Energy-aware execution and allocation of services in mobile networks

Abstract

A method (1) of operating a scheduling function (2) of a functional component (6) of a mobile network (8) is disclosed. The method (1) comprises receiving (102) a message, the message comprising one of: a capability announcement message of a first execution node (82) of the mobile network (8), the capability announcement message comprising a default execution capability or an actual execution capability of the first execution node (82); a scheduling request message for a functional component (6) of a mobile network (8), the scheduling request message comprising an execution requirement of the functional component (6). The method (1) further comprises rescheduling (103A) one or more functional components (6) operating on a first execution node (82) of the mobile network (8) to one or more second execution nodes (82) of the mobile network (8) in response to the received capability advertisement message according to the execution requirements of the respective functional components (6) and the respective execution capabilities of each execution node (82) of the mobile network (8). The method (1) further comprises scheduling (103B) the functional component (6) to one or more second execution nodes (82) of the mobile network (8) according to the execution requirements of the functional component (6) and the respective execution capabilities of each execution node (82) of the mobile network (8) in response to the received scheduling request message. Therefore, energy awareness and zero-carbon resource scheduling are realized in the next generation of network.

Description

Technical Field

[0001] This disclosure relates to the field of mobile communications, and more particularly to various entities and methods for supporting energy-aware execution and allocation of services in mobile networks. Background Technology

[0002] Future 6G mobile networks are expected to provide a fully virtualized platform to dynamically execute network functions and application logic based on containerized applications and / or virtual machines (VMs), while also taking into account the associated carbon footprint.

[0003] However, the use of carbon neutrality and renewable energy sources (such as solar, wind, and hydropower) presents challenges such as energy fluctuations, supply shortages, and even oversupply, making it extremely challenging to match available resources with demand. This often results in the use of less green energy (in cases of supply shortages) or the waste of excess energy (in cases of supply surpluses). Summary of the Invention

[0004] The aim is to overcome the above and other shortcomings.

[0005] The above and other objectives are achieved through the features of the independent claim. Other implementations will be apparent from the dependent claims, the description, and the drawings.

[0006] According to a first aspect of this disclosure, a method for scheduling functional components of a mobile network is provided. The method includes receiving a message comprising one of the following: a capability announcement message of a first execution node of the mobile network, the capability announcement message including the default execution capability or actual execution capability of the first execution node; and a scheduling request message for a functional component of the mobile network, the scheduling request message including the execution requirements of the functional component. The method further includes: in response to the received capability announcement message, rescheduling one or more functional components running on the first execution node of the mobile network to one or more second execution nodes of the mobile network according to the execution requirements of the corresponding functional component and the corresponding execution capability of each execution node of the mobile network. The method further includes: in response to the received scheduling request message, rescheduling the functional component to one or more second execution nodes of the mobile network according to the execution requirements of the functional component and the corresponding execution capability of each execution node of the mobile network.

[0007] By matching the execution capabilities (including energy characteristics) of execution nodes with the execution requirements of functional components to be instantiated, energy awareness and zero-carbon resource scheduling can be achieved in next-generation networks. Therefore, the scheduling function can deploy functional components in the optimal location to meet execution requirements, thus achieving zero-carbon resource scheduling.

[0008] The proposed solution can be applied to systems with fluctuating energy and / or workloads that may have different characteristics (i.e., differences in energy characteristics based on location, differences in workload requirements based on workload type, deadlines, time sensitivity, location sensitivity, etc.). This may be relevant to data centers, wired communications, etc. Specifically, the mobile core architecture will be able to perform energy-aware scheduling and execution on heterogeneous drive platforms and base stations.

[0009] The mobile network used in this article can refer to a telecommunications network that has a wireless link between a fixed network infrastructure and a mobile terminal node.

[0010] The network function or NF used in this document can refer to a functional element within a virtualized network architecture (e.g., a 3GPP 5G / 6G system architecture), characterized by well-defined external interfaces and well-defined functional behaviors.

[0011] The functional components used in this article may refer to network functions (see above) or applications.

[0012] The scheduling function used in this article may refer to a specific NF in the control plane of the 3GPP system architecture, which is used to instantiate other functional components on demand based on the available processing, storage and / or network resources of the virtualized mobile network.

[0013] The execution node used in this article can refer to a structural element of the network architecture used to expose processing, storage, and / or network resources to the virtualized mobile network.

[0014] The default execution capability used in this article refers to the execution capability used when there is no actual execution capability.

[0015] The actual execution capacity used in this article may refer to temporary execution capacity. For example, the computing capacity of an execution node can vary depending on the availability of renewable energy.

