Computing power scheduling method, and device and medium

Abstract

The present application belongs to the technical field of communications. Disclosed are a computing power scheduling method, and a device and a medium. The computing power scheduling method comprises: in response to a computing power request message having been received, on the basis of a flow computing power forwarding table corresponding to the computing power request message, determining a network path and a service instance identifier which correspond to the computing power request message; and on the basis of the network path and the service instance identifier, forwarding the computing power request message to a corresponding computing instance.

Description

Computing power scheduling methods, equipment and media

[0001] Related applications

[0002] This application claims priority to Chinese patent application No. 202410801939.1, filed on June 20, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of communications, and more particularly to a computing power scheduling method, network equipment, and storage medium. Background Technology

[0004] The future development goal of computing power networks is to flexibly meet the integrated needs of various services for computing resources and network connectivity. The computing power network will schedule computing power request messages to appropriate paths and appropriate nodes in multiple computing instances to meet the requirements of the computing power network joint SLA (Service Level Agreement). Furthermore, it will simultaneously realize the load balancing requirements of the network and computing power resource pool to improve the utilization rate of computing power network resources.

[0005] Current computing network technologies often perform joint computing and network calculations based on reported network and computing instance status information to select the optimal service instance and network path to meet computing network SLA requirements and execute the guidance service for user computing power request messages. However, due to the imperfect scheduling mechanism, there is an imbalance in the use of computing network resources. Summary of the Invention

[0006] To achieve the above objectives, embodiments of this application provide a computing power scheduling method, the method comprising the following steps: in response to receiving a computing power request message, determining the network path and service instance identifier corresponding to the computing power request message based on the streaming computing power forwarding table corresponding to the computing power request message; and forwarding the computing power request message to the corresponding computing instance according to the network path and service instance identifier.

[0007] This application embodiment also provides a network device, the network device including: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the computing power scheduling method as described above.

[0008] This application embodiment also provides a storage medium, which is a computer-readable storage medium, and stores a computer program on the storage medium. When the computer program is executed by a processor, it implements the steps of the computing power scheduling method described above. Attached Figure Description

[0009] Figure 1 is a schematic diagram of the composition architecture of an existing computing power network;

[0010] Figure 2 is a schematic diagram of the structure of the operating device of the hardware operating environment involved in the embodiment of this application;

[0011] Figure 3 is a flowchart illustrating the computing power scheduling method according to the first embodiment;

[0012] Figure 4 is a schematic diagram of the computing power network system architecture in an embodiment of this application;

[0013] Figure 5 is a flowchart illustrating the computing power scheduling method according to the second embodiment;

[0014] Figure 6 is a schematic diagram of a computing power network system in an embodiment of this application;

[0015] Figure 7 is a flowchart illustrating the computing power scheduling method according to the third embodiment;

[0016] Figure 8 is a schematic diagram of the forwarding process in an embodiment of this application;

[0017] Figure 9 is a flowchart illustrating the computing power scheduling method according to the fourth embodiment;

[0018] Figure 10 is a schematic diagram of session-level payload packet encapsulation in an embodiment of this application;

[0019] Figure 11 is a schematic diagram of the computing power scheduling device provided in the embodiment of this application.

[0020] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0022] The future development goal of computing power networks is to flexibly meet the integrated needs of various services for computing resources and network connectivity. The computing power network will schedule service request packets to appropriate paths and appropriate nodes in multiple computing instances to meet the joint SLA requirements of computing power networks, and further improve the utilization rate of computing power network resources by simultaneously realizing the load balancing requirements of network and computing power resource pools.

[0023] Referring to Figure 1, which is a schematic diagram of the existing computing power network architecture, a complete and operational computing power network system typically includes three parts: the edge side, the network side, and the computing side. In fact, computing power perception can be achieved by the computing network brain directly connecting to the cloud management platform, or by the egress computing power gateway (EGW) directly connecting to the perception module deployed in the cloud resource pool. This is just a difference in perception methods, and there is no essential difference for the integrated computing network scheduling algorithm and the perception resource model. The following case will use the latter to illustrate the relevant steps.

[0024] Current computing network technologies often select the optimal service instance and network path based on reported network and computing instance status information, and perform guidance services for user computing power request messages to meet the requirements of the computing network service level agreement (SLA). However, this approach generates two main problems:

[0025] Issue 1: Imbalanced use of network resources. To alleviate pressure on the control plane, reporting cycles are typically extended or threshold reporting is used to slow down the update speed of the computing power METRIC. Therefore, within the time interval between two updates to the streaming computing power forwarding table, all computing power service requests originating from a specific ingress device will arrive at the same computing instance via a fixed network path, resulting in an extreme imbalance in the allocation of network and computing power resources. Although the network can use ECMP (Equal-cost multi-path routing) for load balancing, computing power requests will still be excessively concentrated on a single instance.

[0026] Problem 2: High load on the control plane. Sudden surges in traffic requests, due to uneven resource allocation, frequently exceed thresholds in network and computing resource usage, triggering frequent reports and table updates, placing a heavy computational burden on the control plane. While incremental computation and distribution strategies can reduce the pressure of calculating and updating routing tables to some extent, they do not fundamentally solve the root cause of the problem.

[0027] Given the specific application environment and scheduling requirements of computing networks, it is necessary to find a way to effectively solve these two problems while ensuring SLA, so as to achieve more balanced and efficient resource utilization.

[0028] To solve the above technical problems, refer to Figure 2, which is a schematic diagram of the operating device structure of the hardware operating environment involved in the embodiment of this application.

[0029] As shown in Figure 2, the operating device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen and an input unit such as a keyboard; the user interface 1003 may also include standard wired and wireless interfaces. The network interface 1004 may include standard wired and wireless interfaces (such as Wireless-Fidelity (Wi-Fi) interfaces). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0030] Those skilled in the art will understand that the structure shown in Figure 2 does not constitute a limitation on the operating equipment and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0031] As shown in Figure 2, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and a computer program.

[0032] In the operating device shown in Figure 2, the network interface 1004 is mainly used for data communication with other devices; the user interface 1003 is mainly used for data interaction with users; the processor 1001 and memory 1005 in the operating device of this application can be set in the operating device. The operating device calls the computer program stored in the memory 1005 through the processor 1001 and performs the following operations: in response to receiving a computing power request message, based on the streaming computing power forwarding table corresponding to the computing power request message, determines the network path and service instance identifier corresponding to the computing power request message; and forwards the computing power request message to the corresponding computing instance according to the network path and service instance identifier.

