Cloud computing network resource scheduling system and method, storage medium and equipment

Abstract

The invention discloses a cloud computing network resource scheduling system and method, a storage medium and equipment. The cloud computing network resource scheduling system comprises a working load layer, a scheduling control layer and a data center layer, the working load layer is used for providing a data stream demand needing cloud computing, splitting the data stream into a plurality of sub-data packets, and scheduling the sub-data packets to a scheduling queue of the scheduling control layer according to priorities to wait for transmission; the scheduling control layer generates a network resource scheduling strategy by using a deep Q network according to the state information of the sub-data packets to be transmitted and the resource state information of the virtual resource pool reported by the data center layer, and sends a scheduling instruction to the data center layer; and the data center layer abstracts hardware network resources into a configurable virtual resource pool, and allocates the virtual resource pool to execute the scheduling instruction according to the scheduling instruction. According to the method, the network resource scheduling strategy is dynamically generated by adopting the deep Q network, real-time network resource scheduling can be realized in a large-scale and high-dynamic environment, and the calculation complexity is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of network resource scheduling, and in particular to a cloud computing network resource scheduling system, method, storage medium and device. Background Art

[0002] As a mature computing service, cloud computing provides an efficient and cost-effective solution for big data processing, attracting widespread attention and playing a vital role in industrial applications. In recent years, with the growing demand for cloud computing tasks and increasing user expectations for quality of service, the allocation of cloud computing network resources has become increasingly critical, posing greater challenges to resource scheduling and management.

[0003] Currently, traditional static resource allocation methods allocate computing resources in advance based on predefined rules or historical data. This method works well when resource demand is relatively stable, but when faced with dynamically changing workloads, it may lead to resource waste or insufficient resources and fail to adapt to demand fluctuations in a timely manner.

[0004] Faced with the urgent need for cloud computing network resource scheduling, an increasing number of modern technologies are being applied to this task. Some cloud computing resource scheduling approaches employ heuristic algorithms, such as genetic algorithms and particle swarm optimization. While these methods can optimize resource allocation to a certain extent, they rely on initial parameter settings and the specific nature of the problem. However, due to their reliance on initial parameter settings and the specific nature of the problem, they can become trapped in local optima and exhibit high computational complexity, making real-time scheduling difficult in large-scale and highly dynamic environments. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides a cloud computing network resource scheduling system, method, storage medium and device, which adopts a deep Q network to dynamically generate network resource scheduling strategies, can realize real-time network resource scheduling in large-scale and highly dynamic environments, and reduce computational complexity.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solutions: A cloud computing network resource scheduling system includes: a workload layer, a scheduling control layer, and a data center layer;

[0007] The workload layer is used to provide data flow requirements that need to be cloud-computing, and split the data flow into several sub-data packets, which are scheduled according to priority to the scheduling queue of the scheduling control layer to wait for transmission;

[0008] The scheduling control layer generates a network resource scheduling strategy using a deep Q network based on the status information of the sub-data packets waiting to be transmitted and the resource status information of the virtual resource pool reported by the data center layer, and sends a scheduling instruction to the data center layer;

[0009] The data center layer abstracts hardware network resources into configurable virtual resource pools, and allocates virtual resource pools to execute scheduling instructions according to scheduling instructions.

[0010] Furthermore, the scheduling control layer includes: a network monitor, a resource monitor, a scheduling policy module and a service level agreement storage module;

[0011] The network monitor is used to obtain status information of sub-data packets waiting to be transmitted and transmit it to the scheduling strategy module;

[0012] The resource monitor is used to obtain resource status information of the virtual resource pool reported by the data center layer and transmit it to the scheduling strategy module;

[0013] The service level agreement storage module is used to store the service level agreement signed between the cloud computing service provider and the user;

[0014] The scheduling strategy module uses a deep Q network to construct a multi-objective reward function. The status information of sub-packets waiting for transmission, the service level agreement, and the resource status information of the virtual resource pool are used as the state of the deep Q network. The network resource scheduling strategy generated by the corresponding state is used as the action of the deep Q network. The network resource scheduling strategy with the minimum multi-objective reward function is used as the optimal network resource scheduling strategy and is sent to the data center layer.

