Information processing device and requirement derivation method

Abstract

This information processing device derives a requirement of a slice subnet of each communication section constituting a network slice constructed across a plurality of communication sections in a communication network. The information processing device comprises a requirement derivation unit that derives the requirement of the slice subnet of each communication section by solving a multi-armed bandit problem in which each of a plurality of draft requirements is used as an arm, the requirement of the network slice is used as a context, and resources required for the network slice are used as a cost.

Description

Information processing device and requirements derivation method

[0001] The present invention relates to the provisioning of network slices in a communication network (e.g., a 5G network).

[0002] A 5G network consists of User Equipment (UE), base stations (gNB: gNodeB), a 5G Core (5GC), and a transport network connecting the gNB to the inside and outside of the 5GC. Data communications originating from the UE are connected to a Data Network (DN), such as a private network or the internet, via the gNB, transport network, and 5GC. In this specification, following ETSI ZSM 003 [1], the communication section between the UE and the gNB is referred to as the Access Network (AN), the communication section via the transport network is referred to as the Transport Network (TN), and the communication section within the 5GC is referred to as the Core Network (CN). In this specification, references are indicated by numbers such as [1] as above, and the names of the documents corresponding to the numbers are shown at the end of the specification.

[0003] 5G is expected to accommodate a variety of end-to-end (E2E) network requirements, such as eMBB, URLLC, and mMTC. However, due to cost and physical constraints, it is difficult to meet these requirements with a single 5G network. Therefore, network slicing, which constructs a logical network that satisfies each given E2E requirement, is attracting attention, and field experiments are underway.

[0004] E2E network slices are provided by the interconnection of logical networks (or virtual networks) called network slice subnets (or slice subnets) built on AN, TN, and CN.

[0005] M. Iannelli, MR Rahman, N. Choi, L. Wang, "Applying Machine Learning to End-to-end Slice SLA Decomposition," IEEE Conference on Network Softwareization (NetSoft), 2020. DD Vleeschauwer, C. Papagianni, A. Walid, "Decomposing SLAs for Network Slicing," IEEE Communications Letters, 2020. Masaki Kobayashi, Akihito Suzuki, Masahiro Kobayashi, "Examination of Requirements Decomposition Method for Network Slices in a Multi-Domain Environment," IEICE Technical Report, vol. 123, no. 273, CQ2023-38, pp.2-7, 2023.

[0006] In network slice provisioning (slice provisioning), the requirements for slice subnets in each communication segment are derived based on the network slice requirements, and resources in each communication segment are allocated to satisfy those requirements.

[0007] However, conventional technologies have made it difficult to perform efficient slice provisioning for the limited resources in a communication network.

[0008] This invention has been made in view of the above points, and aims to provide a technology for achieving efficient slice provisioning.

[0009] According to the disclosed technology, an information processing device is provided for deriving the requirements for slice subnets of each communication segment that constitute a network slice constructed across multiple communication segments in a communication network, the information processing device comprising a requirements derivation unit that derives the requirements for slice subnets of each communication segment by solving a multi-armed bandit problem in which multiple proposed requirements are represented as arms, the requirements of the network slice are represented as context, and the resources required for the network slice are represented as cost.

[0010] According to the disclosed technology, a technique is provided to enable efficient slice provisioning.

[0011] This figure shows the overview configuration of the architecture that realizes network slice control. This figure shows the network model. This figure shows the system configuration in the embodiment. This is a flowchart for explaining the operation flow of the requirements derivation device 100. This figure shows an example of the device's hardware configuration.

[0012] Hereinafter, embodiments of the present invention (this embodiment) will be described with reference to the drawings. The embodiments described below are merely examples, and the embodiments to which the present invention is applied are not limited to the embodiments described below.

[0013] The following sections will provide a more detailed explanation of network slicing and related issues, followed by a description of the technology related to this embodiment. For convenience, the text of this specification uses standard characters to represent vectors, sets, and other similar entities. It is clear from the context that these characters represent vectors and sets.

