An energy-efficient offloading approach for collaborative mobile edge computing networks with delay constraints

Abstract

The present application relates to an energy-efficient unloading method for a collaborative mobile edge computing network with delay constraints, comprising the following steps: Step 1: Modeling the MEC network scenario, the central server perceives a set of interdependent computationally intensive data generated by IoT devices in the environment; Step 2: Using a directed acyclic graph DAG to characterize the interdependencies between tasks and as the boundary of the decision matrix; Step 3: Complete the setting of constraints, and complete the establishment of the network model, unloading model, and energy consumption model; Step 4: Combined with the obtained DAG graph, according to the optimization algorithm, ensure that the unloading scheme is optimal and the computing delay is minimized. This application can minimize the overall energy consumption and integrate and optimize the MEC network bandwidth utilization and computing resource allocation under the constraints of the specified time limit, thereby improving the practical application significance of the mobile edge computing network.

Description

Technical Field

[0001] The present application belongs to the technical field of electric power Internet of Things, and in particular, relates to an energy-efficient unloading method for a collaborative mobile edge computing network with delay constraints. Background Art

[0002] With the rapid development of technologies such as the Internet of Things (IoT), big data, and cloud computing, many delay-sensitive and computing-intensive application scenarios have emerged, such as autonomous driving and smart healthcare. Traditional IoT devices are limited by their small physical size and lack sufficient computing power and energy supply. Traditional cloud computing (Mobile Cloud Computing, MCC) enables IoT devices to offload tasks to cloud servers and rely on their huge computing power to complete them. However, the offloading process will produce a large delay.

[0003] Mobile Edge Computing (MEC) networks sink computing and storage resources to the edge of IoT device-dense area networks to provide nearby offloading services. However, the computing power of MEC servers is weaker than that of MCC centralized cloud servers. A single MEC server cannot complete a set of intensive computing tasks, and the dependencies between tasks also need to be considered. In addition, the algorithm complexity of the traditional full offloading strategy is relatively large, and it is difficult to obtain the optimal offloading solution based on existing optimization tools. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide an energy-efficient unloading method for a collaborative mobile edge computing network with delay constraints, which reduces the complexity of the algorithm and optimizes the network's latency and energy consumption performance.

[0005] To achieve the above objectives, this application provides the following technical solutions:

[0006] The embodiment of the present application provides an energy efficiency unloading method for a collaborative mobile edge computing network with delay constraints, comprising the following steps:

[0007] Step 1: Model the MEC network scenario in which the central server perceives a set of interdependent computationally intensive data generated by IoT devices in the environment;

[0008] Step 2: Considering the positive non-cyclic characteristics of the dependencies between a set of tasks in time, a directed acyclic graph (DAG) is used to represent the interdependencies between tasks and serve as the boundary of the decision matrix.

[0009] Step 3: According to the computing power, uplink and downlink transmission rates, and computing power allocation of the user end, MEC server end, and cloud end, complete the setting of constraint conditions, and complete the establishment of the network model, offloading model, and energy consumption model;

[0010] Step 4: Combined with the obtained DAG graph, according to the optimization algorithm, under the given energy consumption constraints, the task is offloaded to multiple MEC servers for parallel computing. If necessary, the task will be transferred to the cloud for computing. Finally, the calculation results are sent back to the IoT device to ensure the optimal offloading solution and the minimum computing delay.

[0011] The directed acyclic graph DAG is

[0012] G=(J,ε) (1)

[0013]

[0014] Where J represents the set of tasks generated by an application request, and the ε matrix represents the interdependence between the task sets J, where e ij ∈{0,1},e ij =1 means that the execution of the jth task requires the result of the ith task, e ij = 0 means that the execution of the jth task does not need the result of the ith task, e ij =0(i≥j),

[0015] The data processing frequency of IoT devices is f d , which characterizes the hardware capabilities of IoT devices; the i-th task attribute is initialized as: in Represents the input and output data size, T j ,C i Represent the latency requirement and the number of CPU execution cycles required respectively;

[0016] The total performance of the MEC server is initialized as: K MECS (S U ,S d ,f n ), which represent the uplink and downlink rates and CPU operating frequency that can be freely allocated by the MEC server group; the transmission rate and data processing capacity of the cloud server are represented by S C ,f C To characterize.

