CN111953549A - An online optimized resource management method for maritime edge nodes - Google Patents

Abstract

The invention relates to an online optimization resource management method for maritime edge nodes, belonging to the technical field of communication networks. The invention provides an online optimization resource management algorithm for ensuring the service quality of different applications and improving the efficiency of communication resources. The algorithm requires neither resource cost function nor service quality constraint function on resource scheduling nodes. Furthermore, the gradient information usually required in learning policies is also not required. The algorithm performs resource management of computing resources and communication resources in each time slot based only on the observations of the last time slot. By allocating resources on a slot-by-slot basis, communication costs can be minimized while long-term delay constraints can be satisfied to suit maritime scenarios.

Description

An online optimized resource management method for maritime edge nodes

technical field

The invention relates to the technical field of communication networks, in particular to an online optimization resource management method for maritime edge nodes.

Background technique

With the rapid development of the marine economy, more and more maritime applications are emerging, resulting in more sensor deployment and data generation. The current maritime Internet is limited by limited communication resources and poor transmission environment, and can only provide a low transmission rate. Traditional resource management methods have been unable to cope with the large amount of detection data feedback generated in today's maritime applications. In order to utilize the existing resources more effectively and ensure the QoS (Quality of Service) of the application program, an online optimization resource management algorithm is urgently needed to ensure the service quality of different applications and improve the efficiency of communication resources.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an online optimization resource management algorithm which can ensure the service quality of different applications and improve the efficiency of communication resources in order to overcome the defects of the prior art. The algorithm requires neither resource cost function nor QoS constraint function on resource scheduling nodes. Furthermore, the gradient information usually required in learning policies is also not required. Instead, the algorithm performs resource management of computational and communication resources in each slot only based on the observations of the last slot. By allocating resources on a slot-by-slot basis, communication costs can be minimized while long-term delay constraints can be satisfied to suit maritime scenarios.

The object of the present invention can be realized through the following technical solutions:

An online optimization resource management method for maritime edge nodes, comprising the following steps:

S1. Construct a marine communication network model. In the marine communication network, underwater detectors, underwater robots, underwater unmanned vehicles and other underwater equipment for monitoring, detection and other functions are deployed in the underwater layer. Ships are deployed at the edge layer, and edge nodes such as buoys will collect data from nearby underwater equipment. Data centers, radars, and base stations located at the ground access layer process the data sent by these edge nodes. Focus on the feedback of data collected by these edge nodes. The underwater device produces streams of different applications during the observation period T (a stream is the entity that provides it with QoS, it is a sequence of packets from the transmitter to the receiver). The transmission period is time-slotted, and the Poisson model is used to describe the data distribution generated by different applications, that is, the data arrival rate γ _{i of application i} follows Poisson distribution, and different applications have different QoS requirements;

S2. The available resources on the edge node include communication resources allocated to different application programs i to perform communication processing on the collected information data and computing resources allocated to different application programs i to perform computing processing on the collected information data. For data transmission, edge nodes can allocate the total bandwidth B(t) to different frequency bandwidths for different applications i of t slots. Before data transmission, edge nodes can also allocate different computing resources for different applications i. For different applications i, by given the amount of transmitted data _{li and the allocated bandwidth B i ,} since the communication rate and calculation rate are constant, the transmission time and calculation time of data in time slot t can be determined. At the same time, a decision can be made at time slot t+1 without waiting for the completion of transmission and calculation according to the data information obtained at time t.