[0015] Furthermore, the status information of the sub-data packets waiting for transmission includes: queue information and running status information of the sub-data packets waiting for transmission, wherein the running status information of the sub-data packets waiting for transmission includes: arrival time of the sub-data packets, waiting time, initial data volume, and memory, computing power and bandwidth required for sub-data transmission.

[0016] Furthermore, the resource status information of the virtual resource pool includes: CPU usage, memory usage, bandwidth utilization, running status and number of running virtual machines of each virtual resource pool.

[0017] Furthermore, the multi-objective reward function R is composed of the time reward function R that completes data flow scheduling using the corresponding network resource scheduling strategy. makespan and the energy consumption reward function R using the corresponding network resource scheduling strategy p composition:

[0018] R=λR makespan +(1-λ)R p

[0019] Where λ represents R makespan The weight coefficient of I represents the number of virtual resource pools involved in the network resource scheduling strategy, i represents the index of I, represents the static energy consumption of network resource scheduling by the i-th virtual resource pool, Represents the dynamic energy consumption of network resource scheduling by the i-th virtual resource pool.

[0020] Furthermore, the time reward function R makespan The calculation process is:

[0021]

[0022] Among them, T e Indicates the transmission time of the data stream represents the transmission time of the xth sub-packet; T w represents the waiting time of the data flow, Indicates the waiting time of the xth sub-packet.

[0023] Furthermore, the energy consumption reward function R p The calculation process is:

[0024] i. Obtain the virtual resource pools involved in the network resource scheduling strategy and calculate the resource utilization of each virtual resource pool Among them, N i Indicates the number of sub-packets being processed by the i-th virtual resource pool, Indicates the maximum number of sub-packets that the i-th virtual resource pool can process;

[0025] ii. If U i >0, static energy consumption of the i-th virtual resource pool is the set value; otherwise, the static energy consumption of the i-th virtual resource pool is 0;

[0026] iii. If Dynamic energy consumption of the i-th virtual resource pool Otherwise, the dynamic energy consumption of the i-th virtual resource pool in, represents the optimal resource utilization of the i-th virtual resource pool, α represents U i The coefficient of β represents The coefficient of δ represents The coefficient of

[0027] iv. Superimpose the static energy consumption and dynamic energy consumption of all virtual resource pools in the corresponding network resource scheduling strategy to obtain the energy consumption reward function.

[0028] Furthermore, the present invention also provides a cloud computing network resource scheduling method, comprising the following steps:

[0029] Step S1: Obtain the data stream that needs to be cloud-computing, split the data stream into several sub-data packets, and wait for transmission according to priority;

[0030] Step S2: abstracting the hardware network resources into a configurable virtual resource pool and obtaining resource status information of the virtual resource pool;

[0031] Step S3: Dynamically generate a network resource scheduling strategy using a deep Q network based on the status information of the sub-data packets waiting to be transmitted and the resource status information of the virtual resource pool;

[0032] Step S4: Perform network resource scheduling on the data stream that needs to be cloud-computing according to the generated network resource scheduling policy.

[0033] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the cloud computing network resource scheduling method.

[0034] Furthermore, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the cloud computing network resource scheduling method is implemented.

[0035] Compared with the prior art, the present invention has the following beneficial effects: the cloud computing network resource scheduling system and method of the present invention utilizes a deep Q network to generate a network resource scheduling strategy, and realizes dynamic allocation of network resources for real-time data streams by considering the combined influence of the time reward function and the energy consumption reward function, so that in low-load periods, the generated network resource scheduling strategy increases the energy consumption budget to accelerate tasks and shorten completion time. In high-load periods, the generated network resource scheduling strategy reduces the energy consumption budget to avoid resource contention and ensure that tasks are completed on time, thereby improving network resource utilization. In addition, the weight coefficients of the time reward function and the energy consumption reward function are dynamically adjusted according to the status information of the sub-data packets waiting for transmission and the resource status information of the virtual resource pool reported by the data center layer to adapt to different scenarios; at the same time, the static energy consumption and dynamic energy consumption in the network resource scheduling process are fully considered, and equipment overload is avoided through dynamic energy consumption optimization. It can timely adapt to the scenario of fluctuating network resource scheduling demand and realize real-time scheduling in large-scale and highly dynamic cloud computing tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of a cloud computing network resource scheduling system according to the present invention;

[0037] Figure 2 This is a flow chart of the cloud computing network resource scheduling method of the present invention. DETAILED DESCRIPTION

[0038] The technical solution of the present invention will be further explained below with reference to the accompanying drawings.

