A distributed cloud resource dynamic scheduling method based on a 5G network

Abstract

The application discloses a kind of distributed cloud resource dynamic scheduling methods based on 5G network, it is related to 5G communication and cloud computing resource scheduling field.The method obtains the resource scheduling request of 5G terminal and the resource snapshot of distributed cloud node, according to adjacent resource snapshot, determine resource consumption rate, and combine the time distance correction available resource amount of snapshot from current time, candidate cloud node is screened;Candidate cloud node is carried out based on the condition reservation of resource state version, and obtains cloud resource reservation window;5G network path state is obtained, and 5G network guarantee window and the network cloud overlap window of cloud resource reservation window are determined;According to the required guarantee time length of task, select target cloud node, submit the condition reservation of target node and release remaining reservation resource.The application can reduce the influence of resource snapshot lag and concurrent occupation on scheduling, improve the matching degree of cloud resource and 5G network guarantee time.

Description

Technical Field

[0001] This invention relates to the field of 5G communication and cloud computing resource scheduling, and in particular to a method for dynamic scheduling of distributed cloud resources based on 5G networks. Background Technology

[0002] With the development of 5G communication technology, edge computing, and cloud computing technology, an increasing number of computing tasks generated by 5G terminals are being offloaded to distributed cloud nodes for execution. Distributed clouds typically consist of edge cloud nodes, regional cloud nodes, and central cloud nodes deployed in different locations, each with differences in computing power, storage capacity, network access conditions, and task load. To meet the service requirements of low latency, high bandwidth, and high reliability, it is necessary to select suitable nodes from multiple distributed cloud nodes to execute tasks based on the resource requirements of the terminal tasks and the network service quality requirements.

[0003] Existing distributed cloud resource scheduling methods typically generate resource snapshots based on the resource status (processors, memory, storage, etc.) reported by cloud nodes, and then filter cloud nodes according to task requirements. However, there is usually a time interval between the collection and transmission of resource snapshots and their use by the scheduling system. During this time interval, cloud nodes may experience new task access, resource release, or other reservation operations, causing a discrepancy between the available resource amount recorded in the resource snapshot and the actual available resource amount at the time of scheduling. Especially when tasks arrive in a concentrated manner or node load changes rapidly, it is easy to encounter the problem that the cloud nodes selected based on resource snapshots may no longer be able to provide the corresponding resources when the task is submitted.

[0004] Some scheduling methods mitigate the impact of resource status changes by re-querying node resource status or increasing resource redundancy. However, frequent queries increase communication overhead between cloud nodes and the scheduling system, and concurrent resource changes can still occur between resource query and task submission. Meanwhile, existing methods typically determine cloud resource availability and 5G network path compliance separately, primarily relying on instantaneous metrics such as resource availability, transmission rate, or latency at the time of scheduling for comprehensive scoring, with less consideration given to the correlation between cloud resource availability time and 5G network availability time.

[0005] In the actual task unloading process, even if a cloud node has computing resources that meet the task requirements, if the availability time of the cloud node's resources does not coincide with the time when the 5G network path can meet the quality of service requirements, there may still be situations where task input data cannot be transmitted on time, network service quality degrades during task execution, or reserved cloud resources cannot be effectively utilized.

[0006] To address this, a dynamic scheduling method for distributed cloud resources based on 5G networks is proposed. Summary of the Invention

[0007] The main objective of this invention is to provide a dynamic scheduling method for distributed cloud resources based on 5G networks, which can effectively solve the problems in the background technology.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A method for dynamic scheduling of distributed cloud resources based on 5G networks includes the following steps:

[0010] S1, receive a resource scheduling request sent by a 5G terminal and obtain resource snapshots of multiple distributed cloud nodes. The resource scheduling request includes task parameters, cloud resource requirements and network service quality requirements. The resource snapshots include the available resource quantity, collection time and resource status version identifier of various cloud resources.

[0011] S2, based on the ratio of the decrease in available resources between adjacent resource snapshots of each distributed cloud node within a preset observation period to the collection time interval, determine the resource consumption rate of various cloud resources, and based on the resource consumption rate and the time distance between the latest resource snapshot and the current time, correct the available resources in the latest resource snapshot to obtain the reliable available resources, and determine the distributed cloud node whose reliable available resources meet the cloud resource requirements as a candidate cloud node;

[0012] S3, send a conditional reservation request carrying the corresponding resource status version identifier to the candidate cloud node, and obtain the conditional reservation result returned by the candidate cloud node when the resource status version verification is passed and the current available resource quantity meets the cloud resource requirements. The conditional reservation result includes a conditional reservation identifier, the locked cloud resource quantity and the cloud resource reservation window. The cloud resource reservation window starts at the time when the locked cloud resource can be used and ends at the time when the conditional reservation expires.

[0013] S4, obtain the network path status from the 5G terminal to the candidate cloud node that has been successfully booked under each condition, returned by the 5G network management side, determine the time interval during which the network path status continuously meets the network service quality requirements as the 5G network guarantee window, and determine the time interval during which the 5G network guarantee window and the corresponding cloud resource booking window overlap as the network-cloud overlap window.