[0017] The step 3 is as follows:

[0018] The uninstall decision element is The elements They respectively indicate that the i-th task is executed locally, on the server node, and in the cloud. When the corresponding value is 1, it means that the i-th task is offloaded to the user side, MEC server side, and cloud side for calculation. A single task will only be offloaded once, and the following formula must be satisfied:

[0019]

[0020] For tasks executed locally on the user side, use P L Characterizing the energy consumed by each CPU operation cycle, the energy consumed by executing task i on the user side can be obtained and time taken As shown below:

[0021]

[0022]

[0023] For task i offloaded to MEC server j, that is,

[0024] use Characterize its uplink and downlink transmission rates and CPU operating frequency, and use The energy consumed by receiving and transmitting a data unit can be represented respectively, and the energy consumed by computing task i of MEC server j can be obtained. and computational latency The expression is as follows:

[0025]

[0026]

[0027] Use S C Characterizes the transmission rate of the cloud server. By analogy with the MEC server, the energy consumption can be obtained. and time delay The expression is as follows:

[0028]

[0029]

[0030] in They represent the energy consumed by data upload and download in the corresponding node computing process, Represent the data unit size of the output and output of the i-th task respectively. The uplink and downlink transmission rates and CPU operating frequency are characterized by S CCharacterizes the transmission rate of the cloud server. In order to reduce energy consumption, a single task i is limited to be calculated only at one of the three nodes: the user end, the MEC server end, and the cloud end. The specific offloading scheme is determined by the decision element Determine, and the strict binary variable constraint formula is as follows:

[0031] Considering the interdependence between tasks, the calculation process of a single task may require the calculation results of multiple predecessor tasks. Multiple unloading will affect the overall task logic coherence and cause energy consumption. Therefore, it is stipulated that a single task i can only undergo one unloading process:

[0032]

[0033] For task i, its computation consumes energy E i The total delay T generated by the calculation i They are represented by the following formulas:

[0034]

[0035]

[0036] The step 3 is as follows:

[0037] For IoT devices, when the overall computing energy consumption when offloading to the edge of the network is less than that generated by their own computing, or when they are unable to complete the task computing within a given time due to their own hardware conditions, the offloading process occurs, and ν i Characterizes the CPU operating cycle C required to calculate the task i i and task input data size The ratio of , we get the following formula:

[0038]

[0039] when When , νie is used to represent the complexity rate of task i, as follows:

[0040]

[0041] The IoT device initially considers the data size and required computing cycle of task i, and obtains v by calculation. i >v ie When the task is offloaded to the edge of the network for computing requests.

[0042] Compared with the prior art, the beneficial effects of the present application are: it can minimize the overall energy consumption under the constraints of the specified time limit and integrate and optimize the MEC network bandwidth utilization and computing resource allocation, thereby improving the practical application significance of the mobile edge computing network. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A flow chart of a method according to an embodiment of the present application;

[0045] Figure 2 This is the proposed collaborative MEC network deployment structure diagram;

[0046] Figure 3 It is a DAG representation diagram of the dependencies between tasks;

[0047] Figure 4 Offloading decision flow chart for IoT devices. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0049] The terms "comprises," "comprising," or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0050] See also Figure 1-Figure 4 , the present application provides an energy-efficient unloading method for a collaborative mobile edge computing network with delay constraints, comprising the following steps.