[0014] (Regarding network slicing) As mentioned above, 5G is expected to have use cases with diverse end-to-end (E2E) requirements for the network, such as eMBB, URLLC, and mMTC. However, from the perspective of cost and physical constraints, it is difficult to realize these requirements with a single 5G network. Therefore, network slicing, which constructs a logical network that satisfies each given E2E requirement, is attracting attention, and demonstration experiments are underway.

[0015] In this embodiment, it is assumed that the architecture based on ETSI ZSM 003 [1] is used as the architecture for realizing network slicing. However, the technology according to the present invention is applicable to various architectures, not limited to this architecture.

[0016] An overview of this architecture is shown in Figure 1. As shown in Figure 1, this architecture has a hierarchical structure consisting of two types of Management Domains (MDs). First, there are AN MD2, TN MD3, and CN MD4, which autonomously control resources in the AN, TN, and CN communication segments. Then, there is E2E Service MD1, which comprehensively manages these MDs. In this architecture, E2E network slices are provided by the interconnection of logical networks (which may also be called virtual networks) called network slice subnets built on the AN, TN, and CN.

[0017] Regarding the role of each MD, slice subnets are managed by MD2-4, which control each communication section. E2E Service MD1 is responsible for the interconnection of slice subnets and the provision of slices.

[0018] The APIs between each MD in this architecture are defined in ETSI ZSM 002 [2]. These APIs allow the E2E Service MD to obtain information on the physical network topology and resources, but delegate specific control to the MDs below it. Instead of a single MD directly controlling the resources of the entire 5G network, a hierarchical architecture is adopted consisting of MDs that manage the resources of each communication segment and an MD that oversees the entire network, with the aim of reducing slice management costs [2].

[0019] (Regarding Network Slice Provisioning) The flow of slice provisioning in this architecture is described below. Slice provisioning is the process of constructing slices. First, the slice provider designs the E2E requirements for the network slice as a Network Slice Type (NEST) [3]. Based on the designed NEST, the E2E Service MD derives the requirements for the network slice subnets in each communication section managed by the subordinate MDs. Specifically, the use of SliceProfile [4] defined by 3GPP (registered trademark) is assumed for AN and CN [1].

[0020] Based on the derived requirements, the E2E Service MD requests its subordinate MDs to provision network slice subnets. If a subordinate MD cannot meet the assigned requirements, the E2E Service MD will derive the requirements again.

[0021] Next, we will summarize and explain the key points for deriving network slice subnet requirements in E2E Service MD.

[0022] As an example of a specific requirement, we will explain using availability (the percentage of time during which other simultaneously specified requirements can be met). For example, if an availability requirement of 0.99 is specified for a slice, and the E2E Service MD has an availability requirement P for the slice subnet for the communication interval D ∈ {AN, TN, CN}, then... D Let's assume we have derived this. In this case, in order to achieve the E2E availability requirement, P AN ・P TN ・P CNThe value must be ≥0.99. Furthermore, if extremely high availability requirements are imposed on a specific communication segment, many resources will be needed to meet those requirements. This will reduce the number of provisionable slice subnets in that communication segment, potentially leading to a decrease in the number of successfully provisioned slices. Thus, when deriving slice subnet requirements, it is necessary to consider both satisfying the slice requirements specified in E2E and the resources consumed by the derived results.

[0023] (Purpose of the technology according to this embodiment) As will be described later, this embodiment proposes a requirements derivation algorithm for network slice subnets in a hierarchical architecture and its configuration system. The purpose of the technology according to this embodiment is to maximize the number of network slices that are successfully provisioned by optimizing the requirements derivation of slice subnets in E2E Service MD. Note that the "number of network slices that are successfully provisioned" may also be called the "number of slice provisioned slices".

[0024] (Related Technologies) Here, we will explain related technologies. Specifically, we will explain technologies related to the derivation of slice subnet requirements in hierarchical architectures. First, references [5] [6] (Non-Patent Literature 1, 2) disclose methods for deriving slice subnet requirements using machine learning and risk models for slice requirements such as latency and reliability. However, these methods assume the existence of abundant training data through the generalization of slicing. Therefore, they cannot be applied in the current situation where network slicing is still in the experimental stage.