[0014] S5. Based on the task parameters, the locked cloud resource quantities, and the network path status, determine the required guarantee duration for each candidate cloud node's task. Select a target cloud node from the candidate cloud nodes whose network-cloud overlap window duration is not less than the required guarantee duration for the corresponding task. Send a reservation submission instruction carrying the corresponding conditional reservation identifier to the target cloud node and release the conditional reservation resources of the candidate cloud nodes other than the target cloud node.

[0015] Furthermore, the task parameters include at least the amount of task input data, the amount of task output data, and the amount of task computation; the cloud resource requirements include at least the amount of computing resources, the amount of memory, and the amount of storage resources; and the network service quality requirements include at least the minimum uplink transmission rate, the minimum downlink transmission rate, and the maximum end-to-end latency.

[0016] Furthermore, within the preset observation period, the number of resource snapshots arranged chronologically by acquisition time is: ,and For the first The first distributed cloud node For cloud-like resources, the resource consumption rate is determined according to the following formula:

[0017]

[0018] The amount of trusted available resources is determined according to the following formula:

[0019]

[0020] in, Indicates the first The first distributed cloud node In the first resource snapshot The amount of available resources in a snapshot of cloud-like resources. Indicates the first The time of collection of a resource snapshot, Indicates the first The resource consumption rate of cloud-like resources Indicates the current moment. Indicates the first The preset resource security reserve for cloud-like resources. Indicates the first The amount of reliable and available cloud-like resources; when each type of cloud resource on a distributed cloud node satisfies At that time, the distributed cloud node is identified as a candidate cloud node. Indicating the cloud resource requirement as the first Demand for cloud-like resources.

[0021] Furthermore, the resource status version identifier is updated when cloud resource allocation, cloud resource release, or conditional reservation occurs on the corresponding distributed cloud node; the conditional reservation request also includes the expected task start time and the conditional reservation validity period; the candidate cloud node compares the resource status version identifier carried in the conditional reservation request with the current resource status version identifier. If they match and the current available resource quantity meets the cloud resource requirement, it locks no less than the required cloud resource quantity, generates the conditional reservation identifier, and determines the later of the expected task start time and the earliest available time of the locked cloud resource as the start time of the cloud resource reservation window, and determines the sum of the conditional reservation establishment time and the conditional reservation validity period as the end time of the cloud resource reservation window; if the resource status version identifier is inconsistent or the current available resource quantity does not meet the cloud resource requirement, it returns an updated resource snapshot containing the current resource status version identifier.

[0022] Furthermore, the network path state includes multiple consecutive time slices, each corresponding to a guaranteed uplink transmission rate, a guaranteed downlink transmission rate, and an end-to-end latency. Adjacent time slices that guarantee an uplink transmission rate no lower than the minimum uplink transmission rate, a downlink transmission rate no lower than the minimum downlink transmission rate, and an end-to-end latency no greater than the maximum end-to-end latency are merged to obtain the 5G network guarantee window.

[0023] Furthermore, regarding the first Each candidate cloud node is recorded as its cloud resource reservation window. Record any corresponding 5G network guarantee window as The start and end times of the corresponding cloud overlap window are determined according to the following formula:

[0024]

[0025]

[0026] when At that time, A valid cloud overlap window is defined as one where none of the overlap time intervals corresponding to a candidate cloud node meet the condition. If the candidate cloud node does not have a valid network-cloud overlap window, the conditional reservation resources of the candidate cloud node are released.

[0027] Furthermore, regarding the first The candidate cloud node corresponding to the first For each valid overlapping window in the cloud network, the required duration of the task's protection is determined according to the following formula:

[0028]

[0029] in, Indicates the amount of input data for the task. Indicates the amount of data output by the task. Indicates the computational load of the task. and These represent the guaranteed uplink and downlink transmission rates within the corresponding 5G network guarantee window, respectively. Indicates according to the first The available computing rate is determined by the amount of locked computing resources returned by each candidate cloud node; according to Determine the scheduling margin of the corresponding network-cloud overlap window, and from the candidate cloud nodes with scheduling margins greater than or equal to zero, determine the candidate cloud node with the largest scheduling margin and its corresponding network-cloud overlap window as the target cloud node and target execution window, respectively.

[0030] Furthermore, before sending the reservation submission instruction to the target cloud node, the conditional reservation status corresponding to the target cloud node and the latest network path status returned by the 5G network management side are obtained. Based on the conditional reservation status and the latest network path status, the network-cloud overlap window and its scheduling margin are re-determined. When the conditional reservation status is invalid, or the re-determined scheduling margin is less than the preset safety margin, the conditional reservation resources that are still locked are released, an updated resource snapshot is obtained, and steps S2 to S5 are re-executed. When the conditional reservation status is valid and the re-determined scheduling margin is not less than the preset safety margin, the reservation submission instruction is sent, wherein the preset safety margin is a preset duration threshold that is greater than or equal to zero.