The total delay of application i in time slot t is:

和计算延迟

通信延迟

表示在t时隙应用程序i对收集到的信息数据进行通信处理的延迟，是关于传输数据量l_i(t)和分配的带宽B_i(t)的函数，计算延迟

表示在t时隙应用程序i对收集到的信息数据进行计算处理的延迟，是关于P_i(t)的函数，其中，L_i(t)表示在t时隙应用程序i在压缩前新到达的数据量，P_i(t)表示在t时隙应用程序i分配的计算资源，每个时隙中完成的数据流可以是分数，从而不需要将延迟的计算限制在每个时隙中，B_i(t)表示在t时隙分配给应用程序i的带宽，h₀(B_i(t))表示带宽B_i的传输速率，对于K个应用程序在t时隙分别分配的带宽为{B₁(t),B₂(t)···,B_K(t)}；The delay d _i (t) for different application i in time slot t includes the communication delay

and computational delay

communication delay

Represents the delay in communication processing of the collected information data by application i at time slot t, and is a function of the amount of transmitted data l _i (t) and the allocated bandwidth B _i (t), calculate the delay

represents the delay in the computational processing of the collected information data by application i in time slot t, and is a function of P _i (t), where _Li (t) represents the new arrival of application i in time slot t before compression The amount of data, P _i (t) represents the computational resources allocated by application i in time slot t, the data flow completed in each time slot can be fractional, so that there is no need to limit the calculation of the delay to each time slot, B _i (t) represents the bandwidth allocated to application i in time slot t, h ₀ (B _i (t)) represents the transmission rate of bandwidth B _i , and the bandwidth allocated to K applications in time slot t is { B ₁ (t),B ₂ (t)...,B _K (t)};

S3. Initialize the original iteration y ₁ , λ ₁ , parameters δ and β, step size α and u;

S4. The edge node determines that the resource allocation operation set x _t is obtained based on the iteration value y _t through y _t =x _t +δu _t , where u represents a random unit vector, time slot t starts from 1, and loops from step S4 to step S7 Carry out, the time slot t is incremented by 1 every cycle;

S5. The edge node collects the cost value f _t (x _t ) and the constraint value g _t (x _t ) at the query point;

S6, the edge node updates the dual variable λ _t ;

S7, the edge node updates the communication resource allocation set B _t at the time slot t, and updates the computing resource allocation set P _{t at the time slot t} ;

Among them, the fit metric Fit _T is used to evaluate the degree of constraint satisfaction, and the smaller the Fit _T , the smaller the constraint violation.

The beneficial effects of the present invention are that the implementation of resource management is easier and more suitable for marine scenarios. Model the information feedback problem appropriately. Scheduling time so that decisions are easily processed at each time slot; application flows can be processed and transmitted without time slot constraints; online solutions to optimization problems co-schedule computing and communication resources while guaranteeing delay constraints; with the help of gradient estimation, Resources can be allocated without explicit costs and constraints, which is suitable for existing maritime communication networks.

Description of drawings

1 is a schematic diagram of a maritime communication scenario of an online optimized resource management method for a maritime edge node of the present invention;

2 is a schematic diagram of computing resource allocation of two application programs of the online optimization resource management method of the maritime edge node of the present invention;

3 is a schematic diagram of the allocation of two application program frequencies and time slot resources of the online optimization resource management method of the maritime edge node of the present invention;

Fig. 4 is the average cost comparison of different resource management schemes of the online optimization resource management method of the maritime edge node of the present invention;

FIG. 5 is a comparison of the service quality assurance ratios of different resource management schemes of the online optimization resource management method of the maritime edge node of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

The present invention provides an online optimized resource management method for maritime edge nodes, comprising the following steps:

S1. As shown in Figure 1, a schematic diagram of a maritime communication scenario is given. In the maritime communication network, the underwater layer 1 is equipped with underwater detectors 2, underwater robots 3, underwater unmanned vehicles 4 and other underwater equipment for monitoring, detection and other functions. The edge layer 5 deploys ships 6, buoys 7 and these edge nodes collect data from the underwater equipment near them, and the data center 9, radar 10, base station 11, etc. located in the ground access layer 8, and then send data to these edge nodes. to be processed. Focus on the feedback of data collected by these edge nodes. The transmission period is slotted, and the underwater equipment will generate a series of data packet sequences for different applications within the observation period T. Use the Poisson model to describe the data distribution generated by different applications, that is, the data arrival rate γ _{i of application i} follows the Poisson distribution, and different applications have different QoS requirements;