[0016] In one possible implementation, rescheduling or scheduling may include: matching the default execution capability of each execution node with the execution requirements; matching the actual execution capability of any execution node with matching default execution capability with the execution requirements; selecting one or more second execution nodes from any execution node with matching actual execution capability according to a scheduling policy; balancing the execution load among the one or more second execution nodes; sending corresponding scheduling commands for functional components to the one or more second execution nodes; and receiving corresponding scheduling confirmations for functional components from the one or more second execution nodes.

[0017] The scheduling strategy used in this paper can refer to a specific action process selected from alternative options based on given conditions, in order to guide and determine current and future (scheduling) decisions.

[0018] The equilibrium used in this article can refer to an attempt to distribute multiple operations evenly.

[0019] In one possible implementation, rescheduling or scheduling may also include: resetting the countdown timer; and continuing the selection process in response to the expiration of the countdown timer.

[0020] According to a third aspect of this disclosure, a method for operating the proxy function of an execution node in a mobile network is provided. The method includes: sending one or more capability announcement messages of the execution node to a scheduling function of the mobile network. The capability announcement messages include one of the following: a default execution capability of the execution node and an actual execution capability of the execution node. The method further includes: receiving a scheduling command for a functional component of the mobile network from the scheduling function. The method further includes starting the functional component on the execution node. The method further includes sending a scheduling confirmation for the functional component to the scheduling function.

[0021] The proxy function used in this article may refer to a specific NF of the control plane in the 3GPP system architecture, for example, that is adapted to work on another entity (e.g., an execution node) that is not part of the control plane.

[0022] In one possible implementation, the method may further include receiving one or more of the following from the execution node: the default execution capability of the execution node and the actual execution capability of the execution node.

[0023] According to a fifth aspect of this disclosure, a method for operating a functional component of a mobile network is provided. The method may include: sending a scheduling request message for a functional component of the mobile network to a scheduling function of the mobile network, the scheduling request message including an execution requirement for the functional component.

[0024] In one possible implementation, the default execution capacity may include: the identifier of the execution node, the state of the execution node, the default computing capacity of the execution node, the default energy capacity of the execution node, the default energy capacity prediction of the execution node, the default carbon emissions per gigabyte (GB) of communication, the default carbon emissions per gigabyte (GB) of storage, and the default carbon emissions per floating point operation per second (FLOPS) of computation.

[0025] In one possible implementation, the actual execution capability may include: the identifier of the execution node, the state of the execution node, the actual computing capacity of the execution node, the actual energy capacity of the execution node, the percentage of the actual energy capacity at the execution node that is related to renewable energy supply and exceeds general energy demand, the predicted actual energy capacity of the execution node, the actual carbon emissions per gigabyte (GB) of communication, the actual carbon emissions per gigabyte (GB) of storage, the actual carbon emissions per floating point operation per second (FLOPS) of computation, the identifier of the event based on the capability announcement message, and the validity period of the actual execution capability.

[0026] In one possible implementation, events may include periodic events.

[0027] In one possible implementation, the scheduling strategy may include one of the following: the closest match between the execution requirements and their applicable execution capabilities, the nearest geographical proximity to the user of the functional component, the lowest latency to the user of the functional component, the best performance associated with the execution of the functional component (selecting the location with the best performance when multiple options have the same energy characteristics), the lowest energy cost associated with the execution of the functional component (selecting the location with the lowest energy cost when multiple options have the same energy characteristics), and the lowest carbon emissions associated with the execution of the functional component.

[0028] In one possible implementation, the execution requirements may include: the identifier of the functional component, the job type of the functional component, the quality of service (QoS) of the functional component, the preferred location, the preferred energy, the processing requirements of the functional component, the maximum carbon emissions per gigabyte (GB) of communication, the maximum carbon emissions per gigabyte (GB) of storage, and the maximum carbon emissions per floating point operation per second (FLOPS) of computation.

[0029] In one possible implementation, the job type of the functional component may include one of the following: core network function (NF), application and machine learning workloads.

[0030] In one possible implementation, the job QoS of a functional component may include one of the following: 3GPP QoS level, application QoS level, and machine learning workload QoS level.

[0031] The 3GPP QoS level used in this article can refer to a specific combination of throughput, latency, and / or packet loss rate, represented by a QoS flow identifier (QFI).

[0032] The application QoS level used in this article can refer to a specific combination of throughput, latency, and / or packet loss rate.

[0033] The QoS level for machine learning workloads used in this article can refer to a specific combination of throughput and / or packet loss rate (due to insensitivity to latency).

[0034] In one possible implementation, the functional components may include one of the following: network function (NF) and application.