[0031] A distributed cloud resource dynamic scheduling system based on a 5G network includes:

[0032] The request and snapshot acquisition module is used to receive resource scheduling requests sent by 5G terminals and acquire resource snapshots of multiple distributed cloud nodes. The resource scheduling request includes task parameters, cloud resource requirements and network service quality requirements. The resource snapshot includes the available resource quantity, collection time and resource status version identifier of various cloud resources.

[0033] The trusted resource determination module is used to determine the resource consumption rate of various cloud resources based on the ratio of the decrease in available resources between adjacent resource snapshots to the collection time interval, correct the available resources of the snapshot based on the resource consumption rate and the time distance between the latest resource snapshot and the current time to obtain the trusted available resources, and filter candidate cloud nodes based on the trusted available resources.

[0034] The conditional reservation module is used to send a conditional reservation request carrying the corresponding resource status version identifier to the candidate cloud node, and obtain the conditional reservation identifier, the locked cloud resource quantity, and the cloud resource reservation window returned by the candidate cloud node when the resource status version verification is passed and the current available resource quantity meets the cloud resource requirements. The cloud resource reservation window starts at the time when the locked cloud resource can be used and ends at the time when the conditional reservation expires.

[0035] The window matching module is used to obtain the network path status from the 5G terminal to the candidate cloud node that has been successfully booked under each condition, returned by the 5G network management side, determine the 5G network guarantee window based on the time interval in the network path status that continuously meets the network service quality requirements, and determine the overlapping time interval between the 5G network guarantee window and the corresponding cloud resource booking window as the network-cloud overlapping window.

[0036] The scheduling and execution module is used to determine the required guarantee duration for the task based on the task parameters, the locked cloud resource quantities, and the network path status; determine the target cloud node from the candidate cloud nodes whose duration of network-cloud overlap window is not less than the required guarantee duration for the corresponding task; send a reservation submission instruction carrying the corresponding conditional reservation identifier to the target cloud node; and release the conditional reservation resources of the candidate cloud nodes other than the target cloud node.

[0037] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method steps.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. This invention determines the resource consumption rate based on the changes in available resources in adjacent resource snapshots, and corrects the available resource quantity of the snapshot by combining the time distance between the latest resource snapshot and the current time. This enables the selection criteria for candidate cloud nodes to reflect the resource consumption situation after the resource snapshot is collected, reducing the probability of selecting a node with insufficient resources due to the lag in resource snapshots.

[0040] 2. This invention verifies candidate cloud nodes by carrying a conditional reservation request with a resource status version identifier, and locks the corresponding cloud resources and forms a cloud resource reservation window after successful verification. This can reduce reservation failures caused by concurrent resource changes between resource queries and task submissions, and at the same time avoid multiple tasks repeatedly occupying the same resources based on the same expired resource snapshot.

[0041] 3. This invention determines the cloud resource reservation window and the 5G network guarantee window respectively, and forms a network-cloud overlap window by the overlapping time interval of the two, so that task scheduling no longer depends only on the cloud resource status or network status at a certain moment, but can determine whether the cloud resources and 5G network service quality are effective at the same time during task execution.

[0042] 4. This invention determines the required guarantee time for a task based on the data transmission duration and computation execution duration of the task, and selects the target cloud node by combining the scheduling margin of the network-cloud overlap window. Before the reservation is submitted, it can also be reviewed based on the latest status, thereby improving the task scheduling success rate and releasing reservation resources that have not been adopted or no longer meet the scheduling conditions in a timely manner. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0044] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0046] Example 1

[0047] like Figure 1-2 As shown, this embodiment provides a dynamic scheduling method for distributed cloud resources based on a 5G network, applicable to a cloud-network collaborative scheduling environment consisting of 5G terminals, a scheduling control terminal, multiple distributed cloud nodes, and a 5G network management side. The scheduling control terminal can be deployed on a cloud-network orchestration platform, an edge computing management platform, or a separate resource scheduling server; the distributed cloud nodes can include edge cloud nodes, regional cloud nodes, and central cloud nodes; the 5G network management side is used to provide the network path status between the 5G terminals and each distributed cloud node.

[0048] The scheduling control terminal, distributed cloud nodes, and 5G network management side use a unified time base to record the start and end times of resource snapshots, conditional reservations, and network path states. This unified time base can be achieved through 5G network time synchronization, network time protocols, or precise time protocols. The permissible clock synchronization errors between devices can serve as one of the bases for setting subsequent preset safety margins.

[0049] This implementation method includes the following steps.

[0050] S1 receives resource scheduling requests sent by 5G terminals and obtains resource snapshots of multiple distributed cloud nodes.

[0051] When a 5G terminal generates a computing task that needs to be offloaded, it sends a resource scheduling request to the scheduling control terminal. The resource scheduling request includes a task identifier, task parameters, cloud resource requirements, and network service quality requirements.

[0052] The task parameters include the amount of input data, the amount of output data, and the amount of computation. The input data represents the amount of data that needs to be uploaded from the 5G terminal to the distributed cloud node before the task is executed. The output data represents the amount of data that needs to be returned to the 5G terminal after the task is completed. The computational workload can be expressed as the number of floating-point operations, the number of processor instructions, or the standard computational workload corresponding to the task type.