[0039] like Figure 1 Schematic diagram of a cloud computing network resource scheduling system according to the present invention, which includes: a workload layer, a scheduling control layer, and a data center layer;

[0040] The workload layer is used to provide data flow requirements for cloud computing and split the data flow into several sub-packets. These are scheduled according to priority to the scheduling queue of the scheduling control layer for transmission. The sub-packets can be dynamically allocated to different queues to avoid overloading a single link. Priority scheduling can also prevent high-priority sub-packets from waiting too long.

[0041] The scheduling control layer uses a deep Q network to generate a network resource scheduling strategy based on the status information of sub-packets waiting to be transmitted and the resource status information of the virtual resource pool reported by the data center layer, and sends scheduling instructions to the data center layer. This overcomes the problem of existing static resource allocation schemes that may lead to resource waste or insufficient resources when facing dynamically changing workloads. It can realize dynamic allocation of network resources for real-time data streams, can promptly adapt to demand fluctuations, and achieve real-time scheduling in large-scale and highly dynamic environments.

[0042] The data center layer is composed of a large number of heterogeneous hardware network resources. These hardware network resources vary significantly in configuration and architecture, posing significant challenges to network resource scheduling and management. To address this issue, network virtualization technology abstracts hardware network resources into configurable virtual resource pools. These virtual resource pools are then allocated to execute scheduling instructions. Network virtualization technology breaks the isolation between physical network devices, enabling more flexible optimization of network resource utilization.

[0043] The scheduling control layer in the present invention includes: a network monitor, a resource monitor, a scheduling policy module and a service level agreement storage module;

[0044] The network monitor is used to obtain status information of sub-data packets waiting to be transmitted and transmit it to the scheduling strategy module; specifically, the status information of the sub-data packets waiting to be transmitted includes: queue information and running status information of the sub-data packets waiting to be transmitted, wherein the running status information of the sub-data packets waiting to be transmitted includes: arrival time of the sub-data packets, waiting time, initial data volume, and memory, computing power and bandwidth required for sub-data transmission.

[0045] The resource monitor is used to obtain the resource status information of the virtual resource pool reported by the data center layer and transmit it to the scheduling policy module; specifically, the resource status information of the virtual resource pool includes: CPU utilization, memory utilization, bandwidth utilization, operating status and number of running virtual machines of each virtual resource pool.

[0046] The service level agreement storage module is used to store the service level agreement signed between the cloud computing service provider and the user to ensure satisfactory network performance and service availability;

[0047] The scheduling strategy module uses a deep Q-network to construct a multi-objective reward function. The state information of sub-packets awaiting transmission, service-level agreements, and resource status information of virtual resource pools are used as the state of the deep Q-network. The network resource scheduling strategy generated by the corresponding state is used as the action of the deep Q-network. The network resource scheduling strategy with the minimum multi-objective reward function is regarded as the optimal network resource scheduling strategy and sent to the data center layer. Each virtual resource pool can carry a different number of sub-packets, and each virtual resource pool has two states: open and closed. Therefore, for each deep Q-network state, there can be multiple corresponding network resource scheduling strategies. Therefore, the deep Q-network maps the current state to the corresponding action and evaluates the action value based on the expected reward function. It learns the optimal decision action corresponding to the state, reducing computational complexity.

[0048] The multi-objective reward function R in the present invention is composed of the time reward function R that completes data flow scheduling using the corresponding network resource scheduling strategy makespan and the energy consumption reward function R using the corresponding network resource scheduling strategy p The proposed algorithm is composed of a time reward function and an energy consumption reward function, and the network resources for real-time data streams are dynamically allocated by considering the combined influence of the time reward function and the energy consumption reward function. In this way, during low-load periods, the generated network resource scheduling strategy increases the energy consumption budget to accelerate tasks and shorten completion time. During high-load periods, the generated network resource scheduling strategy reduces the energy consumption budget to avoid resource contention and ensure that tasks are completed on time, thereby improving network resource utilization. In addition, the weight coefficients of the time reward function and the energy consumption reward function are dynamically adjusted according to the status information of the sub-packets waiting for transmission and the resource status information of the virtual resource pool reported by the data center layer to adapt to different scenarios.