[0025] Next, reference [7] (Non-Patent Literature 3) discloses a latency requirement partitioning method based on the ETSI ZSM 003 [1] architecture. However, [7] does not consider maximizing the number of slice provisions by optimizing the derivation of slice subnet requirements. Therefore, although the derivation of slice subnet requirements in hierarchical architectures is achieved in references [5] [6] [7], there are challenges regarding the requirements of the algorithm.

[0026] (Regarding the requirements of the algorithm and the challenges to its implementation) Based on the objectives of the technology related to this embodiment and the trends in related technologies, the requirements for the slice subnet requirement derivation algorithm in a hierarchical architecture are summarized as follows.

[0027] <Requirements> Requirement 1: Maximize the number of slice provisions by optimizing the derivation of requirements. Requirement 2: Do not assume the existence of abundant training data. However, as shown below, these requirements are contradictory and difficult to satisfy simultaneously. Therefore, this embodiment addresses the difficulty of simultaneously satisfying these requirements and aims to solve it.

[0028] <Challenges> To fulfill Requirement 1, a model is needed regarding resource utilization when requesting provisioning of slice subnets from the MDs that control each communication segment, as a result of the requirement derivation. Two specific approaches to model construction are possible: a mathematical model approach based on a detailed analysis of the provisioning process in each MD, and a machine learning approach based on data regarding the behavior of each MD. However, since the MDs that manage each communication segment operate autonomously, their detailed internal behavior cannot be controlled, making the former approach difficult to apply. Therefore, the latter machine learning approach needs to be applied, but if we do not assume the existence of abundant training data, taking Requirement 2 into consideration, this approach is also difficult to apply. For the above reasons, it is difficult to satisfy both Requirement 1 and Requirement 2 simultaneously.

[0029] The following describes the technology related to this embodiment for solving the above-mentioned problems.

[0030] (Outline of Embodiment) In this embodiment, to address the challenges in deriving the requirements for a slice subnet, the requirements derivation device 100, described later, employs an online optimization approach as the algorithmic approach to be executed by the device, which acts according to a certain policy and improves the policy based on the results. Specifically, it performs a formulation based on linear Contextual Bandits with Knapsack (linCBwK [8]), which is a multi-armed bandit problem (MAB) with added resource constraints and context, and derives the requirements for a slice subnet using an algorithm based on this formulation. In other words, the requirements derivation device 100 derives the requirements for a slice subnet by solving a multi-armed bandit problem with added resource constraints and context.

[0031] The following describes linCBwK and its application to deriving requirements for slice subnets in this embodiment.

[0032] <linCBwK> linCBwK is a problem that adds context to Bandits with Knapsack (BwK) [9], which is a problem that adds resource constraints to MAB. Below, we will first explain MAB and then explain its development into linCBwK.

[0033] In MAB, the goal is to maximize the cumulative sum of rewards by sequentially selecting a single arm from multiple options (each option called an arm) that have an unknown reward distribution. The key to algorithms for MAB is balancing two contradictory policies: a well-balanced arm selection aimed at estimating rewards, and an efficient arm selection aimed at maximizing the sum of rewards.

[0034] BwK is an extension of MAB, adding resource constraints and a distribution of resource consumption (cost distribution) for each arm. Each time an arm is selected, resources are consumed according to the cost distribution, and a reward is obtained. The objective is to maximize the cumulative sum of rewards while satisfying the resource constraints, given that the reward and cost distributions are unknown.

[0035] Finally, context is a concept that extends MAB and refers to the feature vector input from the outside when selecting an arm. By changing the parameters of the reward distribution of an arm according to the context, it becomes possible to select an arm that reflects different features of the input. As a problem setting that extends BwK according to the context, linCBwK has been proposed [8].

[0036] <Application of Requirements Derivation for Slice Subnet of linCBwK> Explain how to incorporate the requirements derivation process of the slice subnet into the framework of linCBwK. First, design in advance the requirements derivation plans a i =(R i t,AN ,R i t,TN ,R i t,CN ) (i ∈ {1, ···, K}) for K slice subnets, and regard each requirements derivation plan a i as an arm. Next, use the requirements R t of the network slice given at time t as the context. The requirements R t of the network slice are, for example, availability requirements. Then, use r i to represent the success or failure when selecting the arm a t and attempting provisioning (r t ∈ {0, 1}) as the reward. Here, the reward for success is 1 and the reward for failure is 0.