[0033] For example, before the step of determining the network path and service instance identifier corresponding to the computing power request message based on the streaming computing power forwarding table corresponding to the computing power request message, the method further includes: selecting each computing instance that provides the same service and conforms to the computing network service level agreement; obtaining the corresponding computing power status information and network status information according to the network path corresponding to each computing instance; and performing joint calculation based on the computing power status information and network status information to generate the streaming computing power forwarding table, wherein the streaming computing power forwarding table includes multiple next-hop information.

[0034] For example, the step of determining the network path and service instance identifier corresponding to the computing power request packet based on the streaming computing power forwarding table corresponding to the computing power request packet includes: identifying the virtual route forwarding identifier and packet 5-tuple corresponding to the computing power request packet, wherein the packet 5-tuple includes the destination IP address of the packet; performing entry matching on the streaming computing power forwarding table according to the virtual route forwarding identifier and the destination IP address of the packet; if a corresponding entry is matched, performing a hash operation on the packet 5-tuple to determine the next-hop information corresponding to the computing power request packet; and obtaining the network path and service instance identifier corresponding to the computing power request packet according to the next-hop information corresponding to the computing power request packet.

[0035] For example, after the step of obtaining the network path and service instance identifier corresponding to the computing power request message based on the next-hop information corresponding to the computing power request message, the method further includes: creating a flow affinity table based on the virtual route forwarding identifier, the message quintuple, the network path and the service instance identifier, for forwarding subsequent computing power request messages of the same flow to the computing instance.

[0036] For example, the method further includes the following steps: in response to receiving a computing power request message, determining the network path and service instance identifier corresponding to the computing power request message based on the flow affinity table corresponding to the computing power request message; and forwarding the computing power request message to the corresponding computing instance according to the network path and service instance identifier.

[0037] For example, the step of determining the network path and service instance identifier corresponding to the computing power request packet based on the flow affinity table corresponding to the computing power request packet includes: identifying the virtual route forwarding identifier and packet 5-tuple corresponding to the computing power request packet; determining whether a matching entry in the corresponding flow affinity table is found based on the virtual route forwarding identifier and packet 5-tuple; if a matching entry in the corresponding flow affinity table is found, obtaining the network path and service instance identifier based on the matching entry in the flow affinity table; if no matching entry in the corresponding flow affinity table is found, determining the network path and service instance identifier corresponding to the computing power request packet by querying the flow computing power forwarding table.

[0038] For example, the step of obtaining the corresponding computing power status information and network status information based on the network path corresponding to each computing instance includes: collecting the computing power status information of each computing instance through the computing power sensing component, and collecting the network status information of each network path through the network sensing component.

[0039] For example, after the step of collecting the computing power status information of each computing instance through the computing power sensing component, the method further includes: recording the computing power status information of each computing instance, wherein the computing power status information includes at least one computing power status index; and triggering computing power status diffusion and calculating and updating the streaming computing power forwarding table when the computing power sensing component detects that the change of the at least one computing power status index exceeds the corresponding preset gradient threshold.

[0040] For example, after the step of obtaining the corresponding computing power status information and network status information based on the network path corresponding to each computing instance, the method further includes: calculating or updating the multiple next-hop information based on the computing power status information and / or network status information, wherein the multiple next-hop information includes each network path, each service instance identifier and each next-hop sharing ratio, and the each next-hop sharing ratio is used to flatten and split each next hop.

[0041] For example, the step of forwarding the computing power request message to the corresponding computing instance based on the network path and service instance identifier includes: encapsulating a segment routing header through the ingress gateway based on the network path and service instance identifier; forwarding the computing power request message to the egress gateway based on the segment routing header; and decapsulating the segment routing header through the egress gateway and forwarding the computing power request message to the corresponding computing instance based on the service instance identifier.

[0042] For example, the step of forwarding the computing power request message to the corresponding computing instance based on the service instance identifier includes: identifying the service instance identifier; if the service instance identifier is a first type segment identifier, forwarding the computing power request message to the interface corresponding to the service instance identifier to deliver it to the corresponding computing instance; if the service instance identifier is a second type segment identifier, performing a table lookup in the corresponding logical network based on the logical network identifier indicated by the second type segment identifier and the destination IP address of the message corresponding to the computing power request message, and forwarding the computing power request message to the corresponding computing instance based on the table lookup result.

[0043] First Embodiment

[0044] Referring to Figure 3, which is a flowchart of the computing power scheduling method according to the first embodiment, the method includes the following steps: Step S10, in response to receiving a computing power request message, determining the network path and service instance identifier corresponding to the computing power request message based on the streaming computing power forwarding table corresponding to the computing power request message.

[0045] The computing power scheduling method in this application is applicable to IPv6 / SRv6-based computing power networks (including service-aware networks). It proposes a session-level service traffic load balancing method to achieve fine-grained balanced utilization of computing network resources while meeting the SLA of the computing network for user session quality. This avoids the "polarization" of excessive occupation of individual computing and network resources within the time period of two consecutive distributions of computing power forwarding tables, as well as the problem of high control plane pressure caused by frequent changes in the state of computing power instances.

[0046] Referring to Figure 4, which is a schematic diagram of the computing power network system architecture in this embodiment, the computing power network in this embodiment mainly includes service users, a bearer network, and a service provider. For example, the service user is a client that needs to use computing power services, typically a wireless or wired access terminal; the bearer network is an IP bearer network that provides computing power network services. The computing power network service provides the best network path and service instance selection for specific service requests. Typically, computing power gateway devices are added near the service user and service provider to provide computing power service functions, interconnected by an Underlay network (IP / MPLS) to reduce the intrusion of the Underlay network and facilitate implementation. The EGW (Egress Gateway) connects multiple computing resources and senses the status information of the computing resources, while the IGW (Ingress Gateway) provides computing power service access to the client. The IGW / EGW is usually upgraded from the bearer network router software to enhance the computing power network functionality; the service provider is typically a location server resource deployed close to the user to install various services that are extremely sensitive to latency and bandwidth (such as AR / VR), so that users can access them and obtain a better experience. A service instance is an instance resource that provides these business services; it can accept, process, and respond to service requests.

[0047] The computing network architecture in this embodiment fully complies with the SDN design philosophy. By separating the network control plane from the data forwarding plane, the control logic can be abstracted from the underlying network devices, enabling centralized control. It provides open interfaces and programmability, allowing network devices to be dynamically adjusted as needed and facilitating the integration of third-party applications and services (such as the computing network brain).

[0048] For example, when IGW receives a computing power request message, it can determine the network path and service instance identifier corresponding to the computing power request message by querying a pre-determined streaming computing power forwarding table.

[0049] Step S20: Forward the computing power request message to the corresponding computing instance according to the network path and service instance identifier.