[0053] The cloud resource requirements include computing resource requirements, memory resource requirements, and storage resource requirements. They can also be categorized by task type, including graphics processing unit (GPU) resource requirements, neural network processor (NN) resource requirements, or other acceleration resource requirements. When different distributed cloud nodes use different resource metering units, the scheduling and control terminal converts all resource types to a pre-set unified metering unit.

[0054] The network service quality requirements include a minimum uplink transmission rate, a minimum downlink transmission rate, and a maximum end-to-end latency. In other embodiments, network jitter limits, packet loss rate limits, or network reliability requirements may also be included.

[0055] Each distributed cloud node generates a resource snapshot according to a preset period or when the resource status changes, and sends the resource snapshot to the scheduling control terminal. The resource snapshot includes the cloud node identifier, the available resource quantity for each type of cloud resource, the collection time, and the resource status version identifier.

[0056] The available resource quantity in a snapshot refers to the amount of resources that the corresponding cloud node can continue to allocate at the time of the resource snapshot collection, excluding resources that have already been formally allocated or conditionally reserved. The resource status version identifier can be a monotonically increasing version number or a version value generated based on the current resource status. When a distributed cloud node performs cloud resource allocation, cloud resource release, or conditional reservation, the resource status version identifier is updated.

[0057] S2 determines the amount of trusted available resources based on resource snapshots and filters candidate cloud nodes.

[0058] Because there is a time interval between the collection and transmission of resource snapshots and their use by the scheduling and control end, the scheduling and control end does not directly use the available resource quantity in the latest resource snapshot as the current available resource quantity. Instead, it corrects it based on the resource changes within a preset observation period.

[0059] Specifically, for the first The first distributed cloud node Cloud-like resources are acquired within a preset observation period, arranged chronologically by the time of collection. A resource snapshot, and The resource consumption rate is determined according to the following formula:

[0060]

[0061] in, Indicates the first The first distributed cloud node In the first resource snapshot The amount of available resources in a snapshot of cloud-like resources. Indicates the first The moment when a resource snapshot is collected.

[0062] When the available resource quantity in a subsequent resource snapshot is less than that in a previous resource snapshot, the ratio of the decrease to the acquisition time interval is used to characterize the resource consumption rate within the corresponding time period. When the available resource quantity in a subsequent resource snapshot increases, the resource consumption rate corresponding to that adjacent resource snapshot is set to zero. By selecting a larger resource consumption rate from a preset observation period, a relatively conservative estimate can be made for situations of rapid resource consumption.

[0063] The scheduling and control terminal determines the amount of reliable and available resources according to the following formula:

[0064]

[0065] in, This indicates the amount of available resources in the latest resource snapshot. Indicates the current moment. This indicates the time when the latest resource snapshot was collected. This indicates the time elapsed since the latest resource snapshot was taken. This indicates the preset resource safety reserve. This indicates the amount of reliable and available resources.

[0066] The preset resource security reserve can be set by administrators based on the cloud node type and resource type, or it can be determined based on the resource fluctuation range of the corresponding cloud node within the historical observation period. For edge cloud nodes with rapid load changes, a relatively large resource security reserve can be set; for cloud nodes with relatively stable resource status, a relatively small resource security reserve can be set.

[0067] For each type of cloud resource involved in a resource scheduling request, the scheduling control terminal compares the amount of trusted available resources with the corresponding resource demand. When each type of cloud resource on a distributed cloud node satisfies... When this happens, the distributed cloud node is identified as a candidate cloud node, where, Indicates the first Demand for cloud-like resources.

[0068] The amount of trusted available resources is used by the scheduling and control end to perform preliminary screening of cloud nodes. The current available resources of a distributed cloud node are still based on the resource status recorded by the cloud node when it receives a conditional reservation request.

[0069] When there are no candidate cloud nodes that meet the cloud resource requirements, the scheduling control terminal waits for the resource snapshot to be updated, or re-executes the candidate cloud node screening process after the preset retry interval is reached.

[0070] S3 performs conditional reservations for candidate cloud nodes based on resource status versions.

[0071] The scheduling control terminal sends conditional reservation requests to each candidate cloud node. The conditional reservation request includes a task identifier, cloud resource requirements, resource status version identifier, expected task start time, and conditional reservation validity period.

[0072] After receiving a conditional reservation request, the candidate cloud node compares the resource status version identifier carried in the conditional reservation request with the current resource status version identifier of the candidate cloud node, and queries the current available resource quantity.

[0073] When two resource status version identifiers are identical and the current available resource quantity meets the cloud resource requirement, the candidate cloud node locks the corresponding cloud resource through an atomic operation to avoid being occupied by other tasks between resource verification and resource locking. The actual locked cloud resource quantity is not less than the corresponding cloud resource requirement. When cloud resources are allocated according to a fixed specification, the actual locked cloud resource quantity can be greater than the requested cloud resource requirement.

[0074] After resource locking is completed, the candidate cloud node generates a unique conditional reservation identifier, records the amount of various locked cloud resources, the time when the conditional reservation is established, the time when the conditional reservation expires, and the corresponding task identifier, and updates the resource status version identifier.