[0037] Finally, as a result of attempting provisioning, use the vector c t ∈ R d arranging the d types of network resources required for the network slice as the cost.

[0038] Based on the above assumptions, it becomes possible to incorporate the requirements derivation process of the slice subnet into the framework of linCBwK. By utilizing existing algorithms for linCBwK, it is possible to achieve maximizing the number of network slice provisionings in requirement 1) without assuming the existence of rich learning data, and thus the problem is solved.

[0039] (Example) Below, as an example, the configuration and operation of deriving the requirements for slice subnets using the above method will be described in more detail.

[0040] First, an example of the configuration of a hypothetical 5G network will be described, and then an example of the operation of a system (requirements derivation device 100, etc.) using the technology of this embodiment will be described using that 5G network.

[0041] <Network Model and Input / Output> In this embodiment, the 5G network is assumed to have a tree-type topology G:=(V,E) with height h, where V is the set of nodes and E is the set of links connecting them. However, V AN , V TN , V CN Let V be the set of nodes in each communication interval, then V:=V AN ∪V TN ∪V CN As shown in Figure 2, in this embodiment, the specific nodes constituting each set are assumed to be a gNB in ​​AN, a router in TN, and a UPF responsible for data transfer in 5GC in CN. Also, in this embodiment, gNB A i Route L from to UPFu i ⊂E is assumed to be pre-designed.

[0042] The input to the requirements derivation device 100 is the requirements R for the network slice. t :=(A t ,B t ,P t ) will be provided online. t n is the range of the slice expansion. t Let B be a set of gNBs. t The mean μ t , variance σ t 2 The normal distribution of (B t =N(μ t ,σ t 2 )) and P t This is considered availability. Requirement R tUpon receiving the requirements, the requirements R of the network slice subnet for the communication interval D ∈ {AN, TN, CN} are determined by the requirements R t,D := (A t ,B t ,P t,D Derive ). Note that P t,D This represents the availability required for communication segment D. The derived requirements are passed to MD for the communication segment that meets those requirements.

[0043] Requirements for slice subnets R t,D Upon receiving the MD in each communication segment, it attempts to provision slice subnets to the underlying physical network. During provisioning, the requirements R t,D Set A of gNBs specified by t Identify the slice expansion range and, for the physical network included within that range, determine the bandwidth distribution B. t and availability P t Resources are allocated with this in mind. In this embodiment, the implementation and provisioning process of specific slice subnets in each communication segment is assumed to be as follows.

[0044] <Slice Subnets and Provisioning Process in AN> AN MD uses Resource Blocks (RBs) managed by gNB as resources and allocates these resources to each slice subnet. gNB A i ∈V AN To calculate the required number of RBs, first, the distribution of the bandwidth B t Availability P t,AN The bandwidth value b that satisfies this condition i Let Φ be the cumulative density function of the standard normal distribution, and b i =μ t +σ t ・Φ -1 (P t,AN ) is calculated. Then, using the relationship between RB and bandwidth in 3GPP TS 38.306

[0010] , bandwidth b is calculated. i Calculate the number of RBs required for implementation.

[0045] <Slice Subnets and Provisioning Process in TN> TN MD uses the bandwidth of the links in the transport network as a resource and allocates that bandwidth to each slice subnet. In calculating the bandwidth required for link e ∈ E, in this embodiment, the availability P required for link e is used. t,TN,e Set it as follows:

[0046] This allows each path L i Availability in TN is calculated as follows, and the availability requirement P is specified for TN. t,TN This will be realized.

[0047] Next, the bandwidth distribution B in link e e Regarding this, the size of the set of paths passing through e is n. e =#{L i | e∈L i Using}, B e =N(n e μ,n e σ t 2 ) Therefore, the bandwidth value to be allocated to link e should be as follows.