[0050] In one embodiment, after determining the corresponding network path and service instance identifier based on the streaming computing power forwarding table corresponding to the computing power request message, the computing power request message can be forwarded to the corresponding computing instance according to the network path and service instance identifier.

[0051] For example, a service identifier is used to identify multiple compute instances providing the same service, typically using anycast IP addresses. When multiple compute instances providing the same service are connected to the same EGW, a simple anycast IP address cannot be forwarded on the EGW. In this case, tunnels must be configured between the EGW and each service instance for data transmission to avoid directly using anycast IP addresses for table lookup and forwarding. To instruct compute request packets to enter the correct tunnel for forwarding, an END.DX4 / END DX6 SID can be used as the service instance identifier to indicate that compute request packets should be forwarded to a specific compute instance. This Service SID is generally advertised to the head node in MP-BGP along with anycast IP routes so that when a routed compute service request packet matching the service identifier is received, it is encapsulated in an SRH (Segment Routing Header) to guide the next forwarding operation on the EGW.

[0052] If a single EGW is connected to only one compute instance providing a specific service, then anycast IPs can be used for forwarding from the EGW to the compute instance. In this case, depending on the type of service supported, the Service SID can be of various types, such as END.DT4, END.DT6, or END.DT46.

[0053] For example, the Service SID establishes a one-to-one mapping with computing instances in computing power routing. This way, the computing components in the control plane and the forwarding plane do not need to be aware of or carry the IP address information of the computing instances during service forwarding. The Service SID expresses both the forwarding behavior and the mapping to the computing instance. For distributed computing power routing, since the service identifier uses anycast IP, to maximize compatibility with existing router control plane and forwarding plane mechanisms, MP-BGP EVPN / L3VPN is used to advertise anycast IP routes to nodes while simultaneously carrying the Service SID as the VPN SID. For centralized computing power routing, the Service SID can be distributed to the IGW gateway by the computing network controller along with the anycast IP prefix.

[0054] This embodiment, through the above-described scheme, specifically responds to a received computing power request message by determining the network path and service instance identifier corresponding to the computing power request message based on the flow computing power forwarding table corresponding to the computing power request message; and forwards the computing power request message to the corresponding computing instance according to the network path and service instance identifier. Through a pre-determined flow computing power forwarding table and / or flow affinity table, the network path and service instance identifier corresponding to the computing power request message can be provided. The flow computing power forwarding table ensures that the computing power request message can be correctly routed to the appropriate computing instance, providing a complete computing power scheduling mechanism to achieve more balanced and efficient utilization of computing network resources.

[0055] Second Embodiment

[0056] Referring to Figure 5, which is a flowchart illustrating the computing power scheduling method according to the second embodiment, this embodiment, based on the aforementioned first embodiment, further includes the following steps before step S10: Step S01, selecting computing instances that provide the same service and conform to the computing network service level agreement; Step S02, obtaining corresponding computing power status information and network status information according to the network path corresponding to each computing instance; Step S03, performing joint calculation based on the computing power status information and network status information to generate the streaming computing power forwarding table, wherein the streaming computing power forwarding table includes multiple next-hop information.

[0057] For example, the step of obtaining the corresponding computing power status information and network status information based on the network path corresponding to each computing instance includes: collecting the computing power status information of each computing instance through the computing power sensing component, and collecting the network status information of each network path through the network sensing component.

[0058] In this embodiment of the application, in order to achieve integrated scheduling of computing and network, the computing power network can support different deployment models such as centralized, distributed, and hybrid (centralized + distributed). Dynamic computing power scheduling is achieved by adding new components, including computing power awareness components, network awareness components, computing network computing components, and computing power routing and forwarding components.

[0059] For example, the computing power awareness component is typically deployed on the EGW (refer to Option 2 distributed deployment in Figure 4), or it can be deployed independently on the computing network controller (refer to Option 1 centralized deployment in Figure 4). It is used to collect computing service status information, which typically includes resource utilization, slow request percentage, average service completion time, etc. Computing power information usually changes frequently; to avoid excessive network pressure from frequent updates, the information is usually compressed according to long periods and / or thresholds. In a distributed deployment, to reduce the impact of undelayed network devices, the BGP protocol is typically used to overlay and announce the computing status information to the IGW nodes. Regardless of whether the computing power awareness component is deployed centrally or distributed, the ultimate goal is to send the relevant computing power status information to the computing network computing component for further decision-making.

[0060] For example, network awareness components are typically deployed on various network node devices to collect network status information disseminated by underlay network devices and computing gateways. This information includes link bandwidth, link physical latency, link occupancy, etc. It usually uses the IGP protocol to perform intra-domain TE dissemination, forming a TE-DB on each network node. It can also be deployed on the computing network controller, using BGP-LS / Telemetry to achieve network status awareness and form a global TE-DB. Whether it is a centralized or distributed deployment, the final goal is to send the relevant network status information to the computing network computing components for further decision-making.

[0061] For example, the computing network component is typically deployed in the IGW, but it can also be deployed in the centralized computing network controller. In some cases (e.g., cross-domain), it must be deployed in the centralized computing network controller. Based on the status information collected from the computing power awareness component and the network awareness component, the computing network SLA constraints are used to calculate the computing power forwarding table through the computing network computing component, and then the table is sent to the forwarding component for computing power service request traffic.

[0062] For example, the computing power routing and forwarding component is typically deployed on the IGW / EGW gateway to identify user computing power service request traffic and encapsulate and forward service requests according to the best path and service instance provided by the computing power forwarding table. The undelay network generally does not explicitly distinguish service request traffic and only forwards it according to the SRH carried in the packet.

[0063] Exemplarily, after the step of collecting the computing power state information of each computing instance by the computing power perception component, the following steps are further included: recording the computing power state information of each computing instance, where the computing power state information includes at least one computing power state indicator; when the computing power perception component senses that the change of the at least one computing power state indicator exceeds the corresponding preset gradient threshold, triggering the diffusion of the computing power state and calculating and updating the flow computing power forwarding table.

[0064] Exemplarily, due to the strong time-varying nature of the computing power state, in conventional cases, the computing power is often collected at a fixed long period or reported based on a threshold. The fixed period often misses the rapid change of the computing power state and has limited accuracy, while the threshold-based method often depends on the computing power collection source for control and has poor generality. According to the computing power network system architecture in the foregoing embodiments, in this embodiment, the distributed computing power perception is taken as an example to illustrate the computing power perception design method.

[0065] Exemplarily, add the state information of each computing instance in the EGW computing power perception component, including but not limited to information such as computing delay and available bandwidth.

