A dynamic pricing edge collaborative network optimal admission control method

Abstract

The application discloses a kind of dynamic pricing edge collaborative network optimal admission control method, solve the problem of user unloading delay is too high and system load imbalance.Coupled relationship between communication and calculation is coordinated by dynamic pricing, and each user's data of different sizes can be processed within the maximum delay allowed by the user based on dynamic pricing to give service charges.The optimal threshold structure is solved by forward difference method, and the rejected tasks are coordinated by the central controller to enter other optimal servers to achieve load balancing between servers.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of edge computing, and particularly relates to an optimal admission control method for an edge cooperative network. BACKGROUND

[0002] Mobile edge computing (MEC) is increasingly used in computationally intensive and delay-sensitive tasks, such as virtual reality, online gaming, etc. MEC allows users to offload these types of tasks to network edges such as base stations, and quickly process the tasks to return the results to the users. Since the server is close to the user, it has the advantage of low latency, which can significantly improve the efficiency of communication and computation. However, there are some inherent limitations, such as limited resources, and newly arrived tasks may have excessively high latency due to waiting for the completion of previous tasks, which seriously affects the quality of service (QoS) of users.

[0003] The admission control mechanism is a way to control the overall state by limiting the number. The implementation form of this mechanism is usually a threshold structure or pricing. The former directly determines the maximum number of factors considered, and exceeds the maximum number to directly refuse to enter the system; the latter considers the charging mechanism to avoid too many tasks entering the system.

[0004] Latency is one of the important manifestations of user quality of service, and there is a certain latency in the transmission phase and the computing phase. The transmission phase is mainly the task transmission latency, which is related to bandwidth, channel quality, etc.; the computing phase is mainly the computing waiting latency and computing latency, which is mainly related to the number of waiting tasks and the processing speed of the server. The latency-sensitive tasks generated by users have relatively high requirements for latency, and excessively high latency will lead users to be unwilling to offload tasks.

[0005] The existing work uses admission control to ensure the latency of tasks, often ignoring the coupling relationship between communication and computation, only focusing on part of it, and not considering the processing method of rejected tasks. In addition, it does not consider the revenue of service providers, but only focuses on the quality of service of users. SUMMARY

[0006] In order to overcome the shortcomings of the prior art, the application provides a dynamic pricing edge collaborative network optimal admission control method, which solves the problems of high user offloading delay and system load imbalance. The admission control based on dynamic pricing is introduced into the edge collaborative network, which minimizes the cost of users and guarantees the income of service providers under the condition of ensuring fairness. After the user arrives at the calculation queue buffer, the service cost is given based on dynamic pricing, the coupling relationship between communication and calculation is coordinated through dynamic pricing, and the data of different sizes of each user can be processed within the maximum delay allowed. There is a threshold structure for each server, and the optimal threshold structure is solved by the forward difference method. After the tasks in the calculation queue reach the optimal threshold structure, the central controller coordinates the rejected tasks to enter other optimal servers, and realizes the load balancing between the servers. Through the method, the problems of user delay and server load imbalance can be solved at low cost.

[0007] The technical solution adopted by the application to solve its technical problems comprises the following steps:

[0008] Step 1: constructing an edge collaborative computing network;

[0009] Step 2: constructing a user task offloading model, a communication model and a dynamic pricing model;

[0010] Step 3: using Markov chain to model the calculation queue state and constructing a service provider-income model based on the same;

[0011] Step 4: using the forward difference method to solve the optimal threshold of each server;

[0012] Step 5: the central controller coordinates the rejected tasks to join other servers with optimal conditions.

[0013] Further, the step 1 is specifically:

[0014] The edge collaborative computing network is constructed, the system comprises N users and Q base stations, all the base stations belong to one service provider, and the users are directly associated with the designated base stations;

[0015] The user task offloading is a binary offloading model, that is, the task cannot be split and can only be offloaded to the edge server or processed locally; the generation of the i-th user task obeys the Poisson distribution with the parameter λ i , and the arrival of the task at the calculation queue buffer obeys the Poisson distribution with the parameter .