S2. As shown in FIG. 2, a schematic diagram of computing resource allocation of two application programs is given. Application B is allocated two computing queues for computing and processing data, while application A is allocated one computing queue for computing and processing data. As shown in Figure 3, a schematic diagram of frequency and time slot resource allocation for two applications is given. Application A is allocated more frequency bandwidth than application B. Application A uses 2 time slots for calculation processing, application B uses 1 time slot for calculation processing, application A uses 1 time slot for transmission, and application B uses 2 time slots for transmission. Without limiting data processing and transmission to one time slot, data processing and transmission arriving at time slot t can also be processed and transmitted in time slots after time slot t, and decisions can still be made on a slot-by-slot basis. The total delay of different application i in time slot t is:

和计算延迟

通信延迟

表示在t时隙应用程序i对收集到的信息数据进行通信处理的延迟，是关于传输数据量l_i(t)和分配的带宽B_i(t)的函数，计算延迟

表示在t时隙应用程序i对收集到的信息数据进行计算处理的延迟，是关于P_i(t)的函数，其中，L_i(t)表示在t时隙应用程序i在压缩前新到达的数据量，P_i(t)表示在t时隙应用程序i分配的计算资源，每个时隙中完成的数据流可以是分数，从而不需要将延迟的计算限制在每个时隙中，B_i(t)表示在t时隙分配给应用程序i的带宽，h₀(B_i(t))表示带宽B_i的传输速率，对于K个应用程序在t时隙分别分配的带宽为{B₁(t),B₂(t)···,B_K(t)}；The delay d _i (t) for different application i in time slot t includes the communication delay

and computational delay

communication delay

Represents the delay in communication processing of the collected information data by application i at time slot t, and is a function of the amount of transmitted data l _i (t) and the allocated bandwidth B _i (t), calculate the delay

represents the delay in the computational processing of the collected information data by application i in time slot t, and is a function of P _i (t), where _Li (t) represents the new arrival of application i in time slot t before compression The amount of data, P _i (t) represents the computational resources allocated by application i in time slot t, the data flow completed in each time slot can be fractional, so that there is no need to limit the calculation of the delay to each time slot, B _i (t) represents the bandwidth allocated to application i in time slot t, h ₀ (B _i (t)) represents the transmission rate of bandwidth B _i , and the bandwidth allocated to K applications in time slot t is { B ₁ (t),B ₂ (t)...,B _K (t)};

λ₁,参数δ和β,步长α和u进行初始化，其中，β∈[0,1)是取决于δ的预选常数；S3. Iterate over the original

λ ₁ , parameters δ and β, steps α and u are initialized, where β∈[0,1) is a preselected constant that depends on δ;

基于迭代值y_t通过y_t＝x_t+δu_t得到，其中，u表示随机单位向量，时隙t是从1开始，从步骤S4到步骤S7循环进行，每循环一次时隙t加1；S4. The edge node determines the resource allocation operation set

Based on the iterative value y _t , it is obtained by y _t =x _t +δu _t , where u represents a random unit vector, the time slot t starts from 1, and the cycle proceeds from step S4 to step S7, and the time slot t increases by 1 every cycle;

S5. The edge node collects the cost value f _t (x _t ) and the constraint value g _t (x _t ) at the query point;

更新对偶变量λ_t，其中，μ表示正步长，d表示维度，u表示随机单位向量，[x]⁺＝max{x,0}；S6. The edge node passes the formula

Update the dual variable λ _t , where μ represents the positive step size, d represents the dimension, u represents the random unit vector, [x] ⁺ =max{x,0};

更新在时隙t处的通信资源分配集B_t，通过

更新在时隙t处的计算资源分配集P_t；S7, the edge node passes

Update the communication resource allocation set B _t at time slot t by

update the computational resource allocation set P _{t at time slot t} ;

评估约束满足的程度，Fit_T越小，约束违反越小。where Φ _βX (y)= _{argmin x∈βX} ||xy|| ² is expressed as a projection operator, and βX is a subset of X, using the fit metric

Evaluate the degree of constraint satisfaction, the smaller the Fit _T , the smaller the constraint violation.

