied, the scheduler 220 may determine that the inference server 230 of its own device executes the second inference task. If the scheduler 220 determines that the delay tolerance requirements cannot be satisfied, the scheduler 220 may send execution requests to another execution platform for all other inference tasks, including the second inference task, among the multiple inference tasks.

[0062] The scheduler 220 may determine the destination of the inference task based on the schedule of another execution platform acquired by the schedule acquisition unit 216. The scheduler 220 may determine the destination of the inference task based on the schedule of the inference server 230 of the distributed platform 200 and the schedule of the inference server 130 of the management platform 100 acquired by the schedule acquisition unit 216.

[0063] When the scheduler 220 determines not to execute an inference task on its own inference server 230, it may determine an inference server of another execution platform to request execution of the inference task based on the schedule of the inference server of the other execution platform. For example, when the scheduler 220 determines not to execute an inference task on its own inference server 230 and there are multiple inference servers of other execution platforms that can execute the inference task, the scheduler 220 preferentially selects, from the multiple other execution platforms, an inference server of another execution platform that is closer to the scheduler 220 as the inference server to request execution of the inference task. For example, when the scheduler 220 determines not to execute an inference task on its own inference server 230 and there are multiple inference servers 230 of other distributed platforms 200 that can execute the inference task, the scheduler 220 preferentially selects, from the multiple other distributed platforms 200, an inference server 230 of another distributed platform 200 that is closer to the scheduler 220 as the inference server to request execution of the inference task. This reduces communication delays compared to when requesting the execution of an inference task from an inference server 230 of another distributed infrastructure 200 that is farther away from the own machine, thereby contributing to the rapid execution of the inference task.

[0064] 9 shows an example of the functional configuration of the management infrastructure 100. The management infrastructure 100 includes a storage unit 108, a management unit 110, a RAN control unit 112, an AI management unit 114, a schedule acquisition unit 116, a scheduler 120, and an inference server 130. Note that it is not essential for the management infrastructure 100 to include all of these components.

[0065] The storage unit 108 stores various data. The management unit 110 performs various management operations. The management unit 110 may configure the vDU / vCU 102, the UPF 104, or the App 106.

[0066] The RAN control unit 112 cooperates with the RAN control units 212 of the multiple distribution infrastructures 200 to control the RAN 310 .

[0067] The AI ​​management unit 114 manages AI processing. The AI ​​management unit 114 may execute RAN-controlled AI processing. The AI ​​management unit 114 may execute non-RAN-controlled AI processing. The AI ​​management unit 114 may cause the inference server 130 to execute AI inference processing.

[0068] The schedule acquisition unit 116 acquires the schedule of the inference server of another execution platform. For example, the schedule acquisition unit 116 acquires the schedule of the inference server 230 of the distribution platform 200 from the scheduler 220 of the distribution platform 200.

[0069] The scheduler 120 manages the schedule of the inference server 130 of the management base 100 (sometimes referred to as the "own device") that has the scheduler 120. The inference server 130 executes AI inference processing.

[0070] In response to receiving a request to execute an inference task corresponding to a network slice, the scheduler 120 determines whether or not to execute the inference task on its own inference server 130 based on the delay tolerance requirements of the network slice of the inference task. When the scheduler 120 determines to execute the inference task on its own inference server 130, the scheduler 120 causes its own inference server 130 to execute the inference task.

[0071] The scheduler 120 may receive the inference task data with the metadata attached from the UPF, which receives the inference task data and adds metadata to the inference task data indicating the delay tolerance requirements of the network slice of the inference task. The scheduler 120 may determine whether to execute the inference task on its own inference server 130 based on the delay tolerance requirements of the network slice of the inference task indicated by the metadata.

[0072] For example, if the delay tolerance requirement of the inference task is equal to or less than a predetermined threshold, the scheduler 120 determines to execute the inference task on its own inference server 130. The threshold may be arbitrarily settable and may be changeable after being set.

[0073] For example, if the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, the scheduler 120 may send an execution request for the inference task to a scheduler of another execution platform. If the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, the scheduler 120 may send an execution request for the inference task to the scheduler 220 of the distributed platform 200.