[0035] According to a seventh aspect of this disclosure, a computer program is provided, including program code that, when executed on a computer, performs the method of the first aspect, the method of the third aspect, or the method of the fifth aspect.

[0036] According to a second aspect of this disclosure, a scheduling function for functional components of a mobile network is provided. The scheduling function is used to perform the method of scheduling functional components of the mobile network according to the first aspect.

[0037] According to a fourth aspect of this disclosure, a proxy function for an execution node of a mobile network is provided. The proxy function is used to perform the proxy function of the execution node of the mobile network as described in the third aspect.

[0038] According to a sixth aspect of this disclosure, a functional component of a mobile network is provided. The functional component is used to perform the method of operating the functional component of the mobile network as described in the fifth aspect.

[0039] According to the eighth aspect of this disclosure, a mobile network is provided, including a network repository function (NRF), an execution node, and functional components of the mobile network according to the sixth aspect. The NRF includes scheduling functions for the functional components of the mobile network according to the second aspect. The execution node includes a user plane function (UPF). The UPF further includes proxy functions for the execution node of the mobile network according to the fourth aspect.

[0040] The term "Network Storage Function" or NRF as used in this article may refer to a specific NF in the control plane of the 3GPP system architecture, used to discover available NFs and the services they support.

[0041] The term "User Plane Function" or UPF used in this article may refer to the NF of the data / user plane in the 3GPP system architecture, which is used to process (e.g., packet forwarding, policy enforcement) user data between the radio access network and the data network.

[0042] According to a ninth aspect of this disclosure, an edge data network for a mobile network is provided. The edge data network includes an edge application server (EAS), an execution node, and functional components of the sixth aspect. The EAS includes scheduling functions for the functional components for the mobile network of the second aspect. The execution node includes an edge enabler server (EES). The EES further includes proxy functions for the execution node for the mobile network of the fourth aspect.

[0043] The edge (data) network or EDN used in this article can refer to the architectural modifications made to the 3GPP system architecture according to ETSI Technical Specification 123 558 V17.3.0 (see Section 6.2), and is used for computing and data storage as close as possible to the request point (i.e., the user) to provide low latency and save bandwidth.

[0044] The edge application server (EAS) used in this article can refer to an application server conforming to ETSI technical specification 123 558 V17.3.0 (see section 6.3.6). Client applications residing in the UE connect to the EAS to leverage the services offered by edge computing.

[0045] The edge enabler server (EES) used in this article can refer to a server that provides support functions according to ETSI technical specification 123 558 V17.3.0 (see section 6.3.2).

[0046] According to a tenth aspect of this disclosure, a machine learning system is provided, comprising a scheduling function for a functional component of a mobile network as described in the second aspect, one or more execution nodes having a renewable energy supply, and one or more instances of a functional component of a mobile network as described in the sixth aspect. The corresponding execution node includes an agent function for a corresponding execution node of a mobile network as described in the fourth aspect. The corresponding instances of the functional component include an artificial neural network (ANN) for segmentation learning based on a machine learning workload.

[0047] The renewable energy supply used in this article can refer to energy supply from renewable energy sources such as wind, hydro, and solar power.

[0048] The machine learning used in this paper can refer to a class of methods used to transform sample data (i.e., input data combined with desired output data) into a statistical model, which can even be used to predict unseen samples.

[0049] The artificial neural network (ANN) used in this paper can refer to a specific machine learning method based on the forward propagation of the input data through the ANN and the subsequent backpropagation of the error of the obtained output data relative to the expected output data.

[0050] The segmentation learning used in this article can refer to a specific machine learning method in which a deep neural network (i.e., with multiple neuron layers) is segmented into multiple parts, each of which can be located and trained on different entities / devices.

[0051] In one possible implementation, the supply of renewable energy can exceed the general energy demand at the corresponding execution node. Attached Figure Description

[0052] The aspects and implementations described above will now be explained with reference to the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements.

[0053] The accompanying drawings should be considered as schematic diagrams, and the elements shown are not necessarily displayed to scale. Rather, the various elements are represented as their function and general purpose will become apparent to those skilled in the art.

[0054] Figure 1 A method for scheduling functions of functional components operating a mobile network according to this disclosure is shown;

[0055] Figure 2 It shows Figure 1 The interaction between a more detailed implementation of the method and the method for performing proxy functions of an execution node in a mobile network according to this disclosure;

[0056] Figure 3 It shows Figure 1 The interaction between a more detailed implementation of the method and a method for operating functional components of a mobile network according to this disclosure;

[0057] Figure 4 A mobile network according to this disclosure is shown;

[0058] Figure 5 An edge data network for mobile networks according to this disclosure is shown;

[0059] Figure 6 A machine learning system according to this disclosure is shown. Detailed Implementation

[0060] In the following description, specific aspects of implementations of the present disclosure or aspects in which implementations of the present disclosure may be used are illustrated by way of illustration with reference to the accompanying drawings, which form part of this disclosure. It is understood that implementations of the present disclosure may be used in other ways and include structural or logical variations not shown in the drawings. Therefore, the following detailed description is not intended to be limiting, and the scope of the present disclosure is defined by the appended claims.