[0049] The calculation process of the multi-objective reward function R in the present invention is:

[0050] R=λR makespan +(1-λ)R p

[0051] Where λ represents R makespan The weight coefficient of I represents the number of virtual resource pools involved in the network resource scheduling strategy, i represents the index of I, represents the static energy consumption of network resource scheduling by the i-th virtual resource pool, Represents the dynamic energy consumption of network resource scheduling by the i-th virtual resource pool.

[0052] The time reward function R in this invention makespan The calculation process is:

[0053]

[0054] Among them, T e Indicates the transmission time of the data stream represents the transmission time of the xth sub-packet; T w represents the waiting time of the data flow, Indicates the waiting time of the xth sub-packet.

[0055] The energy consumption reward function R in this invention p The calculation process is:

[0056] i. Obtain the virtual resource pools involved in the network resource scheduling strategy and calculate the resource utilization of each virtual resource pool Among them, N i Indicates the number of sub-packets being processed by the i-th virtual resource pool, Indicates the maximum number of sub-packets that the i-th virtual resource pool can process;

[0057] ii. If U i >0, static energy consumption of the i-th virtual resource pool is the set value; otherwise, the static energy consumption of the i-th virtual resource pool is 0;

[0058] iii. If The dynamic energy consumption of the i-th virtual resource pool increases linearly. Otherwise, the dynamic energy consumption of the i-th virtual resource pool increases nonlinearly at a faster rate. in, represents the optimal resource utilization of the i-th virtual resource pool, α represents U i The coefficient of β represents The coefficient of δ represents The coefficient of

[0059] iv. Superimpose the static energy consumption and dynamic energy consumption of all virtual resource pools in the corresponding network resource scheduling strategy to obtain the energy consumption reward function.

[0060] In one technical solution of the present invention, the deep Q network is trained by the stochastic gradient descent method, which can effectively process large-scale data, reduce computing resource consumption, accelerate convergence, avoid local optimal traps, and is very suitable for online learning and applications in dynamic environments.

[0061] In one technical solution of the present invention, hardware network resources are clustered according to geographical location or logical requirements to form network device groups, and are interconnected through bidirectional high-speed channels, while network resources within the same network device group are connected to each other through local channels.

[0062] like Figure 2 In a technical solution of the present invention, a cloud computing network resource scheduling method is also provided, comprising the following steps:

[0063] Step S1: Obtain the data stream that needs to be cloud-computing by subscribing to real-time data changes in the database or using a message queue system, split the data stream into several sub-data packets, and wait for transmission according to priority;

[0064] Step S2: abstracting the hardware network resources into a configurable virtual resource pool and obtaining resource status information of the virtual resource pool;

[0065] Step S3: Dynamically generate a network resource scheduling strategy using a deep Q network based on the status information of the sub-data packets waiting to be transmitted and the resource status information of the virtual resource pool;

[0066] Step S4: Perform network resource scheduling on the data stream that needs to be cloud-computing according to the generated network resource scheduling policy.

[0067] The cloud computing network resource scheduling method proposed in this paper balances energy consumption and task completion time by adjusting the reward ratios for different optimization objectives in a deep Q network. Experimental results show that compared with the baseline algorithm, the proposed method exhibits significant advantages in balancing energy consumption and task completion time.

[0068] In a technical solution of the present invention, a computer-readable storage medium is further provided, storing a computer program, wherein the computer program enables a computer to execute the cloud computing network resource scheduling method.

[0069] In a technical solution of the present invention, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the cloud computing network resource scheduling method is implemented.

[0070] In the embodiments disclosed herein, computer storage media can be tangible media that can contain or store programs for use by or in conjunction with an instruction execution system, device, or apparatus. Computer storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0071] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0072] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A cloud computing network resource scheduling system, characterized in that: include: Workload layer, scheduling control layer, and data center layer; The workload layer is used to provide data flow requirements that need to be cloud-computing, and split the data flow into several sub-data packets, which are scheduled according to priority to the scheduling queue of the scheduling control layer to wait for transmission; The scheduling control layer generates a network resource scheduling strategy using a deep Q network based on the status information of the sub-data packets waiting to be transmitted and the resource status information of the virtual resource pool reported by the data center layer, and sends a scheduling instruction to the data center layer; The data center layer abstracts hardware network resources into configurable virtual resource pools, and allocates virtual resource pools to execute scheduling instructions according to scheduling instructions.