[0074] When the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, the scheduler 120 may determine, based on the utilization status of the computational resources of its own inference server 130, whether to have its own inference server 130 execute the inference task or to send an execution request for the inference task to a scheduler of another execution platform. For example, when the delay tolerance requirement of the inference task is not equal to or less than a predetermined threshold, if the remaining amount of computational resources of its own inference server 130 is greater than the predetermined threshold, the scheduler 120 has its own inference server 130 execute the inference task, whereas if the remaining amount of computational resources of its own inference server 130 is less than the predetermined threshold, the scheduler 120 sends an execution request for the inference task to a scheduler of another execution platform. The threshold may be set arbitrarily and may be changeable after being set.

[0075] When the scheduler 120 receives execution requests for multiple inference tasks, the scheduler 120 may determine whether to have its own inference server 130 execute the tasks in order of strictest delay tolerance requirements. For example, the scheduler 120 may first determine that the inference server 130 executes an inference task (sometimes referred to as a first inference task) having the strictest delay tolerance requirements among the multiple inference tasks. Next, for an inference task (sometimes referred to as a second inference task) having the next strictest delay tolerance requirements among the multiple inference tasks, the scheduler 120 determines whether the inference server 130 of its own machine can execute the first inference task and then satisfy the delay tolerance requirements of the second inference task. If it determines that the delay tolerance requirements can be satisfied, the scheduler 120 may determine that the inference server 130 of its own machine executes the second inference task. If it determines that the delay tolerance requirements cannot be satisfied, the scheduler 120 may send execution requests to another execution platform for all other inference tasks, including the second inference task, among the multiple inference tasks.

[0076] The scheduler 120 may determine the destination to which an inference task is to be requested based on the schedule of another execution platform acquired by the schedule acquisition unit 116. The scheduler 120 may determine the destination to which an inference task is to be requested based on the schedule of the inference server 230 of the distributed platform 200 acquired by the schedule acquisition unit 116.

[0077] When the scheduler 120 determines not to execute an inference task on its own inference server 130, it may determine an inference server of another execution platform to request execution of the inference task based on the schedules of the inference servers of the other execution platforms. For example, when the scheduler 120 determines not to execute an inference task on its own inference server 130 and there are multiple inference servers of other execution platforms that can execute the inference task, the scheduler 120 preferentially selects, from the multiple other execution platforms, an inference server of another execution platform that is closer to the scheduler 120 as the inference server to request execution of the inference task. For example, when the scheduler 120 determines not to execute an inference task on its own inference server 130 and there are multiple inference servers 230 of a distributed platform 200 that can execute the inference task, the scheduler 120 preferentially selects, from the multiple distributed platform 200, an inference server 230 of a distributed platform 200 that is closer to the scheduler 120 as the inference server 230 to request execution of the inference task.

[0078] 10 schematically illustrates an example of the hardware configuration of a computer 1200 that functions as the management infrastructure 100 or the distribution infrastructure 200. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of an apparatus according to the present embodiment, or can cause the computer 1200 to perform operations associated with the apparatus according to the present embodiment or one or more "parts," and / or can cause the computer 1200 to perform a process according to the present embodiment or steps of the process. Such a program can be executed by the CPU 1212 to cause the computer 1200 to perform specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0079] The computer 1200 according to this embodiment includes a CPU 1212, a GPU 1213, a RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communications interface 1222, a storage device 1224, a DVD drive 1226, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive 1226 may be a DVD-ROM drive, a DVD-RAM drive, or the like. The storage device 1224 may be a hard disk drive, a solid-state drive, or the like. The computer 1200 also includes a ROM 1230 and legacy input / output units such as a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0080] The CPU 1212 operates according to programs stored in the ROM 1230 and the RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires image data generated by the CPU 1212 into a frame buffer or the like provided in the RAM 1214 or into the graphics controller 1216 itself, and causes the image data to be displayed on the display device 1218.

[0081] 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 in the computer 1200. The DVD drive 1226 reads programs or data from a DVD-ROM 1227 or the like and provides them to the storage device 1224. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0082] The ROM 1230 stores therein a boot program or the like that is executed by the computer 1200 upon activation, and / or programs that depend on the hardware of the computer 1200. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via a USB port, a parallel port, a serial port, a keyboard port, a mouse port, etc.

[0083] The programs are provided by a computer-readable storage medium such as a DVD-ROM 1227 or an IC card. The programs are read from the computer-readable storage medium, installed in the storage device 1224, RAM 1214, or ROM 1230, which are also examples of computer-readable storage media, and executed by the CPU 1212. Information processing described in these programs is read by the computer 1200, and brings about cooperation between the programs and the various types of hardware resources described above. An apparatus or a method may be configured by implementing operations or processing of information in accordance with the use of the computer 1200.