[0066] Exemplarily, configure multiple gradient thresholds for the keyword field information affecting the computing power in the EGW computing power perception component. The selected fields are not limited to one. For example, the computing delay x1, x2... xM ms and the available bandwidth y1, y2... yN Mbps can be selected.

[0067] Exemplarily, when the computing power perception component senses that the latest computing delay compared with the last recorded delay or the latest available bandwidth compared with the last recorded available bandwidth exceeds the above gradient threshold, triggering the computing power notification and the joint computing of the computing network and the network. Taking the available bandwidth as an example, the last recorded value is a, and y1 <= a <= y2, and the current latest sensed value is b, and b < y1 or b > y2, then triggering the computing power notification and calculation.

[0068] Regardless of whether the notification and the joint computing of the computing network and the network are triggered according to the gradient threshold, the state information of the instance is refreshed to the current record each time.

[0069] Exemplarily, after the step of obtaining the corresponding computing power state information and network state information according to the network paths corresponding to each computing instance, the following steps are further included:

[0070] Calculating or updating the multiple next-hop information according to the computing power state information and / or network state information, where the multiple next-hop information includes each network path, each service instance identifier, and each next-hop sharing ratio, and each next-hop sharing ratio is used for tiling and splitting each next-hop.

[0071] For example, the computing power awareness component is responsible for collecting status information of computing power instances, including but not limited to service identifiers, IP addresses of computing instances, IP addresses of connected EGW gateways, instance processing latency, and available IO bandwidth of instances. This information is then extended to the computing network components within the head node or centralized computing network controller via protocols such as BGP.

[0072] For example, the network awareness component is used to collect status information of each link in the network, such as link latency and available bandwidth. This information is propagated to the network computing component in the head node via the IGP protocol, or transmitted to the network computing component in the network controller via methods such as BGP-LS / Telemetry.

[0073] Referring to Figure 6, which is a schematic diagram of the computing network system in this embodiment, the computing network computing component establishes a computing network status database based on data collected by the network sensing component and the computing power sensing component. To perform integrated computing network computing according to the computing network SLA requirements, it typically combines network link topology, TE link information, and computing power status information for overall calculation to obtain the optimal computing network path and computing instances. For simplicity, Figure 6 uses a network segment to pre-configure multiple SR-Policies based on the network TE-DB. The computing network computing objective is the selection of SR-Policies and the determination of computing instances. This can be achieved by fusing the measurement parameters and computing status of SR-Policies for a one-time calculation. Alternatively, SR-Policies can be pre-configured; both methods are essentially the same in terms of computing network computing. The difference lies in the fact that pre-configuring SR-Policies provides a limited range of differentiated capabilities, which is an exhaustive approach.

[0074] Referring to Table 1, in this embodiment, based on the collected network status data, the network computing component performs joint calculations according to the network SLA corresponding to the service identifier. When considering experience indicator requirements, it is usually prioritized to check whether the end-to-end network indicators of each combination meet the requirements. Combining Figure 6 and Table 1, when the end-to-end SLA latency does not exceed 4ms, all four entries in the table meet the SLA requirements. In one embodiment, the IO available bandwidth ratio of the calculation instance (e.g., 20%, 30%, 30%, 20%) in the entries that meet the SLA requirements can be used to determine the respective load balancing ratio. In addition, the load balancing ratio can also be determined based on the idle level of the CPU's unified load metric, or it can be determined based on a combination of network available bandwidth and IO available bandwidth. The above is just one example; it is only a difference in control plane perception indicators and calculation algorithms. From the perspective of network load balancing principles, they are essentially the same. Therefore, as long as the calculation of the streaming computing power forwarding table and load balancing ratio for multiple next hops is based on the combination of network and computing power status, it is within the scope of protection of this application. This is also the most significant difference from existing network-level load balancing calculation methods.

[0075] Table 1. Example of service-identifier-oriented computing network data

[0076] For example, the computing component generates a computing power forwarding table based on the calculation results and distributes it to the computing power routing forwarding plane. The computing power forwarding table supports public and private network computing power routing, and its key feature is the use of Virtual Forwarding Instance (VRF-ID) and anycast route prefixes as indexes, while also containing entries consisting of multiple next-hops. Each such next-hop sub-entry contains at least three key elements: SR-Policy, Service SID, and traffic sharing ratio. The forwarding plane uses this forwarding table to guide the forwarding of computing power service request packets. Referring to Table 2, this embodiment provides a specific example of a streaming computing power forwarding table for reference.

[0077] Table 2. Example of Stream Computing Power Forwarding Issued by Control Plane Management

[0078] Typically, chips have a limit on the number of next-hop resources. Therefore, constraints need to be implemented in the control plane to ensure that the next hop of the streaming computing power forwarding table does not exceed the maximum specification number N. When calculating the streaming computing power forwarding table, the control plane can choose to sort the network paths and computing instances from best to worst that meet the network SLA conditions, or randomly select the first N network paths and computing instances.

[0079] Referring to Table 3, in order to maximize the use of the existing IP ECMP forwarding mechanism, for the flow computing power forwarding table that adopts the multi-next-hop UCMP load balancing mechanism, the next hop is split into multiple hops according to the load balancing ratio. The splitting method is shown in Tables 2 and 3. The corresponding UCMP ratio is 2:3:3:2, that is, each of the original next hops is copied into 2, 3, 3, and 2 hops respectively, for a total of 10 next hops. This converts UCMP into ECMP mode, which is completely consistent with the existing ECMP forwarding mechanism and reduces the need for modification of the forwarding plane.

[0080] Table 3. Example of streaming computing power forwarding for computing power request messages

[0081] This embodiment, through the above-described scheme, specifically selects computing instances that provide the same service and conform to the Computing Network Service Level Agreement (CLSA); obtains corresponding computing power status information and network status information based on the network path corresponding to each computing instance; performs joint calculations based on the computing power status information and network status information to generate the streaming computing power forwarding table, wherein the streaming computing power forwarding table includes multiple next-hop information; dynamically adjusts the load sharing ratio according to the remaining computing power and network resources perceived by the control plane, generates a UCMP load-sharing streaming computing power forwarding table supporting multiple next hops, and distributes it to the forwarding plane, ensuring effective utilization and efficient allocation of resources.

[0082] Third Embodiment

[0083] Referring to Figure 7, which is a flowchart illustrating the computing power scheduling method according to the third embodiment, this embodiment is based on any of the above embodiments and proposes the third embodiment of this application. In this third embodiment, the content that is the same as or similar to any of the above embodiments can be referred to the above description and will not be repeated hereafter. Based on this, referring to Figure 7, step S10 includes steps S101-S104: Step S101: Identify the virtual routing forwarding identifier and packet 5-tuple corresponding to the computing power request packet, wherein the packet 5-tuple includes the packet destination IP address; Step S102: Perform entry matching on the streaming computing power forwarding table according to the virtual routing forwarding identifier and the packet destination IP address; Step S103: If a corresponding entry is matched, perform a hash operation based on the packet 5-tuple to determine the next-hop information corresponding to the computing power request packet; Step S104: Obtain the network path and service instance identifier corresponding to the computing power request packet based on the next-hop information corresponding to the computing power request packet.