[0016] Further, the step 2 is specifically:

[0017] Step 2-1: the data transmission rate between the user i and the base station j is:

[0018]

[0019] where B i denotes the bandwidth resource allocated to user i by base station j, h j denotes the channel gain between user and base station j, q j is the transmission power between user and base station j, w is the noise power;

[0020] The transmission delay of the task is:

[0021] Step 2-2: The main factor affecting the user benefit is the delay, so according to the various aspects of the delay, the user benefit expression is obtained as:

[0022]

[0023] where V denotes the benefit that the user can obtain after the task is completed; t is the transmission delay, C1 is a delay penalty constant, C1*t represents the communication cost of the user; E[T|n] represents the expected waiting time when there are n tasks in the system; p(n) represents the cost of the computing service that the task needs to pay when there are n tasks in the queue;

[0024] Consider that the processing time of each task is Rewrite equation (2) as:

[0025]

[0026] The final benefit after the task is completed is a non-negative value;

[0027] Step 2-3: Consider the computing queue delay and communication delay, maximize the service provider's benefit under the condition that the final benefit of equation (3) after the task is completed is non-negative:

[0028]

[0029] Further, the step 3 is specifically:

[0030] In order to capture the randomness of the MEC environment, a Markov chain is used to model the queuing state of the task and the server; the Markov chain is used to describe the random process of state transition from one state to another in the state space; the denotes the probability that there are n tasks in the computing queue of server j, according to the Markov chain stationary equation, it is obtained that:

[0031]

[0032] where, In most queuing models with a single server, the service rate μ is required to be greater than the arrival rate λ; since there is admission control in the current model, arriving tasks can be rejected, so the system remains stable when ρ > 1, thus only the cases of ρ = 1 and ρ ≠ 1 are discussed; where k j represents the total number of tasks allowed to join the computing queue by server j.

[0033] The single-server expectation of computing resources is:

[0034]

[0035] The objective function for the service provider is:

[0036]

[0037] where N j represents the total number of users served by server j, B i represents the bandwidth allocated to user i by the service provider, and B represents the total bandwidth possessed by the service provider.

[0038] Further, the step 4 is specifically:

[0039] Step 4-1: Set q = 1 - ρ, substitute the expression of p(n) into the expression (6) of R j , to obtain:

[0040]

[0041] Step 4-2: Use the forward difference method to solve the optimal threshold k * ; k * satisfies the condition:

[0042] R(k * )-R(k * +1) = 0

[0043] When ρ ≠ 1,

[0044]

[0045] Step 4-3: Multiply both sides of equation (9) by , then divide both sides by ρ x *q 2 , to obtain:

[0046]

[0047] Let x' = x + 2, then:

[0048]

[0049] Let Substitute the above equation, we can get

[0050] Multiply both sides by p F(ρ,v)-x′ , we can get the equation:

[0051]

[0052] To solve the above problem, we introduce the Lambert W function to solve:

[0053]

[0054] Substitute F(p, v) back, we get the optimal threshold k * The expression is:

[0055]

[0056] The function has two specific branches, W0 and W -1 Since when p < 1, only W0 branch exists a positive value, when p > 1, only W -1 exists a branch The positive value;

[0057] Step 4-4: R(k) is extended from Z + to Z, and R has two non-trivial extreme values on Z, one extreme value is the optimal threshold k * , which is a positive local maximum, and is the global maximum on Z + , and the other is a non-positive local minimum, and the maximum value of k * is obtained by selecting the branch of Lambert W function;

[0058]

[0059] Since the values of the above two limits are positive, there exists a value k > 0 such that R(k) > R(0) = 0; Since R(1) > R(0) = 0 ((v-m) > 1), it can be proved that there exists a k' < 0, then k' is a local minimum of R on Z;

[0060] When p < 1, the coefficient of Lambert W function is positive, then W0(x) > W1(x);

[0061] When p > 1, the coefficient of Lambert W function is negative, so when i = -1, the value is larger;

[0062] Therefore, when p < 1, i = 0; when p > 1, i = -1;

[0063] then k * The expression is as follows:

[0064]

[0065] Further, the step 5 is specifically as follows:

[0066] A central controller cooperation mechanism is used; the central controller has the following functions: first, when the number of tasks in the computing queue of all servers reaches the threshold value, stop the transmission of the communication queue, as the first layer of admission control; second, provide suggestions for the rejected tasks to enter the buffer of other servers, and the single server as the second layer of admission control;

[0067] The central controller finds a server that meets the following three conditions for the rejected task:

[0068] (1) the number of queued tasks is the least;

[0069] (2) has not reached its threshold value;

[0070] (3) p(n) is non-negative;

[0071] The above conditions are respectively (1) to ensure that the task is processed in the shortest time; (2) to ensure that the task will not be rejected by the current server; (3) the server can obtain a certain benefit if it processes the task, which ensures the income of the service provider; the task is processed within the range of its time delay, so the service quality of the user is also guaranteed;

[0072] If the server cannot find a server that meets the above three conditions, it stops the task transmission of all base stations, and at this time the task generated by the user can only be processed locally.