2. A cloud computing network resource scheduling system according to claim 1, characterized in that: The scheduling control layer includes: a network monitor, a resource monitor, a scheduling policy module and a service level agreement storage module; The network monitor is used to obtain status information of sub-data packets waiting to be transmitted and transmit it to the scheduling strategy module; The resource monitor is used to obtain resource status information of the virtual resource pool reported by the data center layer and transmit it to the scheduling strategy module; The service level agreement storage module is used to store the service level agreement signed between the cloud computing service provider and the user; The scheduling strategy module uses a deep Q network to construct a multi-objective reward function. The status information of the sub-packets waiting for transmission, the service level agreement, and the resource status information of the virtual resource pool are used as the state of the deep Q network. The network resource scheduling strategy generated by the corresponding state is used as the action of the deep Q network. The network resource scheduling strategy with the minimum multi-objective reward function is used as the optimal network resource scheduling strategy and is sent to the data center layer.

3. A cloud computing network resource scheduling system according to claim 2, characterized in that: The status information of the sub-data packets waiting for transmission includes: queue information and running status information of the sub-data packets waiting for transmission, wherein the running status information of the sub-data packets waiting for transmission includes: arrival time, waiting time, initial data volume, and memory, computing power and bandwidth required for sub-data transmission.

4. A cloud computing network resource scheduling system according to claim 2, characterized in that: The resource status information of the virtual resource pool includes: CPU usage, memory usage, bandwidth utilization, running status and number of running virtual machines of each virtual resource pool.

5. A cloud computing network resource scheduling system according to claim 2, characterized in that: The multi-objective reward function R is composed of the time reward function R that completes data flow scheduling using the corresponding network resource scheduling strategy. makespan and the energy consumption reward function R using the corresponding network resource scheduling strategy p composition: R=λR makespan +(1-λ)R p Where λ represents R makespan The weight coefficient of I represents the number of virtual resource pools involved in the network resource scheduling strategy, i represents the index of I, represents the static energy consumption of network resource scheduling by the i-th virtual resource pool, Represents the dynamic energy consumption of network resource scheduling by the i-th virtual resource pool.

6. A cloud computing network resource scheduling system according to claim 5, characterized in that: The time reward function R makespan The calculation process is: Among them, T e Indicates the transmission time of the data stream represents the transmission time of the xth sub-packet; T w represents the waiting time of the data flow, Indicates the waiting time of the xth sub-packet.

7. A cloud computing network resource scheduling system according to claim 5, characterized in that: The energy consumption reward function R p The calculation process is: i. Obtain the virtual resource pools involved in the network resource scheduling strategy and calculate the resource utilization of each virtual resource pool Among them, N i Indicates the number of sub-packets being processed by the i-th virtual resource pool, Indicates the maximum number of sub-packets that the i-th virtual resource pool can process; ii. If U i >0, static energy consumption of the i-th virtual resource pool is the set value; otherwise, the static energy consumption of the i-th virtual resource pool is 0; iii. If Dynamic energy consumption of the i-th virtual resource pool Otherwise, the dynamic energy consumption of the i-th virtual resource pool in, represents the optimal resource utilization of the i-th virtual resource pool, α represents U i The coefficient of β represents The coefficient of δ represents The coefficient of iv. Superimpose the static energy consumption and dynamic energy consumption of all virtual resource pools in the corresponding network resource scheduling strategy to obtain the energy consumption reward function.

8. A cloud computing network resource scheduling method of the cloud computing network resource scheduling system according to any one of claims 1 to 7, characterized in that: The steps include: Step S1: Obtain the data stream that needs to be cloud-computing, split the data stream into several sub-data packets, and wait for transmission according to priority; Step S2: abstracting the hardware network resources into a configurable virtual resource pool and obtaining resource status information of the virtual resource pool; Step S3: Dynamically generate a network resource scheduling strategy using a deep Q network based on the status information of the sub-data packets waiting to be transmitted and the resource status information of the virtual resource pool; Step S4: Perform network resource scheduling on the data stream that needs to be cloud-computing according to the generated network resource scheduling policy.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to execute the cloud computing network resource scheduling method according to claim 8.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the cloud computing network resource scheduling method according to claim 8 is implemented.

Images