[0084] For example, after the step of obtaining the network path and service instance identifier corresponding to the computing power request message based on the next-hop information corresponding to the computing power request message, the method further includes:

[0085] Based on the virtual route forwarding identifier, packet 5-tuple, network path, and service instance identifier, a flow affinity table is created. The network path and service instance identifier are used as flow affinity attributes to forward subsequent computing power request packets of the same class to the computing instance.

[0086] For example, the computing power scheduling method further includes the following steps: in response to receiving a computing power request message, determining the network path and service instance identifier corresponding to the computing power request message based on the flow affinity table corresponding to the computing power request message; and forwarding the computing power request message to the corresponding computing instance according to the network path and service instance identifier.

[0087] For example, the step of determining the network path and service instance identifier corresponding to the computing power request packet based on the flow affinity table corresponding to the computing power request packet includes: identifying the virtual route forwarding identifier and packet 5-tuple corresponding to the computing power request packet; determining whether a matching entry in the corresponding flow affinity table is found based on the virtual route forwarding identifier and packet 5-tuple; if a matching entry in the corresponding flow affinity table is found, obtaining the network path and service instance identifier based on the matching entry in the flow affinity table; if no matching entry in the corresponding flow affinity table is found, determining the network path and service instance identifier corresponding to the computing power request packet by querying the flow computing power forwarding table.

[0088] Referring to Figure 8, which is a schematic diagram of the forwarding process in this embodiment of the application, the forwarding process of the computing power service request message in this embodiment mainly includes at least one of the following: Step 1: The ingress IGW computing power gateway receives the user's computing power service request; Step 2: The IGW identifies the corresponding VRF-ID based on the interface that received the computing power service request message. When the VRF-ID is not equal to zero, it indicates private network forwarding of the computing power route; when the VRF-ID is equal to zero, it indicates public network forwarding of the computing power route; Step 3: Query the flow computing power forwarding table based on the VRF-ID and the destination IP of the message, and perform the longest match of the computing power forwarding table by combining the VRF-ID and the destination IP address of the message. If a match is found, proceed to Step 4 to obtain the flow affinity table; otherwise, discard the message; Step 4: Query the flow affinity table based on the VRF-ID and the message 5-tuple. If a match is found, obtain the corresponding SR-Policy and Service. If the SID is entered, proceed to step 6 to encapsulate and forward the packet; otherwise, proceed to step 6 to query the computing power forwarding table. Step 5: Perform a hash operation based on the packet's five-tuple information, determine the appropriate next-hop entry in the streaming computing power forwarding table, obtain the corresponding SR-Policy and Service SID, and create a streaming affinity table. This table contains key information such as VRF-ID, 5-tuple, SR-Policy, Service SID, and aging time, ensuring that subsequent packets of the same type are routed to the same compute instance for processing. Step 6: The IGW gateway encapsulates an SRH based on the SR-Policy and Service SID obtained in steps 4 and 5, and forwards the IPv6 compute request packet. The underlay network device performs IP forwarding based on the IPv6 and SRH packet headers. Step 7: The IGW gateway removes the SRH encapsulation and performs forwarding based on the Service SID. When the Service SID is END.DX, the packet is forwarded to the corresponding tunnel based on END.DX. When the Service SID is END.DT (there is only one compute instance), the packet is forwarded by looking up the destination IP address in the corresponding VRF based on the VRF-ID indicated by END.DT. Step 8: The compute instance receives the compute request packet, performs subsequent processing, and responds with the corresponding compute service request response packet.

[0089] For example, when different users make computing power requests, the process of steps 1-8 is repeated. The difference is that in step 5, the next hop obtained by hashing is different according to the different results of the five-tuple HASH, that is, different network paths (SR-Policy) and computing instances (Service SID) are hit. This achieves the effect of session-level load balancing for different users before the next update of the streaming computing power forwarding table after the scheduling result is issued.

[0090] In this embodiment, the control plane uses gradient thresholds to trigger computations to reduce the computational pressure on the computing power announcement and streaming computing power forwarding table. At the same time, the forwarding plane uses the aforementioned session-level load balancing mechanism, so that after a table is created, each computing power session accesses multiple different computing instances and network paths. This further reduces the probability that the status indicators of each computing instance exceed the gradient thresholds, thereby avoiding the burden on the control plane caused by frequent computing power announcements, computing network calculations, and table creation, forming a low-frequency negative feedback system oriented towards computing network awareness, routing calculations, and service forwarding.

[0091] This embodiment, through the above scheme, specifically identifies the virtual routing forwarding identifier and packet 5-tuple corresponding to the computing power request packet. Based on the virtual routing forwarding identifier and the destination IP address of the packet, it performs entry matching on the streaming computing power forwarding table, and then performs a hash operation on the packet 5-tuple to determine the next-hop information corresponding to the computing power request packet. Based on the next-hop information corresponding to the computing power request packet, it obtains the network path and service instance identifier corresponding to the computing power request packet; or it obtains the network path and service instance identifier by querying the flow affinity table, so as to forward the computing power request packet to the corresponding computing instance. By using a session-level load balancing mechanism, within the cycle of updating the streaming computing power forwarding table, it can dynamically guide the traffic of different sessions to different paths and computing instances, effectively avoiding excessive concentration of resources, reducing the pressure on the control plane perceived resources and the burden of managing down-table entries.

[0092] Fourth embodiment

[0093] Referring to Figure 9, which is a flowchart illustrating the computing power scheduling method according to the fourth embodiment, this embodiment is based on any of the above embodiments and proposes the fourth embodiment of this application. In this fourth embodiment, content that is the same as or similar to any of the above embodiments can be referred to the above description and will not be repeated hereafter. Based on this, referring to Figure 9, step S20 includes steps S201-S203: Step S201: Encapsulate a segment routing header through the ingress gateway according to the network path and service instance identifier; Step S202: Forward the computing power request message to the egress gateway according to the segment routing header; Step S203: Decapsulate the segment routing header through the egress gateway and forward the computing power request message to the corresponding computing instance according to the service instance identifier.