[0073] The beneficial effects of the present application are as follows:

[0074] 1) The present application coordinates the coupling relationship between communication and calculation by establishing a two-layer admission control mechanism;

[0075] 2) The present application uses dynamic pricing to provide more fair services for users, and realizes the balance between the income of the service provider and the service quality of the user;

[0076] 3) The present application uses a central controller to realize coordinated calculation of the edge network, greatly improves the resource utilization rate of the server in the network, and improves the income of the service provider. BRIEF DESCRIPTION OF DRAWINGS

[0077] Figure 1 is the scene diagram in the embodiment of the present application.

[0078] Figure 2 is the solution diagram in the embodiment of the present application. DETAILED DESCRIPTION

[0079] The application will be further described below in connection with the drawings and examples.

[0080] The application aims at the problems of the prior art and proposes a queue-aware optimal admission control scheme in an edge collaboration network based on dynamic pricing, which is used for guaranteeing the quality of service of task-intensive users and the unbalanced network load.

[0081] The object of the application can be achieved by adopting the following technical scheme:

[0082] A queue-aware optimal admission control scheme in an edge collaboration network based on dynamic pricing comprises the following steps:

[0083] Step one: constructing an edge collaboration computing network;

[0084] Step two: constructing a user task offloading model, a communication model and a dynamic pricing model;

[0085] Step three: using a Markov chain to model the computing queue state and constructing a service provider = revenue model therefrom;

[0086] Step four: using a forward difference method to solve the optimal threshold value of each server;

[0087] Step five: a central controller coordinating other servers with the optimal joining condition of the rejected tasks;

[0088] The step one specifically comprises constructing an edge collaboration network system, which comprises N users and Q base stations, all the base stations belong to one service provider, and the users are directly associated with specific base stations. The user task offloading is a binary offloading model, i.e. the tasks cannot be split and can only be offloaded to an edge server or processed locally. The generation of each user task obeys a Poisson distribution with a parameter λ i , and the arrival of the tasks in the computing queue buffer zone obeys a Poisson distribution with a parameter .

[0089] The step two specifically comprises that the data transmission rate between the user i and the base station j is:

[0090]

[0091] wherein B i represents the bandwidth resource allocated to the user i by the base station j, h j represents the channel gain between the user and the base station j, q j represents the transmission power between the user and the base station j, and w represents the noise power. Then the transmission delay of the task is:

[0092] The main factor that affects the user's benefit is the delay, so according to the various aspects of the delay, the benefit expression of the user is

[0093] V - C1*E[T|n] - p(n) - C1*t≥ 0

[0094] In order to ensure the quality of service of the user, it is necessary to ensure that the final benefit after the completion of the task is non-negative. Wherein V represents the income that the user can obtain after the completion of the task. T is the transmission delay, C1 is the delay penalty constant, C1*t represents the communication cost of the user, E[T|n] represents the expected waiting time when there are n tasks in the system, and p(n) represents the computing service fee that the task needs to pay when there are n tasks in the queue. Since the transmission delay of the user is considered, when the data packet is large, the fee paid in the computing queue will be reduced. The processing time of each task is considered as After rewriting the formula, it is

[0095]

[0096] In order to ensure the user's income, the computing queue delay and the communication delay are considered. In order to stimulate the service provider to provide more services, the service provider's income is maximized under the condition that the above formula is non-negative.

[0097]

[0098]

[0099]

[0100] Step three, in order to capture the randomness of the MEC environment, the queuing state of the task and the server is modeled using Markov chain. Markov chain is used to describe the random process of state transition from one state to another in the state space. Use P(n) represents the probability that there are n tasks in the computing queue of server j, according to the stationary equation of Markov chain, it is obtained that

[0101]

[0102] Wherein, In most queuing models with a single server, the service rate μ must be greater than the arrival rate λ. Since there is admission control in the current model, the arriving tasks may be rejected, so when ρ>1, the system still remains stable, so only the cases of ρ=1 and ρ≠1 are discussed.