[0051] 1) First, the MEC network framework is modeled. The scenario considered is that an IoT device generates a group of interdependent task groups, which are distributed to M MEC servers for calculation according to the collaborative relationship. If the MEC server lacks the corresponding computing power or cannot complete the task within the constraint time, it needs to be uploaded to the cloud for calculation, and finally the calculation results are sent back to the IoT device to complete the offload request.

[0052] Then the decision offloading process is modeled. The interdependencies between tasks are represented by a directed acyclic graph (DAG), which serves as the boundary condition of the subsequent decision matrix.

[0053] G=(J,ε) (1)

[0054]

[0055] where e ij ∈{0,1},e ij =1(e ij = 0) means that the execution of the jth task requires (does not require) the result of the ith task. Considering that the interdependence between tasks follows the positive non-cyclic characteristics of the time dimension, e ij =0(i≥j).

[0056] The data processing frequency of IoT devices is f d , which characterizes the hardware capabilities of IoT devices; the i-th task attribute is initialized as: in Represents the input and output data size, T j ,C i Represent the latency requirement and the number of CPU execution cycles required respectively; the total performance of the MEC server is initialized as: K MECS (S U ,S d ,f n ), which represent the uplink and downlink rates and CPU operating frequency that can be freely allocated by the MEC server group; the transmission rate and data processing capacity of the cloud server are represented by S C ,f C To characterize.

[0057] 2) The uninstall decision element is When the corresponding value is 1, it means that the i-th task is offloaded to the user side, MEC server side, and cloud side for calculation. A single task will only be offloaded once, and the following formula must be satisfied:

[0058]

[0059] For tasks executed locally on the user side, use P LCharacterizing the energy consumed by each CPU operation cycle, the energy consumed by executing task i on the user side can be obtained and time taken As shown below:

[0060]

[0061]

[0062] For task i offloaded to MEC server j, that is,

[0063] use Characterize its uplink and downlink transmission rates and CPU operating frequency, and use Respectively representing the energy consumed by receiving and transmitting a data unit, the energy consumed by computing task i of MEC server j can be obtained and computational latency The expression is as follows:

[0064]

[0065]

[0066] If MEC server j cannot complete the computing task, it will transfer the i-th task to the cloud for computing, that is,

[0067] Use S C Characterizes the transmission rate of the cloud server. By analogy with the MEC server, we can get the energy consumption and time delay The expression is as follows:

[0068]

[0069]

[0070] In order to reduce energy consumption, a single task i is limited to be calculated at only one of the three nodes: the user end, the MEC server end, and the cloud end. The specific offloading scheme is determined by the decision element. Determine, and the strict binary variable constraint formula is as follows:

[0071]

[0072] Considering the interdependence between tasks, the calculation process of a single task may require the calculation results of multiple predecessor tasks. Multiple unloading will affect the logical coherence of the overall task and cause energy consumption. Therefore, it is stipulated that a single task i can only undergo one unloading process. The following constraints need to be satisfied on the MEC server and cloud side:

[0073]

[0074] For task i, its computation consumes energy E i The total delay T generated by the calculation i They are represented by the following formulas:

[0075]

[0076]

[0077] 3) The unloading process is not random. Appropriate unloading decision conditions are formulated on the IoT device side and the MEC server side.

[0078] In practical terms, on the one hand, it can reduce the computing pressure on cloud servers, and on the other hand, it can effectively optimize and integrate bandwidth resources and improve spectrum utilization.

[0079] For IoT devices, the offloading process occurs when the overall computing energy consumption when offloading to the edge of the network is less than that generated by the computing itself, or when the computing task cannot be completed within a given time due to the limitations of its own hardware conditions. The energy consumption of the offloading process includes the transmission of raw data. and receive computational data from MEC server nodes The energy consumption of the unloading process is only related to the size of the input and output data. When IoT devices are used, they will generate offloading requirements.

[0080] Use ν i Characterizes the CPU operating cycle C required to calculate the task i i and task input data size The ratio is Then we get the following formula:

[0081]

[0082] when When using ν ie Characterize the complexity of task i as follows:

[0083]

[0084] The IoT device initially considers the data size and required computing cycle of task i, and obtains v by calculation i >v ie When the task is offloaded to the edge of the network for computing requests.