[0075] Candidate cloud nodes determine the start time of the cloud resource reservation window as the later of the desired task start time and the earliest available time of the locked cloud resources, and determine the end time of the cloud resource reservation window as the sum of the conditional reservation establishment time and the conditional reservation validity period. Within the cloud resource reservation window, the locked cloud resources remain available for use by the corresponding tasks.

[0076] When the start time of a cloud resource reservation window is not earlier than its end time, the candidate cloud node determines that the conditional reservation has failed and releases the locked cloud resources. When the conditional reservation is successful, the candidate cloud node returns the conditional reservation result to the scheduling control terminal. The conditional reservation result includes the conditional reservation identifier, the actual locked amount of various types of cloud resources, the cloud resource reservation window, and the current resource status version identifier.

[0077] When the resource status version identifier carried in the conditional reservation request is inconsistent with the current resource status version identifier, or when the current available resource quantity cannot meet the cloud resource demand, the candidate cloud node will not perform resource locking and will return a conditional reservation failure result and an updated resource snapshot to the scheduling control terminal. The scheduling control terminal can recalculate the trusted available resource quantity based on the updated resource snapshot.

[0078] If no reservation submission instruction is received within the valid period of the conditional reservation, the candidate cloud node will determine the corresponding conditional reservation as invalid and automatically release the locked cloud resources. Conditional reservations are used to temporarily reserve resources and do not directly trigger task execution; only after receiving a valid reservation submission instruction will the candidate cloud node convert the conditional reservation resources into resources occupied by the formal task.

[0079] S4, determine the 5G network protection window and the network-cloud overlap window.

[0080] For each candidate cloud node that successfully completes a conditional reservation, the scheduling control terminal sends a network path status acquisition request to the 5G network management side. The network path status acquisition request includes the 5G terminal identifier, the network access address of the candidate cloud node, the network service quality requirements, and the time range to be queried.

[0081] The 5G network management side can determine the network path status between the 5G terminal and the candidate cloud node based on the load of the radio access network, the load of the 5G bearer network, network slicing resources, service quality flow configuration, the cell where the terminal is located, and the network access status of the candidate cloud node.

[0082] The network path state is represented by continuous time slices. Each time slice includes the start time, end time, guaranteed uplink transmission rate, guaranteed downlink transmission rate, and end-to-end latency. The time slice length can be set according to the service type and the rate of network state change, for example, 100 milliseconds, 500 milliseconds, or 1 second.

[0083] The scheduling control terminal determines the time slices that can guarantee an uplink transmission rate no lower than the minimum uplink transmission rate, a downlink transmission rate no lower than the minimum downlink transmission rate, and an end-to-end delay no greater than the maximum end-to-end delay as qualified time slices, and merges qualified time slices that are consecutive in time to obtain one or more 5G network guarantee windows.

[0084] Regarding the first Each candidate cloud node is recorded as its cloud resource reservation window. Record any corresponding 5G network guarantee window as The start and end times of the overlapping window of the network and cloud are determined according to the following formula:

[0085]

[0086]

[0087] when At that time, A valid cloud overlap window is identified when all overlap time intervals corresponding to a candidate cloud node do not meet the condition. If the cloud resource availability time of the candidate cloud node does not effectively overlap with the 5G network guarantee time, the scheduling control terminal sends a conditional reservation release instruction to the candidate cloud node.

[0088] S5 determines the target cloud node based on the required mission duration and the overlapping window of the network and cloud.

[0089] The scheduling control terminal determines the available computing speed of candidate cloud nodes based on the actual amount of computing resources locked in the conditional reservation results. Specifically, the correspondence between different processor types, processor core counts, graphics processor types, acceleration resource counts, and available computing speeds under different task types can be established in advance through benchmark task tests.

[0090] The scheduling control terminal queries the corresponding relationship based on the task type and the actual amount of locked computing resources to obtain the first... Available computing speed corresponding to each candidate cloud node .therefore, Instead of directly using the computing resource requirements before conditional reservation, the amount of computing resources actually locked after the candidate cloud node conditional reservation is successful is determined.

[0091] Regarding the first The candidate cloud node corresponding to the first For each valid overlapping window in the cloud network, the required duration of the task's protection is determined according to the following formula:

[0092]

[0093] in, and These represent the amount of input data and the amount of output data for the task, respectively, and are uniformly converted to bits during calculation. Indicates the computational load of the task; and These represent the guaranteed uplink and downlink transmission rates within the corresponding 5G network guarantee window, respectively. This indicates the available computing speed corresponding to the actual locked computing resources.

[0094] To make the calculation results relatively conservative, when the guaranteed transmission rates for different time slices within the same 5G network guarantee window are different, the minimum guaranteed uplink transmission rate and the minimum guaranteed downlink transmission rate within that window can be used as the respective values. and .

[0095] When the task input data has already been stored in the candidate cloud node, it can be... Set to zero; when the task result does not need to be returned to the 5G terminal, it can be set to zero. Set to zero. Fixed time consumption such as task environment initialization can be factored into the task's computational load, or taken into account when setting a preset safety margin.