[0061] For example, it should be understood that the disclosure relating to the described method also applies to the corresponding apparatus or system for performing the method, and vice versa. For instance, if one or more specific method steps are described, the corresponding apparatus may include one or more units, such as functional units for performing the described one or more method steps (e.g., a unit performing one or more steps; or multiple units, each performing one or more of the multiple steps), even if such one or more units are not explicitly described or shown in the accompanying drawings. Furthermore, for example, if a specific apparatus is described based on one or more units (e.g., functional units), the corresponding method may include a step for performing the function of one or more units (e.g., a step performing the function of one or more units; or multiple steps, each performing the function of one or more of the multiple units), even if such one or more steps are not explicitly described or shown in the accompanying drawings. Further, it is understood that, unless otherwise explicitly stated, features of the various exemplary implementations and / or aspects described herein can be combined with each other.

[0062] Figure 1 A method 1 is shown for scheduling function 2 of functional component 6 of mobile network 8 according to the present disclosure.

[0063] The mobile network 8 is represented by a dashed rectangle and includes related functional elements, namely the functional component 6 on the left, the scheduling function 2, and the multiple proxy functions 4 of the corresponding execution node 82 on the right.

[0064] Method 1 includes step 102, receiving a message. The message may be a capability announcement message from the first execution node 82 of the mobile network 8 (and arrives from one of the right-hand proxy functions 4) or a scheduling request message for the functional component 6 of the mobile network 8 (and arrives from the left-hand functional component 6).

[0065] Capability announcement messages (if any) include announcements of the default or actual execution capabilities of the first execution node 82.

[0066] The scheduling request message (if any) includes the execution requirements of functional component 6.

[0067] In response to the received capability notification message, method 1 further includes step 103A: rescheduling one or more functional components 6 running on the first execution node 82 of the mobile network 8 to one or more second execution nodes 82 of the mobile network 8 according to the execution requirements of the corresponding functional component 6 and the corresponding execution capabilities of each execution node 82 of the mobile network 8.

[0068] It should be noted that one or more second execution nodes 82 may include the first execution node 82. In other words, receiving a capability announcement message does not necessarily mean that one or more functional components 6 running on the first announcement execution node 82 will be actually rescheduled.

[0069] In response to the received scheduling request message, method 1 further includes step 103B: scheduling the functional component 6 to one or more second execution nodes 82 of the mobile network 8 according to the execution requirements of the functional component 6 and the corresponding execution capabilities of each execution node 82 of the mobile network 8.

[0070] In summary, the scheduling request of functional component 6 triggers the scheduling of an instance of functional component 6 (103B), while the capability announcement of proxy function 4 triggers the rescheduling of the instance of functional component 6 running on execution node 82 represented by proxy function 4 (103A).

[0071] Figure 2 It shows Figure 1 The interaction between the more detailed implementation of method 1 and the agent function 4 of the execution node 82 of the mobile network 8 according to this disclosure.

[0072] On the proxy function 4 and its execution node 82 side, method 3 may include step 301: receiving one or more of the following from execution node 82: the default execution capability of execution node 82 and the actual execution capability of execution node 82.

[0073] Default execution capabilities may include: a unique identifier for execution node 82, the status of execution node 82 (e.g., online, offline), the default computing capacity of execution node 82 (in floating point operations per second, FLOPS), the default energy capacity of execution node 82 (in kW for the current notification period), the default energy capacity prediction for execution node 82 (in kW for the next notification period), the default carbon emissions per gigabyte (GB) of communication (in kg / GB), the default carbon emissions per gigabyte (GB) of storage (in kg / GB), and the default carbon emissions per FLOPS of computation (in kg / GFLOPS).

[0074] Actual execution capability may include: a unique identifier for execution node 82, the status of execution node 82, the actual computing capacity of execution node 82 (in FLOPS), the actual energy capacity of execution node 82 (in kW), the percentage of actual energy capacity at execution node 82 that is related to renewable energy supply and exceeds general energy demand (i.e., excess energy expressed as %), the predicted actual energy capacity of execution node 82 (in kW), the actual carbon emissions per gigabyte (GB) of communication (in kg / GB), the actual carbon emissions per gigabyte (GB) of storage (in kg / GB), the actual carbon emissions per FLOPS calculated (in kg / GFLOPS), the identifier of the event based on the capability announcement message, and the validity period of the actual execution capability (i.e., the current announcement period).