[0048] <Slice Subnets and Provisioning Process in CN> CN MD uses the bandwidth in UPF as a resource and allocates that resource to each slice subnet. Bandwidth distribution in UPFu B u Regarding this, the size of the base station set that u accommodates is n. u =#A t Using this, Bu = N(n u μ,n u σ t 2 ) And the bandwidth value to be allocated to UPF is set as follows.

[0049] <System Configuration Example>Figure 3 shows a system configuration example in this embodiment. As shown in Figure 3, this system includes a requirement derivation device 100 and a control unit 130 in the E2E Service MD, and a control unit 210 and a resource allocation unit 220 in each MD. Also, Figure 3 shows a physical network 300 that is the target of resource allocation.

[0050] Note that the control unit 130 may be included in the requirement derivation device 100. The requirement derivation device 100 may be referred to as an information processing device. Also, "control unit 130 + requirement derivation device 100" may be referred to as an information processing device.

[0051] The control unit 130, the control unit 210, and the resource allocation unit 220 may each be a device. For example, the control unit 130, the control unit 210, and the resource allocation unit 220 may each be referred to as a control device 130, a control device 210, and a resource allocation device 220.

[0052] Regarding the functional blocks belonging to the E2E Service MD, the control unit 130 is responsible for controlling the input / output to the E2E Service MD and controlling the requirement derivation device

[0053] As shown in Figure 3, the control unit 130 receives the requirement R of the network slice t and inputs the requirement and the resource (cost) to the requirement splitting device 100. Also, the control unit 130 receives the requirements R of the slice subnet from the requirement derivation device 100 t AN 、R t TN 、R t CN [[ID=2']

[0054] Also, the control unit 130 requests the control unit t AN 、R t TN 、R t CNPerform a slice subnet provisioning request based on this. Also, the control unit 130 receives the resource state of the physical network from the control unit 210.

[0055] The requirement derivation device 100 includes a requirement derivation unit 110 and a storage unit 120. The requirement derivation unit 110 cooperates with the storage unit 120 and executes an algorithm for linCBwK [8] by the operations shown in the flowchart of FIG. 4 described later.

[0056] In this embodiment, each arm of linCBwK represents a proposal for deriving availability requirements, and the requirement derivation unit 110 uses the slice requirement R given as context t to select the optimal arm a i =(P i,AN , P i,TN , P i,CN ). The storage unit 120 stores the rewards and costs observed in the past. The requirement derivation unit 110 reads the rewards and costs observed in the past from the storage unit and uses them for arm selection

[0057] Next, each functional unit constituting the subordinate MD will be described. The control unit 210 is responsible for controlling the input to the corresponding MD and managing the physical network 300 in each communication section. Also, the resource allocation unit 220 calculates the resources of the physical network 300 that satisfy the requirements R t,D of the slice subnet specified for the communication section D.

[0058] For example, if the control unit 210 and the resource allocation unit 220 are in the MD that manages the AN, the requirements R of the slice subnet are input from the control unit 210 to the resource allocation unit 220, and the requirements R t AN are input from the resource allocation unit From the control unit 210, the resources necessary for realizing R t AN are output.

[0059] <Execution flow> First, referring to the flowchart of FIG. 4, the operation flow of the requirement derivation device 100 will be described.

[0060] In S101, the requirement derivation unit 110 determines the requirements R of the network slicet Receive.

[0061] In S102, the requirements derivation unit 110 updates the parameters used to estimate the reward and cost based on the previously observed reward and cost read from the storage unit 120.

[0062] In S103, the requirements derivation unit 110 calculates estimated reward and estimated cost for each arm using the updated parameters and selects the appropriate arm. The estimated reward is the estimated reward that can be obtained if provisioning is performed based on the requirements corresponding to the arm in question, and the estimated cost is the estimated cost that can be obtained if provisioning is performed based on the requirements corresponding to the arm in question.

[0063] An appropriate arm is one that offers high rewards and low costs. For example, one could choose the arm where "estimated reward ÷ estimated cost" is the highest.

[0064] Subsequently, the requirements derivation unit 110 receives the reward and cost observed based on the provisioning attempted based on the obtained arm.

[0065] In S104, if resources are depleted, that is, if the observed cost exceeds the cumulative resource constraint, processing is interrupted; otherwise, in S105, the parameters used for learning are updated and processing continues. In this way, the algorithm for linCBwK [8] repeatedly selects arms while updating the reward and cost distribution for sequentially given inputs.