[0096] The scheduling margin for the overlapping window of NetCloud is determined according to the following formula:

[0097]

[0098] when This indicates that the corresponding network-cloud overlap window can cover the total time required for task input data upload, task calculation, and task output data download. The scheduling control terminal selects the candidate cloud node with the largest scheduling margin from among the candidate cloud nodes with a scheduling margin greater than or equal to zero, and determines the corresponding network-cloud overlap window as the target execution window.

[0099] When multiple candidate cloud nodes have the same scheduling margin, the candidate cloud node with the earlier start time of the target execution window can be selected first; when the start time of the target execution window is also the same, the selection can be further based on the cloud resource occupation cost or the network distance between the candidate cloud node and the 5G terminal.

[0100] Before sending the reservation submission command, the scheduling control terminal queries the conditional reservation status from the target cloud node and obtains the latest network path status from the 5G network management side. Based on the latest network path status, the scheduling control terminal re-determines the 5G network guarantee window, the network-cloud overlap window, and the scheduling margin.

[0101] The preset safety margin is a preset duration threshold greater than or equal to zero, used to compensate for factors such as network status acquisition latency, clock synchronization errors, task startup delays, and resource status changes. The preset safety margin can be set according to the task type and scheduling environment, for example, to 1 second, 2 seconds, or 5 seconds.

[0102] When the conditional reservation status of the target cloud node is invalid, or the re-determined scheduling margin is less than the preset safety margin, the scheduling control terminal does not send a reservation submission instruction, releases the conditional reservation resources that are still locked in each candidate cloud node, obtains the updated resource snapshot, and then re-executes steps S2 to S5.

[0103] When the conditional reservation status of the target cloud node is valid and the redefined scheduling margin is not less than the preset safety margin, the scheduling control terminal sends a reservation submission instruction carrying the conditional reservation identifier and the target execution window to the target cloud node, and at the same time releases the conditional reservation resources of other candidate cloud nodes.

[0104] After receiving the reservation submission instruction, the target cloud node verifies the conditional reservation identifier and its status, converts the conditional reservation resources into resources for the formal task, and starts the task within the target execution window. The task can start at the beginning of the target execution window, or no later than... It is started at the right time to ensure that the task can be completed before the target execution window ends.

[0105] For example, in a resource scheduling request sent by a 5G terminal, the computing resource requirement is 8 computing resource units, the memory requirement is 16GB, the storage resource requirement is 40GB, the minimum uplink transmission rate is 80Mbit / s, the minimum downlink transmission rate is 40Mbit / s, and the maximum end-to-end latency is 20ms.

[0106] For the first candidate cloud node, in the resource snapshot taken at 10:00:00, the available resource quantity for computing resources was 16 computing resource units; in the latest resource snapshot taken at 10:00:05, the available resource quantity for computing resources was 14 computing resource units. The current time is 10:00:07, and the preset resource safety reserve is 1 computing resource unit. Therefore, the computing resource consumption rate is:

[0107]

[0108] The corresponding amount of trusted available resources is:

[0109]

[0110] Since 12.2 computing resource units is no less than 8 computing resource units, and the amount of trusted available resources corresponding to the memory and storage resources of this cloud node also meets the corresponding requirements, it is determined as a candidate cloud node.

[0111] After successful conditional reservation, the cloud resource reservation window returned by the first candidate cloud node is from 10:00:08 to 10:00:28. The 5G network guarantee window corresponding to the first candidate cloud node returned by the 5G network management side is from 10:00:05 to 10:00:25. Therefore, the corresponding network-cloud overlap window is from 10:00:08 to 10:00:25, lasting for 17 seconds.

[0112] The cloud resource reservation window returned by the other candidate cloud node is from 10:00:10 to 10:00:35, and the corresponding 5G network guarantee window is from 10:00:12 to 10:00:26. Therefore, the duration of its network-cloud overlap window is 14 seconds.

[0113] Assuming the task input data volume is 400 Mbit, the task output data volume is 80 Mbit, and the task computation volume is 600 GFLOPs, and the first candidate cloud node has a guaranteed uplink transmission rate of 100 Mbit / s and a guaranteed downlink transmission rate of 50 Mbit / s, and the actual locked computing resources correspond to an available computing rate of 100 GFLOPs / s, then the required guaranteed duration for the task is:

[0114] Second

[0115] The scheduling margin for the first candidate cloud node is:

[0116] Second

[0117] The second candidate cloud node has a guaranteed uplink transmission rate of 80 Mbit / s, a guaranteed downlink transmission rate of 40 Mbit / s, and an available computing rate of 150 GFLOP / s. Therefore, the required duration for its task is:

[0118] Second

[0119] The scheduling margin for the second candidate cloud node is 3 seconds. Therefore, the scheduling control terminal prioritizes the first candidate cloud node as the target cloud node.

[0120] Before sending the reservation submission command, the scheduling control terminal re-acquires the conditional reservation status and latest network path status of the first candidate cloud node. When the duration of the re-determined network-cloud overlap window is 15 seconds, its scheduling margin is 3.4 seconds. If the preset safety margin is 2 seconds, the re-determined scheduling margin is not less than the preset safety margin. The scheduling control terminal sends the reservation submission command to the first candidate cloud node and releases the conditional reservation resources of the second candidate cloud node.