[0075] For example, the events that trigger capability notification messages may include periodic events, and the event identifier may indicate this.

[0076] Method 3 further includes step 302: sending one or more capability announcement messages of execution node 82 to the scheduling function 2 of mobile network 8.

[0077] The corresponding capability notification message includes one of the following: the default execution capability of execution node 82 and the actual execution capability of execution node 82.

[0078] It should be noted that, for example, the execution capability can be a recently received execution capability or a recently resent execution capability that is periodically resent.

[0079] As mentioned above Figure 1 The capability announcement of the proxy function 4 triggers the rescheduling (103A) of the instance of the functional component 6 running on the execution node 82 represented by the proxy function 4.

[0080] Therefore, on the scheduling function 2 side, the rescheduling (103A) step may include step 104, matching the default execution capabilities of each execution node 82 with the execution requirements. This completes a coarse filtering of the execution nodes 82 of the mobile network 8.

[0081] The rescheduling (103A) step may further include step 105, which matches the actual execution capability of any execution node 82 with matching default execution capabilities to the execution requirements. This achieves fine-grained filtering of execution nodes 82 that remain relevant to the mobile network 8.

[0082] The rescheduling (103A) step may also include step 106, selecting one or more second execution nodes 82 from any execution node 82 with matching actual execution capabilities according to the scheduling policy.

[0083] Specifically, the scheduling strategy may include one of the following: the closest match between the execution requirements and their applicable execution capabilities (possibly based on predefined metrics), the nearest geographical proximity to the user of functional component 6, the lowest latency relative to the user of functional component 6, the best performance associated with the execution of functional component 6, the lowest energy cost associated with the execution of functional component 6, and the lowest carbon emissions associated with the execution of functional component 6.

[0084] The rescheduling (103A) step may also include step 107, balancing the execution load among one or more second execution nodes 82. It should be noted that in the case of a single second execution node 82, the balancing step is meaningless and is skipped.

[0085] The rescheduling (103A) step may also include step 108: resetting the countdown timer to an expiration time after a specific period for periodic evaluation of optional optimization opportunities.

[0086] The rescheduling (103A) step may also include step 109, sending a corresponding scheduling command for functional component 6 to one or more second execution nodes 82.

[0087] On the agent function 4 and its execution node 82 side, method 3 further includes step 309, receiving a scheduling command for functional component 6 from scheduling function 2.

[0088] Method 3 also includes step 310, which involves starting functional component 6 on execution node 82.

[0089] Method 3 also includes step 311, sending a scheduling confirmation for functional component 6 to scheduling function 2.

[0090] On the scheduling function 2 side, the rescheduling (103A) step may also include step 111, receiving a corresponding scheduling confirmation for functional component 6 from one or more second execution nodes 82.

[0091] In response to the scheduling confirmation, method 1 proceeds to the receiving (102) step.

[0092] However, the rescheduling (103A) step may also include the following step 112: continuing the selection (106) step in response to the expiration of the countdown timer in step 108.

[0093] It should be noted that the continue (112) step can also be merged into the receive (102) step, thus the receive (102) step acts as a "callback state".

[0094] Figure 3 It shows Figure 1 The interaction between a more detailed implementation of method 1 and method 5 of the functional component 6 of operating mobile network 8 according to this disclosure.

[0095] Functional component 6 may include one of the following: network function (NF) and application.

[0096] On the functional component 6 side, method 5 includes step 502, sending a scheduling request message for functional component 6 of mobile network 8 to scheduling function 2 of mobile network 8.

[0097] The scheduling request message includes the execution requirements of functional component 6.

[0098] The execution requirements may include: a unique identifier for functional component 6, the job type of functional component 6, the quality of service (QoS) of functional component 6, preferred location, preferred energy source (e.g., wind, hydro, and solar), processing requirements for functional component 6, maximum carbon emissions per gigabyte (GB) of communication (in kg / GB), maximum carbon emissions per gigabyte (GB) of storage (in kg / GB), and maximum carbon emissions per floating point operation per second (FLOPS) of computation (in kg / GFLOPS).

[0099] The job types for functional component 6 can include one of the following: core network functions, applications, and machine learning workloads.

[0100] The job QoS of functional component 6 may include one of the following: 3GPP QoS level, application QoS level, machine learning workload QoS level (i.e., non-latency sensitive).

[0101] As mentioned above Figure 1 The scheduling request of functional component 6 triggers the scheduling of an instance of functional component 6 (103B).