[0084] For example, when communication is performed between computer 1200 and an external device, CPU 1212 may execute a communication program loaded into RAM 1214 and instruct communication interface 1222 to perform communication processing based on the processing described in the communication program. Under the control of CPU 1212, communication interface 1222 reads transmission data stored in a transmission buffer area provided in RAM 1214, storage device 1224, DVD-ROM 1227, or a recording medium such as an IC card, and transmits the read transmission data to a network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0085] Furthermore, the CPU 1212 may cause all or a necessary portion of a file or database stored in an external recording medium such as the storage device 1224, the DVD drive 1226 (DVD-ROM 1227), an IC card, etc. to be read into the RAM 1214, and may perform various types of processing on the data on the RAM 1214. The CPU 1212 may then write back the processed data to the external recording medium.

[0086] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. The CPU 1212 may perform various types of processing on data read from the RAM 1214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequences of the programs, and write the results back to the RAM 1214. The CPU 1212 may also search for information in a file, database, etc. on the recording medium. For example, if multiple entries, each having an attribute value of a first attribute associated with an attribute value of a second attribute, are stored on the recording medium, the CPU 1212 may search for an entry whose attribute value of the first attribute matches a specified condition from among the multiple entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0087] The above-described programs or software modules may be stored in a computer-readable storage medium on or near the computer 1200. 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 also be used as a computer-readable storage medium, thereby providing the programs to the computer 1200 via the network.

[0088] The blocks in the flowcharts and block diagrams in the present embodiments may represent stages of a process in which an operation is performed or "parts" of a device responsible for performing the operation. Particular stages and "parts" may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable storage medium, and / or a processor provided with computer-readable instructions stored on a computer-readable storage medium. The dedicated circuitry may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. The programmable circuitry may include reconfigurable hardware circuits, such as field programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), including AND, OR, XOR, NAND, NOR, and other logical operations, flip-flops, registers, and memory elements.

[0089] A computer-readable storage medium may include any tangible device capable of storing instructions that are executed by an appropriate device, such that a computer-readable storage medium having instructions stored thereon comprises an article of manufacture, including instructions that can be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of computer-readable storage media may include electronic, magnetic, optical, electromagnetic, and semiconductor storage media. More specific examples of computer-readable storage media may include floppy disks, diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), electrically erasable programmable read-only memories (EEPROMs), static random access memories (SRAMs), compact disc read-only memories (CD-ROMs), digital versatile discs (DVDs), Blu-ray discs, memory sticks, integrated circuit cards, and the like.

[0090] The computer readable instructions may include either assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and conventional procedural programming languages ​​such as the “C” programming language or similar programming languages.

[0091] The computer-readable instructions may be provided to a general-purpose computer, a special-purpose computer, or another programmable data processing device processor or programmable circuit, either locally or via a local area network (LAN), a wide area network (WAN) such as the Internet, so that the processor or programmable circuit of the programmable data processing device, such as a computer, executes the computer-readable instructions to generate means for performing the operations specified in the flowcharts or block diagrams. Here, the computer may be a personal computer (PC), a tablet computer, a smartphone, a workstation, a server computer, a general-purpose computer, a special-purpose computer, or the like, or may be a computer system in which multiple computers are connected. Such a computer system in which multiple computers are connected is also called a distributed computing system, and is a broad definition of computer. In a distributed computing system, multiple computers collectively execute a program by each executing a portion of the program and passing data between the computers as needed during program execution.

[0092] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc. A computer may have one processor or multiple processors. In a multiprocessor system with multiple processors, each processor executes a portion of a program and passes data between processors as needed during program execution, allowing the multiple processors to collectively execute the program. For example, in multitasking, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at each time slice. In this case, which portion of a program each processor executes changes dynamically. Which portion of a program each of the multiple processors executes may also be statically determined by multiprocessor-aware programming.

[0093] Although the present invention has been described above using the embodiments, the technical 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 and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention.

[0094] It should be noted that the execution order of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and that the processes can be performed in any order unless the output of a previous process is used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is described using "first," "next," etc. for convenience, this does not mean that the processes must be performed in this order.