[0103] The expected value of the computing resource of the single server is

[0104]

[0105] Then the objective function for the service provider is

[0106]

[0107]

[0108]

[0109] Step four, using forward difference method to solve the optimal threshold k * Set q = 1 - p, substitute the expression of p(n) into the expression of R j , we can get

[0110]

[0111] Using forward difference method, solve the optimal threshold k * , k * satisfy the condition

[0112] R(k * )-R(k * +1) = 0

[0113] When p ≠ 1,

[0114]

[0115] Multiply both sides of the equation by , and then divide both sides by p x *q 2

[0116] After that, we can get

[0117]

[0118] Let x' = x + 2, then

[0119]

[0120] Let Substitute the above equation, we can get

[0121] Multiply both sides by p F(ρ,v)-x′ , and then substitute x = x' - 2, we can get the equation

[0122]

[0123] To solve the above problem, Lambert W function is introduced to solve it.

[0124]

[0125] After the F(ρ,v) is met, the optimal threshold k * The expression of k

[0126]

[0127] The function has two specific branches, W0 and W -1 Since when ρ<1, only W0 branch exists a positive value, when ρ>1, only W -1 branch exists a positive value.

[0128] R(k) is extended from Z + to Z, and R has two non-trivial extreme values on Z. One extreme value is the optimal threshold k * , which is a positive local maximum, and is the global maximum on Z + , and the other is a non-positive local minimum. By selecting the branch of Lambert W function, the maximum of k * is obtained.

[0129]

[0130] Since the values of the above two limits are positive, there exists a value k>0 such that R(k)>R(0)=0. Since R(1)>R(0)=0 ((v-m)>1), it can be proved that there exists a k'<0, then k' is a local minimum of R on Z. When ρ<1, the coefficient of Lambert W function is positive, then W0(x)>W1(x). When ρ>1, the coefficient of Lambert W function is negative, so when i=-1, the value is larger. Therefore, when ρ<1, i=0; when ρ>1, i=-1.

[0131] The expression of k * is as follows:

[0132]

[0133] The step five, a service provider usually has multiple base stations, in order to better use the computing resources of each base station server, avoid resource utilization imbalance, through the use of central controller cooperation mechanism. The role of the central controller is: first, when the number of tasks in all server computing queues reaches the threshold value, stop the transmission of the communication queue, as the first layer of access control; second, provide suggestions for the rejected tasks to enter the buffer area of other servers, and the single server as the second layer of access control.

[0134] ​The central controller finds a server for the rejected task that satisfies the following three conditions simultaneously.

[0135] (1) The number of queued tasks is the least.

[0136] (2) It has not reached its threshold.

[0137] (3) p(n) is non-negative

[0138] The above conditions are respectively (1) to ensure that the task is processed in the shortest time; (2) to ensure that the task will not be rejected by the current server; (3) to ensure that the server can obtain a certain benefit if it processes the task, thereby ensuring the benefit of the service provider. The task is processed within the allowed range of its time delay, so the quality of service of the user is also guaranteed.

[0139] If the server cannot find a server that satisfies the above three conditions, it stops the task transmission of all base stations, and at this time the tasks generated by the user can only be processed locally.

Claims

1. A method for dynamic pricing edge-cooperative network optimal admission control, characterized in that, Comprising the following steps: Step 1: constructing an edge collaborative computing network; Step 2: constructing a user task offloading model, a communication model, and a dynamic pricing model; Step 3: modeling the computing queue state using a Markov chain and constructing a service provider-revenue model therefrom; Step 4: solving the optimal threshold value of each server using a forward difference method; Step 5: a central controller coordinating the joining of rejected tasks to other servers that are optimal in terms of conditions; The step 2 is specifically: Step 2-1: the data transmission rate between user i and base station j is: where B i denotes the bandwidth resource allocated by the base station j to the user i, h j denotes the channel gain between the user and the base station j, q j is the transmission power between the user and the base station j, and w is the noise power; The transmission delay of the task is: Step 2-2: the main factor affecting the user benefit is the delay, so according to various aspects of the delay, the benefit expression of the user is: V-C1*E[T|n]-p(n)-C1*t>0 (2) Where V represents the benefit that the user can obtain after completing the task; t is the transmission delay, C1 is a delay penalty constant, C1*t represents the communication cost of the user; E[T|n] represents the expected waiting time when there are n tasks in the system; p(n) represents the computing service fee that the task needs to pay when there are n tasks in the queue; Consider the processing time for each task to be Rewriting equation (2) gives The final benefit after completing the task is a non-negative value; Step 2-3: considering the computing queue delay and the communication delay, the revenue of the service provider is maximized under the condition that the final benefit of the task after completing the task is non-negative according to formula (3):