[0102] It should be noted that the scheduling (103A) step and the rescheduling (103A) step attempt to map functional component 6 and its execution requirements to execution node 82 and its execution capabilities, respectively.

[0103] Therefore, on the scheduling function 2 side, the scheduling (103B) steps may include those already combined Figure 2 Steps 104 to 112 are explained in more detail.

[0104] Figure 4 A mobile network 8 according to this disclosure is shown.

[0105] According to the described 3GPP implementation, functional component 6, scheduling function 2 and proxy function 4 can be executed in the core network of mobile network 8, respectively.

[0106] Therefore, the functional component 6 to be instantiated can be included in the mobile network 8, the scheduling function 2 can be included in the network repository function (NRF) 81 of the mobile network 8, and the proxy function 4 can be included in the user plane function (UPF) 821 of the mobile network 8. The UPF 821 itself can be included in the execution node 82 represented by the proxy function 4. Therefore, the instance of functional component 6 can be hosted by the UPF 821, which in turn is hosted by the exemplary execution node 82.

[0107] It should be noted that the identifier is Nnrf The UPF-NRF interface is extended based on the default and actual execution capabilities of the execution node 82. Specifically, this may involve modifications to 3GPP specification 29.510 (see sections 5.2.2.2NFRegister and 5.2.2.3NFUpdate).

[0108] Figure 5 An edge data network 9 of a mobile network 8 according to this disclosure is shown.

[0109] According to the described implementation, functional component 6 can be executed in the core network of mobile network 8, while scheduling function 2 and proxy function 4 can be executed in the edge data network 9 of mobile network 8, respectively.

[0110] To this end, the instantiated functional component 6 may be included in the mobile network 8, the scheduling function 2 may be included in the edge application server (EAS) 91 of the edge data network 9 of the mobile network 8, and the proxy function 4 may be included in the edge enabler server (EES) 92 of the edge data network 9. The EES 92 itself may be included in the execution node 82 represented by the proxy function 4. Therefore, an instance of functional component 6 may be hosted by the EES 92, which in turn is hosted by the exemplary execution node 82.

[0111] It should be noted that it is labeled EDGE-3 The EAS-EES interface is extended based on the default and actual execution capabilities of execution node 82. Specifically, this may involve modifications to 3GPP technical specification 23.558 (see sections 8.4.3.4.2 / Eees_EASRegistration_Request and 8.4.3.4.3 / Eees_EASRegistration_Update).

[0112] Figure 6 A machine learning system 10 according to this disclosure is shown.

[0113] Machine learning requires significant energy for training. Sustainability is crucial, which can be achieved by allowing machine learning workloads to be dedicated to data centers with renewable energy sources.

[0114] Therefore, the machine learning system 10 includes one or more execution nodes 82 that have a renewable energy supply and are represented by corresponding agent functions 4.

[0115] At the corresponding execution node 82, renewable energy supply can exceed general energy demand. In particular, the period of excessive renewable (and carbon-neutral) energy supply may be used for machine learning purposes.

[0116] Based on the described implementation method, one or more instances of functional component 6 can be instantiated.

[0117] Relevant examples include artificial neural networks (ANNs) 61.

[0118] For example, the functional component 6, which includes a deep neural network, can be subdivided into multiple parts / instances that can be trained according to a segmentation learning method and subsequently used for segmentation inference.

[0119] Machine learning system 10 also includes scheduling function 2 of functional component 6.

[0120] As described above, a scheduling request for functional component 6 triggers scheduling (103B) of one or more instances of functional component 6 at one or more execution nodes 82. In the case of multiple components / instances, the corresponding scheduling request can be applied.

[0121] One or more instances of functional component 6 can perform (segmented) learning processes using an excess of renewable energy supply, based on machine learning workloads.

[0122] The resulting one or more trained ANN 61s can later be used in the (segmentation) inference process.

[0123] This disclosure has been described in conjunction with various implementations. However, those skilled in the art, through practice of the claimed matters and by studying the accompanying drawings, this disclosure, and the independent claims, will understand and arrive at other variations. In the claims and the description, the word "comprising" does not exclude other elements or steps, and "an" does not exclude a plurality. A single element or other unit may fulfill the function of several entities or items listed in the claims. The listing of certain measures in dissimilar dependent claims does not imply that combinations of these measures cannot be used in advantageous implementations. Computer programs may be stored or distributed on suitable media, such as optical storage media or solid-state media provided together with or as part of other hardware, and may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