[0121] In another embodiment, the present invention also provides a distributed cloud resource dynamic scheduling system based on a 5G network, including a request and snapshot acquisition module, a trusted resource determination module, a condition reservation module, a window matching module, and a scheduling execution module.

[0122] The request and snapshot acquisition module is used to receive resource scheduling requests and acquire resource snapshots of distributed cloud nodes; the trusted resource determination module is used to calculate resource consumption rate and trusted available resource quantity, and filter candidate cloud nodes; the conditional reservation module is used to send conditional reservation requests carrying resource status version identifiers, and acquire conditional reservation identifiers, locked cloud resource quantities, and cloud resource reservation windows; the window matching module is used to acquire 5G network path status, determine 5G network guarantee windows and network-cloud overlap windows; the scheduling execution module is used to determine the required guarantee duration and scheduling margin for the task, select target cloud nodes, send reservation submission instructions, and release other conditionally reserved resources.

[0123] Each of the above modules can be implemented by one or more processors executing computer programs stored in memory, or they can be deployed on different servers and work together through a communication network. The computer programs can be stored in a hard disk, solid-state drive, read-only memory, random access memory, flash memory, or other computer-readable storage media.

[0124] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic scheduling of distributed cloud resources based on 5G networks, characterized in that, Includes the following steps: S1, receive a resource scheduling request sent by a 5G terminal and obtain resource snapshots of multiple distributed cloud nodes. The resource scheduling request includes task parameters, cloud resource requirements and network service quality requirements. The resource snapshots include the available resource quantity, collection time and resource status version identifier of various cloud resources. S2, based on the ratio of the decrease in available resources between adjacent resource snapshots of each distributed cloud node within a preset observation period to the collection time interval, determine the resource consumption rate of various cloud resources, and based on the resource consumption rate and the time distance between the latest resource snapshot and the current time, correct the available resources in the latest resource snapshot to obtain the reliable available resources, and determine the distributed cloud node whose reliable available resources meet the cloud resource requirements as a candidate cloud node; S3, send a conditional reservation request carrying the corresponding resource status version identifier to the candidate cloud node, and obtain the conditional reservation result returned by the candidate cloud node when the resource status version verification is passed and the current available resource quantity meets the cloud resource requirements. The conditional reservation result includes a conditional reservation identifier, the locked cloud resource quantity and the cloud resource reservation window. The cloud resource reservation window starts at the time when the locked cloud resource can be used and ends at the time when the conditional reservation expires. S4, obtain the network path status from the 5G terminal to the candidate cloud node that has been successfully booked under each condition, returned by the 5G network management side, determine the time interval during which the network path status continuously meets the network service quality requirements as the 5G network guarantee window, and determine the time interval during which the 5G network guarantee window and the corresponding cloud resource booking window overlap as the network-cloud overlap window. S5. Based on the task parameters, the locked cloud resource quantities, and the network path status, determine the required guarantee duration for each candidate cloud node's task. Select a target cloud node from the candidate cloud nodes whose network-cloud overlap window duration is not less than the required guarantee duration for the corresponding task. Send a reservation submission instruction carrying the corresponding conditional reservation identifier to the target cloud node and release the conditional reservation resources of the candidate cloud nodes other than the target cloud node.

2. The method for dynamic scheduling of distributed cloud resources based on a 5G network according to claim 1, characterized in that, The task parameters include at least the amount of task input data, the amount of task output data, and the amount of task computation; the cloud resource requirements include at least the amount of computing resources, memory, and storage resources; and the network service quality requirements include at least the minimum uplink transmission rate, the minimum downlink transmission rate, and the maximum end-to-end latency.

3. The method for dynamic scheduling of distributed cloud resources based on a 5G network according to claim 2, characterized in that, Within the preset observation period, the number of resource snapshots arranged chronologically by acquisition time is: ,and For the first The first distributed cloud node For cloud-like resources, the resource consumption rate is determined according to the following formula: The amount of trusted available resources is determined according to the following formula: in, Indicates the first The first distributed cloud node In the first resource snapshot The amount of available resources in a snapshot of cloud-like resources. Indicates the first The time of collection of a resource snapshot. Indicates the first Resource consumption rate of cloud-like resources Indicates the current moment. Indicates the first The preset resource security reserve for cloud-like resources. Indicates the first The amount of reliable and available cloud-like resources; when each type of cloud resource on a distributed cloud node satisfies At that time, the distributed cloud node is identified as a candidate cloud node. Indicating the cloud resource requirement, the first... Demand for cloud-like resources.

4. The method for dynamic scheduling of distributed cloud resources based on a 5G network according to claim 1, characterized in that, The resource status version identifier is updated when cloud resource allocation, cloud resource release, or conditional reservation occurs on the corresponding distributed cloud node; the conditional reservation request also includes the expected task start time and the validity period of the conditional reservation. The candidate cloud node compares the resource status version identifier carried in the conditional reservation request with the current resource status version identifier. If they match and the current available resources meet the cloud resource requirements, it locks no fewer than the required cloud resources, generates the conditional reservation identifier, and determines the later of the expected task start time and the earliest available time of the locked cloud resources as the start time of the cloud resource reservation window. The sum of the conditional reservation establishment time and the conditional reservation validity period is determined as the end time of the cloud resource reservation window. If the resource status version identifiers do not match or the current available resources do not meet the cloud resource requirements, it returns an updated resource snapshot containing the current resource status version identifier.