2. The method of claim 1, wherein, The step 1 is specifically: An edge collaborative computing network is constructed, the system includes N users and Q base stations, all base stations are owned by a service provider, and the user is directly associated with a designated base station; User task offloading is a binary offloading model, i.e., tasks are not split and can only be offloaded to edge servers or handled locally; the generation of the ith user task obeys a Poisson distribution with parameter i By the additivity of the Poisson distribution, it is known that the generation of tasks arriving at the computing queue buffer obeys a Poisson distribution with parameter .

3. The method of claim 1, wherein, The step 3 is specifically: To capture the randomness of the MEC environment, Markov chains are used to model the queuing state of tasks and servers; Markov chains are used to describe a stochastic process that transitions from one state to another in a state space; using denotes the probability that server j has n tasks in its computation queue, according to the Markov chain stationary equation, it is obtained that: wherein, In most queuing models with a single server, the service rate μ is required to be greater than the arrival rate λ; since there is admission control in the current model, arriving tasks can be rejected, so the system remains stable when ρ > 1, so only the cases of ρ = 1 and ρ ≠ 1 are discussed; wherein k j represents the total number of tasks allowed to join the computing queue by server j; The expected value of the computing resource of a single server is: Then the objective function of the service provider is: where N j denotes the total number of users served by server j, B i denotes the bandwidth allocated to user i by the service provider, B denotes the total bandwidth owned by the service provider.

4. The method of claim 1, wherein, The step 4 is specifically: Step 4-1: Set q = 1 - p, and substitute the expression of p(n) into the expression (6) of R j to obtain: Step 4-2: Solve the optimal threshold k using the forward difference method * ; k * The condition to be met is: R(k * )-R(k * +1)=0 When ρ≠1, Step 4-3: multiply both sides of equation (9) by and divide both sides by p x* q 2 and we have Let x'=x+2, then: Let Substituting the above equation, we have Both sides of the same ride ρ F(ρ,v)-x′ Again, x = x '-2 back, you can get the equation: To solve the above problem, the Lambert W function is introduced to solve: After substituting F(ρ, v) back, the optimal threshold k is obtained * The expression is: The function has two specific branches, W0 and W -1 Since when p < 1, only the W0 branch exists a is positive, when p > 1, only the W -1 branch exists is positive; Step 4-4: R(k) is extended to Z + and R has two non-trivial extrema on Z, one, k * , is a positive local maximum, which is a global maximum on Z + , and the other is a non-positive local minimum, which gives the maximum of k * by choosing the branch of the Lambert W function; Since the values of the above two limits are positive, there exists a value k>0 such that R(k)>R(0)=0; Since R(1)>R(0)=0 ((v-m)>1), it can be proved that there exists a k'<0, then k' is a local minimum of R on Z; When p < 1, the coefficient of the Lambert W function is positive, then W0(x) > W1(x); When ρ>1, the coefficient of the Lambert W function is negative, so when i=-1, the value is larger; Therefore, when ρ<1, i=0; when ρ>1, i=-1; then k * The expression for k is as follows:

5. The method of claim 1, wherein, The step 5 is specifically as follows: A central controller cooperation mechanism is used; the functions of the central controller are: first, when the number of tasks in the computing queue of all servers reaches the threshold value, the transmission of the communication queue is stopped, serving as the first layer of access control; Second, provide suggestions for the rejected tasks to enter the buffer of other servers, and a single server serves as the second layer of access control; The central controller finds a server that meets the following three conditions for the rejected task: (1) the number of queued tasks is the least; (2) it has not reached its threshold value; (3) p(n) is non-negative; The above conditions respectively ensure that the task is processed in the shortest time (1), that the task is not rejected by the current server (2), and that the server can obtain a certain benefit if it processes the task (3), thereby ensuring the revenue of the service provider; the task is processed within the allowed range of its delay, so the service quality of the user is also guaranteed; If the server cannot find a server that meets the above three conditions, stop all base station task transmission, at this time the user generated task can only be handled locally.

Images