[0095] 10 System, 100 Management infrastructure, 102 vDU / vCU, 104 UPF, 106 App, 108 Memory unit, 110 Management unit, 112 RAN control unit, 114 AI management unit, 116 Schedule acquisition unit, 120 Scheduler, 130 Inference server, 200 Distribution infrastructure, 202 vDU / vCU, 204 UPF, 206 App, 208 Memory unit, 210 Management unit, 212 RAN control unit, 214 AI management unit, 216 Schedule acquisition unit, 220 Scheduler, 230 Inference server, 300 Radio base station, 310 RAN, 410 Low-delay slice, 420 Low-delay slice, 430 Low-delay slice, 440 Non-low-delay slice, 510 Data, 512 Metadata, 520 data, 522 metadata, 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 comprising a plurality of execution platforms that control a RAN (Radio Access Network), each of the plurality of execution platforms having an inference server that executes AI (Artificial Intelligence) inference processing, and a scheduler that manages the schedule of the inference server, wherein the scheduler, upon receiving a request to execute an inference task corresponding to a network slice, determines whether or not to execute the inference task on the inference server under its own management based on the delay tolerance requirements of the network slice for the inference task, and if it determines that the inference task should be executed on the inference server under its own management, causes the inference server under its own management to execute the inference task.

2. The system of claim 1, wherein the scheduler determines to execute the inference task on the inference server under its management if the delay tolerance requirement of the inference task is equal to or less than a predetermined threshold.

3. The system described in claim 2, wherein the scheduler sends a request to execute the inference task to a scheduler of another execution platform if the delay tolerance requirement of the inference task is not equal to or less than the predetermined threshold.

4. The system described in claim 2, wherein, when the delay tolerance requirement of the inference task is not below the predetermined threshold, the scheduler determines, based on the utilization status of the computing resources of the inference server under its own management, whether to have the inference task executed by the inference server under its own management or to send a request to execute the inference task to a scheduler of another execution platform.

5. A system described in any one of claims 1 to 4, wherein each of the multiple execution platforms has a schedule acquisition unit that acquires the schedule of the inference server owned by the other execution platforms, and when the scheduler decides not to execute the inference task on the inference server under its own management, it determines the inference server of the other execution platform to request execution of the inference task based on the schedule of the inference server of the other execution platform.

6. The system described in claim 5, wherein, when the scheduler decides not to execute the inference task on the inference server under its own management, if there are multiple inference servers of the other execution platforms that can execute the inference task, the scheduler preferentially selects, from among the multiple other execution platforms, the inference server of the other execution platform that is closest to its own execution platform as the inference server to be requested to execute the inference task.

7. The system described in any one of claims 1 to 4, further comprising a UPF that receives data of the inference task and sends metadata indicating the delay tolerance requirements of the network slice to the scheduler in addition to the data of the inference task, wherein the scheduler decides whether to execute the inference task on the inference server under its management based on the delay tolerance requirements of the network slice of the inference task indicated by the metadata.

8. A system according to any one of claims 1 to 4, wherein the plurality of execution platforms includes a plurality of distributed platforms and a management platform that manages the plurality of distributed platforms.

9. An execution platform comprising: an execution unit that executes a RAN control function that controls a RAN (Radio Area Network); an inference server that executes AI (Artificial Intelligence) inference processing; and a scheduler that manages the schedule of the inference server, wherein the scheduler, upon receiving a request to execute an inference task corresponding to a network slice, determines whether or not to execute the inference task on the inference server under its own management based on the delay tolerance requirements of the network slice for the inference task, and, if it decides to execute the inference task on the inference server under its own management, controls the inference server under its own management to execute the inference task.

10. A program for causing a computer having a RAN control function for controlling a RAN (Radio Area Network) and an inference server for executing AI (Artificial Intelligence) inference processing to execute the following steps: a decision step for determining whether or not to execute an inference task corresponding to a network slice in response to receiving a request to execute the inference task based on the delay tolerance requirements of the network slice for the inference task; and a control step for controlling the inference server under the control of the computer to execute the inference task if it is determined in the decision step that the inference task will be executed by the inference server under the computer's management.

11. A control method executed by a computer having a RAN control function that controls a RAN (Radio Area Network) and an inference server that executes AI (Artificial Intelligence) inference processing, the control method comprising: a decision step that, in response to receiving a request to execute an inference task corresponding to a network slice, decides whether or not to execute the inference task in the inference server based on the delay tolerance requirements of the network slice for the inference task; and a control step that, if it is decided in the decision step that the inference task will be executed in the inference server under the control of the computer, controls the inference server under the control of the computer to execute the inference task.

Images