Claims

1. A method (1) for operating a scheduling function (2), the scheduling function being used for a functional component (6) of a mobile network (8), the method (1) comprising: - Receive message (102), which includes one of the following: - A capability announcement message of the first execution node (82) of the mobile network (8), the capability announcement message including the default execution capability or actual execution capability of the first execution node (82); - A scheduling request message for the functional component (6) of the mobile network (8), the scheduling request message including the execution requirements of the functional component (6); - In response to the received capability notification message, based on the execution requirements of the corresponding functional component (6) and the corresponding execution capabilities of each execution node (82) of the mobile network (8), one or more functional components (6) running on the first execution node (82) of the mobile network (8) are rescheduled (103A) to one or more second execution nodes (82) of the mobile network (8); - In response to the received scheduling request message, the functional component (6) is scheduled (103B) to one or more second execution nodes (82) of the mobile network (8) according to the execution requirements of the functional component (6) and the corresponding execution capabilities of each execution node (82) of the mobile network (8).

2. The method (1) according to claim 1, wherein, Rescheduling (103A) one or more functional components (6) running on the first execution node (82) of the mobile network (8) to one or more second execution nodes (82) of the mobile network (8), or scheduling (103B) the functional components (6) to one or more second execution nodes (82) of the mobile network (8), includes: - Match the default execution capabilities of each execution node (82) with the execution requirements (104); - Match the actual execution capability of any execution node (82) with the execution requirement with the execution requirement (105); - Select (106) one or more second execution nodes (82) from any execution node (82) with matching actual execution capabilities according to the scheduling policy; - Balance (107) the execution load among the one or more second execution nodes (82); - Send (109) a corresponding scheduling command for the functional component (6) to the one or more second execution nodes (82); - Receive (111) a corresponding scheduling confirmation for the functional component (6) from the one or more second execution nodes (82).

3. The method (1) according to claim 2, wherein, Rescheduling (103A) one or more functional components (6) running on the first execution node (82) of the mobile network (8) to one or more second execution nodes (82) of the mobile network (8), or scheduling (103B) the functional components (6) to one or more second execution nodes (82) of the mobile network (8), further includes: - Reset (108) countdown timer; - In response to the expiration of the countdown timer, continue with step (112) of selecting (106).

4. A method (3) for operating a proxy function (4), the proxy function being used by an execution node (82) of a mobile network (8), the method (3) comprising: - Send (302) one or more capability announcement messages of the execution node (82) to the scheduling function (2) of the mobile network (8), wherein the corresponding capability announcement message includes one of the following: - The default execution capability of the execution node (82), - The actual execution capability of the execution node (82); - Receive (309) a scheduling command for the functional component (6) of the mobile network (8) from the scheduling function (2); - Start (310) the functional component (6) on the execution node (82); - Send (311) a scheduling confirmation for the functional component (6) to the scheduling function (2).

5. The method (3) according to claim 4, further comprising: - Receive (301) one or more of the following from the execution node (82): - The default execution capability of the execution node (82), - The actual execution capability of the execution node (82).

6. A method (5) for operating a functional component (6), said functional component for a mobile network (8), said method (5) comprising: - Send (502) a scheduling request message for the functional component (6) of the mobile network (8) to the scheduling function (2) of the mobile network (8), the scheduling request message including the execution requirements of the functional component (6).

7. The method (1) according to claim 2 or 3, or the method (3) according to claim 4 or 5, or the method (5) according to claim 6, wherein, The default execution capabilities include: - The identifier of the execution node (82), - The state of the execution node (82), - The default computing capacity of the execution node (82), - The default energy capacity of the execution node (82), - The default energy capacity prediction of the execution node (82), - Default carbon emissions per gigabyte (GB) of communication - Default carbon emissions per gigabyte (GB) of storage - Calculates the default carbon emissions based on the number of floating-point operations per second (FLOPS).

8. The method (1) according to any one of claims 2, 3, and 7, or the method (3) according to any one of claims 4, 5, and 7, or the method (5) according to claim 6 or 7, wherein, The actual execution capability includes: - The identifier of the execution node (82), - The state of the execution node (82), - The actual computing capacity of the execution node (82), - The actual energy capacity of the execution node (82), - At the execution node (82), the percentage of actual energy capacity that is related to renewable energy supply and exceeds general energy demand. - The actual energy capacity prediction of the execution node (82), - Actual carbon emissions per gigabyte (GB) of communication - Actual carbon emissions per gigabyte (GB) of storage - Calculate the actual carbon emissions in floating-point operations per second (FLOPS). - The identifier of the event based on the capability announcement message. - The validity period of the actual execution capability.

9. The method (1) according to any one of claims 2 or 3 and 7 or 8, or the method (3) according to any one of claims 4 or 5 and 7 or 8, or the method (5) according to any one of claims 6 to 8, wherein, The events include periodic events.