[0094] For example, the step of forwarding the computing power request message to the corresponding computing instance based on the service instance identifier includes: identifying the service instance identifier; if the service instance identifier is a first type segment identifier, forwarding the computing power request message to the interface corresponding to the service instance identifier to deliver it to the corresponding computing instance; if the service instance identifier is a second type segment identifier, performing a table lookup in the corresponding logical network based on the logical network identifier indicated by the second type segment identifier and the destination IP address of the message corresponding to the computing power request message, and forwarding the computing power request message to the corresponding computing instance based on the table lookup result.

[0095] Referring to Figure 10, which is a schematic diagram of session-level load balancing packet encapsulation in an embodiment of this application, as shown in Figure 10, the uplink stages of session-level computing network load balancing mainly include from the terminal to the ingress computing gateway, from the ingress computing gateway to the egress computing gateway, and from the egress computing gateway to the computing instance.

[0096] For example, in the terminal to the ingress computing power gateway stage, in order to reduce terminal modification, the DA is used to directly encapsulate the anycast destination address that represents the computing power table service identifier. Based on this destination address, the computing power service message is sent to the ingress gateway through access network routing.

[0097] For example, in the stage from the ingress computing power gateway to the egress computing power gateway, the SR policy and Service ID information of each next hop are obtained through packet DA computing power routing addressing. An IPv6 header is added and the SR policy and Service ID (END.DX in the figure) are encapsulated into the corresponding SRH. IPv6 hop-by-hop forwarding is achieved through the outer IPv6 header and SRH.

[0098] For example, in the stage from the outbound computing gateway to the compute instance, upon encountering the Service SID (END.DX), the SRH encapsulation of the packet is removed. Based on the IP tunnel interface or general interface mapped to the Service SID (END.DX), the computing request packet is directly forwarded from the corresponding tunnel interface. A tunnel header is encapsulated according to the IP tunnel type. The destination IP in the tunnel header is the compute instance IP, and the source IP is the Egress gateway IP, corresponding to the remote IP and local IP configured in the gateway tunnel, respectively. Here, the tunnel can be of any type, such as GRE, GIF, etc. When the outbound computing gateway connects to only one compute instance providing the same service, the Service SID can be END.DT, and the computing request packet (DA) can be directly forwarded without encapsulating an external tunnel.

[0099] To maintain consistency with the uplink, the downlink packet forwarding still uses tunnel forwarding encapsulation directly from the compute instance to the egress gateway. Encapsulation at other stages is a standard VPN encapsulation and forwarding process that is not based on compute power routing addressing, which will not be elaborated in this embodiment.

[0100] In this embodiment, the session-level load balancing differs from the non-session-level load balancing in its forwarding lookup mechanism and load-sharing mechanism, but the packet encapsulation at each stage is completely consistent. The main difference in the forwarding behavior of session-level load balancing lies in the ingress computing power gateway. Therefore, the session-level load balancing function can support smooth upgrades based on computing power routing and can be selectively enabled.

[0101] This embodiment, through the above scheme, specifically encapsulates a segment routing header based on the network path and service instance identifier at the ingress gateway; forwards the computing power request message to the egress gateway based on the segment routing header; and decapsulates the segment routing header at the egress gateway and forwards the computing power request message to the corresponding computing instance based on the service instance identifier. This ensures that the computing power request message can be correctly routed to the appropriate computing instance, providing a complete computing power scheduling mechanism to achieve more balanced and efficient utilization of computing network resources.

[0102] Fifth embodiment

[0103] Referring to Figure 11, which is a schematic diagram of the computing power scheduling device provided in an embodiment of this application, the device can be mounted on or is the terminal device in the above-described method embodiments. The computing power scheduling device shown in Figure 11 can be used to perform some or all of the functions in the method embodiments described above. As shown in Figure 11, the computing power scheduling device includes the following modules: a response module, used to determine the network path and service instance identifier corresponding to the computing power request message based on the streaming computing power forwarding table corresponding to the computing power request message in response to receiving the computing power request message; and a forwarding module, used to forward the computing power request message to the corresponding computing instance according to the network path and service instance identifier.

[0104] For example, before the step of determining the network path and service instance identifier corresponding to the computing power request message based on the streaming computing power forwarding table corresponding to the computing power request message, the method further includes: selecting each computing instance that provides the same service and conforms to the computing network service level agreement; obtaining the corresponding computing power status information and network status information according to the network path corresponding to each computing instance; and performing joint calculation based on the computing power status information and network status information to generate the streaming computing power forwarding table, wherein the streaming computing power forwarding table includes multiple next-hop information.

[0105] For example, the step of determining the network path and service instance identifier corresponding to the computing power request packet based on the streaming computing power forwarding table corresponding to the computing power request packet includes: identifying the virtual route forwarding identifier and packet 5-tuple corresponding to the computing power request packet, wherein the packet 5-tuple includes the destination IP address of the packet; performing entry matching on the streaming computing power forwarding table according to the virtual route forwarding identifier and the destination IP address of the packet; if a corresponding entry is matched, performing a hash operation on the packet 5-tuple to determine the next-hop information corresponding to the computing power request packet; and obtaining the network path and service instance identifier corresponding to the computing power request packet according to the next-hop information corresponding to the computing power request packet.

[0106] For example, after the step of obtaining the network path and service instance identifier corresponding to the computing power request message based on the next-hop information corresponding to the computing power request message, the method further includes: creating a flow affinity table based on the virtual route forwarding identifier, the message quintuple, the network path and the service instance identifier, for forwarding subsequent computing power request messages of the same flow to the computing instance.

[0107] For example, the apparatus is further configured to perform the following steps: in response to receiving a computing power request message, determining the network path and service instance identifier corresponding to the computing power request message based on the flow affinity table corresponding to the computing power request message; and forwarding the computing power request message to the corresponding computing instance according to the network path and service instance identifier.

[0108] For example, the step of determining the network path and service instance identifier corresponding to the computing power request packet based on the flow affinity table corresponding to the computing power request packet includes: identifying the virtual route forwarding identifier and packet 5-tuple corresponding to the computing power request packet; determining whether a matching entry in the corresponding flow affinity table is found based on the virtual route forwarding identifier and packet 5-tuple; if a matching entry in the corresponding flow affinity table is found, obtaining the network path and service instance identifier based on the matching entry in the flow affinity table; if no matching entry in the corresponding flow affinity table is found, determining the network path and service instance identifier corresponding to the computing power request packet by querying the flow computing power forwarding table.

[0109] For example, the step of obtaining the corresponding computing power status information and network status information based on the network path corresponding to each computing instance includes: collecting the computing power status information of each computing instance through the computing power sensing component, and collecting the network status information of each network path through the network sensing component.