The following further proves the beneficial effects of the present invention through corresponding experimental data:

Through simulation experiments, 5 applications with different delay constraints are used, which will be processed and relayed by edge nodes, and the delay constraints are different for different applications. Unless otherwise stated, the delay constraints here are randomly chosen from [0.5,500]ms. The time slot (resource allocation period) unit here is 10 milliseconds, including 10 frames (in practice, one frame usually lasts for 1 millisecond). The observation time is 1000 milliseconds, or 100 resource allocation cycles. If not specified, data arriving by application i follows a Poisson process, where γ _i =1000*i. The available resources of the edge node are B=10000 and P=8000, and the transmitted data and processed data per unit resource are 1 and 10, respectively. Observation time T=1000 ms, step size μ=α=0.05/T. In the experiments, the observation time, resources and capabilities (communication and computation) of edge nodes are just examples, all the above can be adjusted according to different situations.

Figure 4 shows a comparison of the average cost of different resource management schemes. Uniform, random, and proposed scenarios that allocate resources equally to each application flow, respectively. It can be seen that the proposed scheme significantly reduces the average cost compared to the other two schemes. At certain burst points, the proposed scheme can still keep costs low.

Figure 5 shows the comparison of service quality assurance ratios of different resource management schemes. Uniform, random, and proposed scenarios that allocate resources equally to each application flow, respectively. It can be seen that compared with the other two schemes, the proposed scheme has a guarantee probability of QoS close to 1, which can better meet the QoS requirements of each application.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited to this. Any person familiar with the technical field can easily think of various equivalents within the technical scope disclosed by the present invention. Modifications or substitutions should be included within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (1)

1. An online optimized resource management method for maritime edge nodes, comprising the following steps: S1. Use the Poisson model to describe the data distribution generated by different applications, that is, the data arrival rate γ i of different application _i follows Poisson distribution, and different applications have different QoS requirements; S2. Calculate the total delay of application i in time slot t, and the total delay of different application i in time slot t is:

The delay d _i (t) for different application i in time slot t includes the communication delay

and computational delay

communication delay

Represents the delay in communication processing of the collected information data by application i at time slot t, and is a function of the amount of transmitted data l _i (t) and the allocated bandwidth B _i (t), calculate the delay

represents the delay in the computational processing of the collected information data by application i in time slot t, and is a function of P _i (t), where _Li (t) represents the new arrival of application i in time slot t before compression The amount of data, P _i (t) represents the computational resources allocated by application i in time slot t, the data flow completed in each time slot can be fractional, so that there is no need to limit the calculation of the delay to each time slot, B _i (t) represents the bandwidth allocated to application i in time slot t, h ₀ (B _i (t)) represents the transmission rate of bandwidth B _i , and the bandwidth allocated to K applications in time slot t is { B ₁ (t),B ₂ (t)...,B _K (t)}; S3. Iterate over the original

λ ₁ , parameters δ and β, steps α and u are initialized, where β∈[0,1) is a preselected constant that depends on δ; S4. The edge node determines the resource allocation operation set

Based on the iterative value y _t , it is obtained by y _t =x _t +δu _t , where u represents a random unit vector, the time slot t starts from 1, and the cycle proceeds from step S4 to step S7, and the time slot t increases by 1 every cycle; S5. The edge node collects the cost value f _t (x _t ) and the constraint value g _t (x _t ) at the query point; S6. The edge node passes the formula

Update the dual variable λ _t , where μ represents the positive step size, d represents the dimension, [x] ⁺ =max{x,0}; S7, the edge node passes

Update the communication resource allocation set B _t at time slot t by

update the computational resource allocation set P _{t at time slot t} ; where Φ _βX (y)=arg min _x∈βX ||xy|| ² is expressed as a projection operator, and βX is a subset of X, using the fit metric

Evaluate the degree of constraint satisfaction, the smaller the Fit _T , the smaller the constraint violation.

Images