10. The method (1) according to any one of claims 2 or 3 and 7 to 9, or the method (3) according to any one of claims 4 or 5 and 7 to 9, or the method (5) according to any one of claims 6 to 9, wherein, The scheduling strategy includes one of the following: - The closest match between the execution requirements and the applicable execution capabilities. - The user's nearest geographical proximity relative to the functional component (6), - The lowest latency for the user relative to the functional component (6), - Optimal performance associated with the execution of the functional component (6), - The lowest energy cost associated with the execution of the functional component (6), - The lowest carbon emissions associated with the execution of the functional component (6).

11. The method (1) according to any one of claims 2 or 3 and 7 to 10, or the method (3) according to any one of claims 4 or 5 and 7 to 10, or the method (5) according to any one of claims 6 to 10, wherein, The execution requirements include: - Identifier of the functional component (6), - The job type of the functional component (6), - The Quality of Service (QoS) of the functional component (6), - Preferred location, - Preferred energy source - The processing requirements of the functional component (6), - Maximum carbon emissions per gigabyte (GB) of communication - Maximum carbon emissions per gigabyte (GB) of storage - Calculate the maximum carbon emissions in floating-point operations per second (FLOPS).

12. The method (1) according to any one of claims 2 or 3 and 7 to 11, or the method (3) according to any one of claims 4 or 5 and 7 to 11, or the method (5) according to any one of claims 6 to 11, wherein, The job type of the functional component (6) includes one of the following: - Core Network Functions (NF) - Applications - Machine learning workloads.

13. The method (1) according to any one of claims 2 or 3 and 7 to 12, or the method (3) according to any one of claims 4 or 5 and 7 to 12, or the method (5) according to any one of claims 6 to 12, wherein, The job QoS of the functional component (6) includes one of the following: - 3GPP QoS level, - Apply QoS level, - QoS levels for machine learning workloads.

14. The method (1) according to any one of claims 2 or 3 and 7 to 13, or the method (3) according to any one of claims 4 or 5 and 7 to 13, or the method (5) according to any one of claims 6 to 13, wherein, The functional component (6) includes one of the following: - Network Functions (NF) - Application.

15. A computer program (7) comprising program code that, when executed on a computer, performs the method (1) according to any one of claims 2 or 3 and 7 to 14, or the method (3) according to any one of claims 4 or 5 and 7 to 14, or the method (5) according to any one of claims 6 to 14.

16. A scheduling function (2) for a functional component (6) of a mobile network (8), a method (1) for performing the scheduling function (2) for operating the functional component (6) of the mobile network (8) according to any one of claims 1 to 3 and 7 to 14.

17. A proxy function (4) for an execution node (82) of a mobile network (8) for performing a method (3) of the proxy function (4) of the execution node (82) of the mobile network (8) according to any one of claims 4 or 5 and 7 to 14.

18. A functional component (6) for a mobile network (8) for performing a method (5) of the functional component (6) for operating the mobile network (8) according to any one of claims 6 to 14.

19. A mobile network (8), comprising: - Network storage function NRF (81), the network storage function includes: - Scheduling function (2) of the functional component (6) for the mobile network (8) according to claim 16; - Execution node (82), the execution node includes: - User plane functionality UPF (821), the user plane functionality includes: - Proxy function (4) of the execution node (82) for the mobile network (8) according to claim 17; - Functional component (6) of the mobile network (8) according to claim 18.

20. An edge data network (9) for use in a mobile network, the edge data network (9) comprising: - Edge Application Server (EAS) (91), the edge application server comprising: - Scheduling function (2) of the functional component (6) for the mobile network (8) according to claim 16; - Execution node (82), the execution node includes: - Edge Enabled Server (EES) (92), the edge enabled server comprising: - Proxy function (4) of the execution node (82) for the mobile network (8) according to claim 17; - Functional component (6) of the mobile network (8) according to claim 18.

21. A machine learning system (10), comprising: - Scheduling function (2) of the functional component (6) for the mobile network (8) according to claim 16; - One or more execution nodes (82) with renewable energy supply, wherein the respective execution nodes (82) include: - Proxy function (4) of the corresponding execution node (82) for the mobile network (8) according to claim 17; - One or more instances of the functional component (6) of the mobile network (8) according to claim 18, wherein a corresponding instance of the functional component (6) includes: - Artificial Neural Network (ANN) (61) is used for segmentation inference and segmentation learning based on machine learning workloads.

22. The machine learning system (10) according to claim 21, wherein, At the corresponding execution node (82), the renewable energy supply exceeds general energy demand.

Images