5. The method for dynamic scheduling of distributed cloud resources based on a 5G network according to claim 2, characterized in that, The network path state includes multiple consecutive time slices, each corresponding to a guaranteed uplink transmission rate, a guaranteed downlink transmission rate, and an end-to-end latency. Adjacent time slices that guarantee an uplink transmission rate no lower than the minimum uplink transmission rate, a downlink transmission rate no lower than the minimum downlink transmission rate, and an end-to-end latency no greater than the maximum end-to-end latency are merged to obtain the 5G network guarantee window.

6. The method for dynamic scheduling of distributed cloud resources based on a 5G network according to claim 5, characterized in that, Regarding the first Each candidate cloud node is recorded as its cloud resource reservation window. Record any corresponding 5G network guarantee window as The start and end times of the corresponding cloud overlap window are determined according to the following formula: when At that time, A valid cloud overlap window is defined as one where none of the overlap time intervals corresponding to a candidate cloud node meet the condition. If the candidate cloud node does not have a valid network-cloud overlap window, the conditional reservation resources of the candidate cloud node are released.

7. A method for dynamic scheduling of distributed cloud resources based on a 5G network according to claim 6, characterized in that, Regarding the first The candidate cloud node corresponding to the first For each valid overlapping window in the cloud network, the required duration of the task is determined according to the following formula: in, Indicates the amount of input data for the task. Indicates the amount of data output by the task. Indicates the computational load of the task. and These represent the guaranteed uplink and downlink transmission rates within the corresponding 5G network guarantee window, respectively. Indicates according to the first The available computing rate is determined by the amount of locked computing resources returned by each candidate cloud node; according to Determine the scheduling margin of the corresponding network-cloud overlap window, and from the candidate cloud nodes with scheduling margins greater than or equal to zero, determine the candidate cloud node with the largest scheduling margin and its corresponding network-cloud overlap window as the target cloud node and target execution window, respectively.

8. A method for dynamic scheduling of distributed cloud resources based on a 5G network according to claim 7, characterized in that, Before sending the reservation submission command to the target cloud node, the conditional reservation status corresponding to the target cloud node and the latest network path status returned by the 5G network management side are obtained. The network-cloud overlap window and its scheduling margin are re-determined based on the conditional reservation status and the latest network path status. When the conditional reservation status is invalid, or the re-determined scheduling margin is less than the preset safety margin, the conditional reservation resources that are still locked are released, an updated resource snapshot is obtained, and steps S2 to S5 are re-executed. When the conditional reservation status is valid and the re-determined scheduling margin is not less than the preset safety margin, the reservation submission command is sent, wherein the preset safety margin is a preset duration threshold that is greater than or equal to zero.

9. A distributed cloud resource dynamic scheduling system based on a 5G network, characterized in that, include: The request and snapshot acquisition module is used to receive resource scheduling requests sent by 5G terminals and acquire resource snapshots of multiple distributed cloud nodes. The resource scheduling request includes task parameters, cloud resource requirements and network service quality requirements. The resource snapshot includes the available resource quantity, collection time and resource status version identifier of various cloud resources. The trusted resource determination module is used to determine the resource consumption rate of various cloud resources based on the ratio of the decrease in available resources between adjacent resource snapshots to the collection time interval, correct the available resources of the snapshot based on the resource consumption rate and the time distance between the latest resource snapshot and the current time to obtain the trusted available resources, and filter candidate cloud nodes based on the trusted available resources. The conditional reservation module is used to send a conditional reservation request carrying the corresponding resource status version identifier to the candidate cloud node, and obtain the conditional reservation identifier, the locked cloud resource quantity, and the cloud resource reservation window returned by the candidate cloud node when the resource status version verification is passed and the current available resource quantity meets the cloud resource requirements. The cloud resource reservation window starts at the time when the locked cloud resource can be used and ends at the time when the conditional reservation expires. The window matching module is used to obtain the network path status from the 5G terminal to the candidate cloud node that has been successfully booked under each condition, returned by the 5G network management side, determine the 5G network guarantee window based on the time interval in the network path status that continuously meets the network service quality requirements, and determine the overlapping time interval between the 5G network guarantee window and the corresponding cloud resource booking window as the network-cloud overlapping window. The scheduling and execution module is used to determine the required guarantee duration for the task based on the task parameters, the locked cloud resource quantities, and the network path status; determine the target cloud node from the candidate cloud nodes whose duration of network-cloud overlap window is not less than the required guarantee duration for the corresponding task; send a reservation submission instruction carrying the corresponding conditional reservation identifier to the target cloud node; and release the conditional reservation resources of the candidate cloud nodes other than the target cloud node.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1 to 8.

Images