[0110] For example, after the step of collecting the computing power status information of each computing instance through the computing power sensing component, the method further includes: recording the computing power status information of each computing instance, wherein the computing power status information includes at least one computing power status index; and triggering computing power status diffusion and calculating and updating the streaming computing power forwarding table when the computing power sensing component detects that the change of the at least one computing power status index exceeds the corresponding preset gradient threshold.

[0111] For example, after the step of obtaining the corresponding computing power status information and network status information based on the network path corresponding to each computing instance, the method further includes: calculating or updating the multiple next-hop information based on the computing power status information and / or network status information, wherein the multiple next-hop information includes each network path, each service instance identifier and each next-hop sharing ratio, and the each next-hop sharing ratio is used to flatten and split each next hop.

[0112] For example, the step of forwarding the computing power request message to the corresponding computing instance based on the network path and service instance identifier includes: encapsulating a segment routing header through the ingress gateway based on the network path and service instance identifier; forwarding the computing power request message to the egress gateway based on the segment routing header; and decapsulating the segment routing header through the egress gateway and forwarding the computing power request message to the corresponding computing instance based on the service instance identifier.

[0113] For example, the step of forwarding the computing power request message to the corresponding computing instance based on the service instance identifier includes: identifying the service instance identifier; if the service instance identifier is a first type segment identifier, forwarding the computing power request message to the interface corresponding to the service instance identifier to deliver it to the corresponding computing instance; if the service instance identifier is a second type segment identifier, performing a table lookup in the corresponding logical network based on the logical network identifier indicated by the second type segment identifier and the destination IP address of the message corresponding to the computing power request message, and forwarding the computing power request message to the corresponding computing instance based on the table lookup result.

[0114] In existing computing network technologies, the control plane, based on the target constraints of the computing network SLA (Quality of Service Agreement), uses the perceived computing power METRIC information and network TE link information to calculate the optimal computing instance and network path that satisfies the computing network SLA in this scheduling. In the embodiments of this application, the control plane performs a single scheduling calculation and orchestration from the ingress gateway to all service instances that can provide the same service and meet the computing network SLA requirements, as well as their reachable paths and load sharing ratios.

[0115] For example, the control plane determines the traffic allocation ratio for each instance and path by evaluating the available resources (such as remaining bandwidth, processing load, storage space, etc.) and network resources (such as remaining interface bandwidth, available bandwidth of SRv6-POLICY, etc.) of the compute instance based on the network path corresponding to each compute instance.

[0116] For example, the forwarding plane is based on a multi-next-hop streaming computing power forwarding table. The key indexes include VRFID and service identifier prefixes. The next-hop information includes SRv6-POLICY and Service SID used to direct forwarding request packets to computing instances, as well as traffic sharing ratio information. Service request packets are forwarded according to the load sharing ratio, while maintaining the affinity of session flows, thereby achieving traffic load balancing of computing power network service requests at the session level.

[0117] For example, the control plane introduces a state recording and gradient threshold management model for computing instances in the computing power awareness component. That is, it records the previous computing power instance state value of each computing instance. By comparing the latest value of the current awareness with the previous value, if the state of the computing power instance crosses the threshold range, it triggers the diffusion of computing power state and the calculation of the control plane. Combined with the forwarding plane, it realizes the ability to achieve multiple next-hop session-level load balancing capabilities, which reduces the probability of each computing instance state exceeding the threshold. Overall, it greatly reduces the frequency of computing power awareness, notification information transmission, and calculation frequency of the control plane.

[0118] For example, in this embodiment of the application, in order to reduce the computational pressure of computing power announcements and streaming computing power forwarding tables, the control plane uses gradient thresholds to trigger the process. At the same time, the forwarding plane uses the aforementioned session-level load balancing mechanism, so that after a table is created, each computing power session accesses multiple different computing instances and network paths. This further reduces the probability that the status indicators of each computing instance exceed each gradient threshold, thereby avoiding the burden on the control plane caused by frequent computing power announcements, computing network calculations, and table creation, forming a low-frequency negative feedback system oriented towards computing network awareness, routing calculations, and service forwarding.

[0119] For example, although the network load balancing calculations and entries involved in the embodiments of this application are somewhat similar to the concepts of ECMP / UCMP in traditional network domains, the latter focuses on reaching the same destination network segment to achieve balanced utilization of network resources, while the former focuses on reaching the same type of service to achieve balanced utilization of network resources and computing resources.

[0120] This application differs significantly from related technologies in its technical implementation, specifically in three aspects: 1) Regarding the calculation method for multiple next-hops in the streaming computing power forwarding table, this application comprehensively considers the dynamic status of computing power and multiple network indicators (such as available bandwidth, latency, packet loss, etc.), while network-level ECMP / UCMP mainly considers static indicators such as port bandwidth and COST value; 2) In terms of table design, this application combines network paths and mapped computing instances to form the next hop, while network-level ECMP / UCMP often only uses adjacent connection IPs or SR-Policies as multiple next hops; 3) In the forwarding layer computing power network application, directly applying network ECMP / UCMP can only solve the network-level load balancing problem of reaching the same computing instance, but cannot solve the load balancing problem of connecting multiple computing instances under the same gateway, while this application can solve this problem.

[0121] In this embodiment, by implementing session-level load balancing on the computing network, each computation can generate multiple next hops that meet the computing network SLA requirements. These next hops simultaneously define the network path and the computing instance. This design allows coexistence with service-identifier-oriented computing network best selection mechanisms, adopting differentiated strategies based on different service identifiers, thereby avoiding the limitations of relying solely on a single computing network best selection. By directly mapping the corresponding computing instance using a specific SERVICE SID, it is not necessary to advertise the computing instance IP address to all computing nodes in the computing network, ensuring compatibility with existing EVPN / L3VPN MP-BGP protocol processes. The computation result adds the SERVICE SID to the multi-next-hop attribute field of the streaming computing power forwarding table, combining it with the SR-Policy to form a unique entry, participating in UCMP load balancing. The UCMP load balancing mechanism is finally converted to the ECMP mechanism at the forwarding plane, which greatly reduces changes at the forwarding layer and ensures compatibility with existing IP ECMP mechanisms. Based on the remaining computing power and network resources perceived by the control plane, the load balancing ratio is dynamically adjusted to generate a UCMP load balancing streaming computing power forwarding table supporting multiple next hops, which is then distributed to the forwarding plane. This approach ensures the effective utilization and efficient allocation of resources. The control plane supports gradient thresholds on the computing power awareness component. When the computing power awareness state differs from the previous state by a cross-domain gradient value, computing power announcements and computing power routing calculations are performed, directly reducing the pressure on the control plane. Combined with session-level load balancing on the forwarding plane for multiple computing instances and network paths, the magnitude and probability of state changes for each computing instance can be directly reduced, thereby reducing the pressure on the control plane's awareness state changes from the source and reducing the pressure on each stage of control plane awareness, announcement, and calculation.