[0066] Next, we will explain the operation of the entire system. First, the requirements R derived by the requirements derivation device 100 in E2E Service MD t,D Regarding R t,D Availability requirement P included t,D P i AN ・P t TN ・P t CN ≥P t If the requirements for availability P for E2E slices are not met, tIt is not possible to satisfy this condition.

[0067] Therefore, the control unit 130 in the E2E Service MD is R t,D Availability requirement P included t,D P i AN ・P t TN ・P t CN ≥P t If it is determined that the requirement is not met, the reward and cost for deriving that requirement are both set to 0, and the requirement deriving device 100 is requested to derive the requirement again until the same arm is selected. When making the request, the requirement deriving device 100 is informed that the reward and cost for deriving that requirement are both 0.

[0068] Conversely, P i AN ・P t TN ・P t CN ≥P t If this is the case, the availability requirements for the slice are met.

[0069] Therefore, the control unit 130, i AN ・P t TN ・P t CN ≥P t If that is the case, then R t,D The MD passes on the resource allocation unit 220 allocates the resources necessary to realize the requirements, and the control unit 210 controls the R t,D The cost required to realize (specifically, the resources consumed) is observed, and this cost, along with the reward of 1, is passed to the requirements derivation unit 110 via the control unit 130. Subsequently, the slice subnet is provisioned, and the requirements R t The process is complete.

[0070] In this embodiment, the context may include the topology of the gNB or transport network to be expanded. Furthermore, while other MAB algorithms with added resource constraints and context exist besides linCBwK [8], these methods can also be incorporated into the requirements derivation device 1100.

[0071] (Example Hardware Configuration) Any of the devices described in this embodiment (information processing device, requirements derivation device, control device, resource allocation device) can be realized, for example, by having a computer run a program. This computer may be a physical computer or a virtual machine on the cloud.

[0072] In other words, the device can be realized by using hardware resources such as the CPU and memory built into a computer to execute a program corresponding to the processing performed by the device. The program can be recorded on a computer-readable recording medium (such as portable memory), saved, and distributed. It can also be provided via a network, such as the Internet or email.

[0073] Figure 5 shows an example of the hardware configuration of the computer described above. The computer in Figure 5 has a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, etc., all of which are interconnected by a bus BS. The computer may also be equipped with a GPU.

[0074] The program that enables processing on the computer is provided on a recording medium 1001, such as a CD-ROM or memory card. When the recording medium 1001 containing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001; it may also be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program as well as necessary files and data.

[0075] The memory device 1003 reads and stores a program from the auxiliary storage device 1002 when a program startup command is received. The CPU 1004 implements the functions related to the memory device 1003 according to the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to a network, etc. The display device 1006 displays a GUI (Graphical User Interface) etc., based on a program. The input device 1007 consists of a keyboard and mouse, buttons, or a touch panel, etc., and is used to input various operation commands. The output device 1008 outputs the calculation results.

[0076] (Summary of Embodiments, Effects, etc.) As described above, the technology described in these embodiments enables efficient slice provisioning with limited resources. As a result of efficient slice provisioning, slice providers can meet the needs of more customers.

[0077] More specifically, the technology according to this embodiment makes it possible to maximize the number of slice provisions in network slicing implemented by a hierarchical architecture, without assuming requirement 1) the existence of abundant training data.

[0078] In other words, in this embodiment, the challenge of simultaneously satisfying requirement 1 and requirement 2 was addressed by a formulation and algorithm based on linCBwK [8], which is a type of online optimization.

[0079] The following additional information is disclosed regarding the embodiments described above.