[0122] The computing power scheduling device provided in this application embodiment is similar in implementation principle and beneficial effect to the technical solution shown in the corresponding method embodiment above, and will not be described again here.

[0123] It should be noted that the streaming forwarding table in this application has the following two deployment methods: 1) Deployment at the forwarding plane: The first packet directly matches the target IP through the forwarding plane, obtains the next-hop information based on the hash result of the 5-tuple, and this information is directly used to guide service forwarding and trigger the control plane to create a flow affinity table and distribute it to the forwarding plane. Subsequent packets are forwarded according to the flow affinity table; 2) Deployment at the control plane: The first packet is sent to the control plane, where the target IP is found and the next-hop information is obtained based on the hash result of the 5-tuple. Then, it is encapsulated and forwarded, and a flow affinity table is generated and sent to the forwarding plane to directly guide the forwarding of subsequent packets. Both deployment schemes are specific embodiments of this application and are within the protection scope of this application.

[0124] Furthermore, this application embodiment also provides a network device, the network device including: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the computing power scheduling method as described above.

[0125] Furthermore, this application embodiment also provides a storage medium, which is a computer-readable storage medium, and stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of the computing power scheduling method described above.

[0126] In this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0127] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0128] The above are merely some embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A computing power scheduling method, wherein, The computing power scheduling method includes: In response to receiving a computing power request message, the network path and service instance identifier corresponding to the computing power request message are determined based on the streaming computing power forwarding table corresponding to the computing power request message; The computing power request message is forwarded to the corresponding computing instance based on the network path and service instance identifier.

2. The computing power scheduling method as described in claim 1, wherein, Before the step of determining the network path and service instance identifier corresponding to the computing power request message based on the stream computing power forwarding table corresponding to the computing power request message, the method further includes: Select computing instances that provide the same service and comply with the computing network service level agreement; Obtain the corresponding computing power status information and network status information based on the network path corresponding to each computing instance; The streaming computing power forwarding table is generated by jointly calculating the computing power status information and the network status information, wherein the streaming computing power forwarding table includes multiple next-hop information.

3. The computing power scheduling method as described in claim 2, wherein, The step of determining the network path and service instance identifier corresponding to the computing power request message based on the stream computing power forwarding table corresponding to the computing power request message includes: Identify the virtual route forwarding identifier and the packet 5-tuple corresponding to the computing power request packet, wherein the packet 5-tuple includes the packet destination IP address; The stream computing power forwarding table is matched with entries based on the virtual route forwarding identifier and the destination IP address of the packet; If a matching entry is found, a hash operation is performed based on the message quintuple to determine the next-hop information corresponding to the computing power request message; Based on the next-hop information corresponding to the computing power request message, obtain the network path and service instance identifier corresponding to the computing power request message.

4. The computing power scheduling method as described in claim 3, wherein, After the step of obtaining the network path and service instance identifier corresponding to the computing power request message based on the next-hop information corresponding to the computing power request message, the method further includes: Based on the virtual route forwarding identifier, packet 5-tuple, network path, and service instance identifier, a flow affinity table is created to forward subsequent computing power request packets of the same class to the computing instance.

5. The computing power scheduling method as described in claim 4, wherein, The method further includes: In response to receiving a computing power request message, the network path and service instance identifier corresponding to the computing power request message are determined based on the flow affinity table corresponding to the computing power request message; The computing power request message is forwarded to the corresponding computing instance based on the network path and service instance identifier.

6. The computing power scheduling method as described in claim 5, wherein, The step of determining the network path and service instance identifier corresponding to the computing power request message based on the flow affinity table corresponding to the computing power request message includes: Identify the virtual route forwarding identifier and the packet 5-tuple corresponding to the computing power request message; Based on the virtual route forwarding identifier and the packet 5-tuple, determine whether a matching entry for the corresponding flow affinity table is found; If a matching entry in the corresponding flow affinity table is found, the network path and service instance identifier are obtained based on the matching entry in the flow affinity table. If no matching entry is found in the flow affinity table, the network path and service instance identifier corresponding to the computing power request message are determined by querying the flow computing power forwarding table.

7. The computing power scheduling method as described in claim 2, wherein, The step of obtaining the corresponding computing power status information and network status information based on the network path corresponding to each computing instance includes: The computing power status information of each computing instance is collected by the computing power awareness component, and the network status information of each network path is collected by the network awareness component.

8. The computing power scheduling method as described in claim 7, wherein, Following the step of collecting the computing power status information of each computing instance through the computing power sensing component, the method further includes: Record the computing power status information of each computing instance, wherein the computing power status information includes at least one computing power status index; When the computing power sensing component detects that the change of at least one computing power status indicator exceeds the corresponding preset gradient threshold, it triggers computing power status diffusion and calculates and updates the streaming computing power forwarding table.

9. The computing power scheduling method as described in claim 8, wherein, Following the step of obtaining the corresponding computing power status information and network status information based on the network path corresponding to each computing instance, the method further includes: The multiple next-hop information is calculated or updated based on the computing power status information and / or network status information. The multiple next-hop information includes each network path, each service instance identifier, and each next-hop load sharing ratio. The each next-hop load sharing ratio is used to flatten and split each next hop.

10. The computing power scheduling method as described in claim 1 or 5, wherein, The step of forwarding the computing power request message to the corresponding computing instance based on the network path and service instance identifier includes: The ingress gateway encapsulates the segment routing header based on the network path and service instance identifier; The computing power request message is forwarded to the egress gateway according to the segment routing header; The segment routing header encapsulation is removed through the egress gateway, and the computing power request message is forwarded to the corresponding computing instance according to the service instance identifier.

11. The computing power scheduling method as described in claim 10, wherein, The step of forwarding the computing power request message to the corresponding computing instance based on the service instance identifier includes: Identify the service instance identifier; If the service instance identifier is a first type segment identifier, the computing power request message is forwarded to the interface corresponding to the service instance identifier and delivered to the corresponding computing instance. When the service instance is identified as a second type segment identifier, a table is looked up in the corresponding logical network according to the logical network identifier indicated by the second type segment identifier, based on the destination IP address of the computing power request message. Based on the table lookup result, the computing power request message is forwarded to the corresponding computing instance.

12. A network device, wherein, The network device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the computing power scheduling method as described in any one of claims 1 to 11.

13. A storage medium, wherein, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the computing power scheduling method as described in any one of claims 1 to 11.

Images