[0080] <Notes> (Note 1) An information processing device for deriving the requirements of a slice subnet for each communication segment that constitutes a network slice constructed across multiple communication segments in a communication network, comprising: a memory; and at least one processor connected to the memory, wherein the processor derives the requirements of a slice subnet for each communication segment by solving a multi-armed bandit problem in which multiple proposed requirements are each arm, the requirements of the network slice are the context, and the resources required for the network slice are the cost. (Note 2) The information processing device according to Note 1, wherein the processor selects one arm based on an estimated reward and an estimated cost, and receives the observed reward and cost for provisioning attempted based on the requirements corresponding to the selected arm. (Note 3) The information processing device according to Note 2, wherein in provisioning in a communication segment, resources for the communication segment are allocated based on bandwidth distribution and availability requirements specified for the communication segment. (Appendix 4) A requirements derivation method executed by an information processing device for deriving the requirements of slice subnets for each communication segment that constitute a network slice constructed across multiple communication segments in a communication network, comprising a requirements derivation step of deriving the requirements of slice subnets for each communication segment by solving a multi-armed bandit problem in which multiple proposed requirements are each arm, the requirements of the network slice are the context, and the resources required for the network slice are the cost. (Appendix 5) A non-temporary storage medium storing a program for causing a computer to function as an information processing device described in any one of Appendix 1 to 3.

[0081] Although this embodiment has been described above, the present invention is not limited to this specific embodiment, and various modifications and changes are possible within the scope of the gist of the invention as described in the claims.

[0082] [References] [1] ETSI ISG ZSM, "ETSI GS ZSM 003 v1.1.1," 2021. [2] ETSI ISG ZSM, "ETSI GS ZSM 002 v1.1.1," 2019. [3] GSMA, "Generic Network Slice Template Version 8.0," 2022. [4] 3GPP, "3GPP TS 28.541 version 17.4.0," 2021. [5] M. Iannelli, MR Rahman, N. Choi, L. Wang, "Applying Machine Learning to End-to-end Slice SLA Decomposition," IEEE Conference on Network Softwarization (NetSoft), 2020. [6] DD Vleeschauwer, C. Papagianni, A. Walid, "Decomposing SLAs for Network Slicing," IEEE Communications Letters, 2020. [7] Masaki Kobayashi, Akihito Suzuki, Masahiro Kobayashi, "Study on Requirements Decomposition Method for Network Slices in a Multi-Domain Environment," IEICE Technical Report, vol. 123, no. 273, CQ2023-38, pp.2-7, 2023. [8] S. Agrawal, NR Devanur, "Linear Contextual Bandits with Knapsacks," Conference on Neural Information Processing Systems (NIPS), 2016. [9] A. Badanidiyuru, R. Kleinberg, A. Slivkins, "Bandits with Knapsacks," IEEE 54th Annual Symposium on Foundations of Computer Science (FOCS), 2018.

[0010] 3GPP, "3GPP TS 38.306 version 17.0.0," 2022.

[0083] 100 Requirements Derivation Device 110 Requirements Derivation Unit 120 Storage Unit 130, 210 Control Unit 220 Resource Allocation Unit 300 Physical Network Unit 1000 Drive Device 1001 Recording Medium 1002 Auxiliary Storage Device 1003 Memory Device 1004 CPU 1005 Interface Device 1006 Display Device 1007 Input Device 1008 Output Device

Claims

1. An information processing device for deriving the requirements for slice subnets of each communication segment that constitute a network slice constructed across multiple communication segments in a communication network, comprising a requirements derivation unit that derives the requirements for slice subnets of each communication segment by solving a multi-armed bandit problem in which multiple proposed requirements are used as arms, the requirements of the network slice as the context, and the resources required for the network slice as the cost.

2. The information processing apparatus according to claim 1, wherein the requirements derivation unit selects one arm based on the estimated reward and the estimated cost, and receives the observed reward and cost for the provisioning attempted based on the requirements corresponding to the selected arm.

3. The information processing apparatus according to claim 2, wherein, in provisioning a communication segment, resources are allocated to the communication segment based on the bandwidth distribution and availability requirements specified for that communication segment.

4. A requirements derivation method executed by an information processing device for deriving the requirements of slice subnets for each communication segment that constitute a network slice constructed across multiple communication segments in a communication network, the requirements derivation method comprising a requirements derivation step of deriving the requirements of slice subnets for each communication segment by solving a multi-armed bandit problem in which multiple proposed requirements are used as arms, the requirements of the network slice as the context, and the resources required for the network slice as the cost.

Images