[0085] In this example, the task generation and DAG structure matrix analysis, offloading process, and network decision generation in the process of a task offloading request of IoT devices are carried out.

[0086] Modeling at the mathematical level, but the optimization problem after the transformation of traditional 0-1 unloading decision elements is often NP-hard and cannot be solved with the help of corresponding algorithms.

[0087] 4) The present invention relaxes the decision element to make it a continuous variable in the interval [0,1], so the problem is transformed into a convex optimization problem. The convex optimization problem (P) obtained after the relaxation process is as follows:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096] The obtained optimized decision plan is then used as the probability value of the subsequent randomized approximate decision algorithm (randomized rounding technique), and finally the optimal decision plan is obtained. Finally, with the help of convex optimization solvers such as Gurobi and CVX, the optimal offloading decision plan for the task computing request of this IoT device can be obtained, which can minimize the overall energy consumption under the constraint of the specified time limit and integrate and optimize the bandwidth utilization and computing resource allocation of the MEC network, thereby improving the practical application significance of the mobile edge computing network.

[0097] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An energy-efficient offloading approach for collaborative mobile edge computing networks with delay constraints, It is characterized in that The following steps are involved: Step 1: Model the MEC network scenario in which the central server perceives a set of interdependent computationally intensive data generated by IoT devices in the environment; Step 2: Considering the positive non-cyclic characteristics of the dependencies between a set of tasks in time, a directed acyclic graph (DAG) is used to represent the interdependencies between tasks and serve as the boundary of the decision matrix. Step 3: According to the computing power, uplink and downlink transmission rates, and computing power allocation of the user end, MEC server end, and cloud end, complete the setting of constraint conditions, and complete the establishment of the network model, offloading model, and energy consumption model; Step 4: Combined with the obtained DAG graph, according to the optimization algorithm, under the given energy consumption constraint, the task is offloaded to multiple MEC servers for parallel computing. If the MEC server lacks the corresponding computing power or cannot complete the task within the constraint time, it needs to be uploaded to the cloud for computing. Finally, the calculation results are sent back to the IoT device to ensure the optimal offloading solution and the minimum computing delay. The step 3 is as follows: The uninstall decision element is The elements They respectively indicate that the i-th task is executed locally, on the server node, and in the cloud. When the corresponding value is 1, it means that the i-th task is offloaded to the user side, MEC server side, and cloud side for calculation. A single task will only be offloaded once, and the following formula must be satisfied: For tasks executed locally on the user side, use P L Characterizing the energy consumed by each CPU operation cycle, the energy consumed by executing task i on the user side can be obtained and time taken As shown below: For task i offloaded to MEC server j, that is, use Characterize its uplink and downlink transmission rates and CPU operating frequency, and use Respectively representing the energy consumed by receiving and transmitting a data unit, the energy consumed by computing task i of MEC server j can be obtained and computational latency The expression is as follows: Use S C Characterizes the transmission rate of the cloud server. By analogy with the MEC server, the energy consumption can be obtained. and time delay The expression is as follows: in They represent the energy consumed by data upload and download in the corresponding node computing process, Respectively represent the data unit size of the output and output of the i-th task, and use The uplink and downlink transmission rates and CPU operating frequency are characterized by S C Characterizes the transmission rate of the cloud server. In order to reduce energy consumption, a single task i is limited to be calculated only at one of the three nodes: the user end, the MEC server end, and the cloud end. The specific offloading scheme is determined by the decision element Determine, and the strict binary variable constraint formula is as follows: Considering the interdependence between tasks, the calculation process of a single task may require the calculation results of multiple predecessor tasks. Multiple unloading will affect the overall task logic coherence and cause energy consumption. Therefore, it is stipulated that a single task i can only undergo one unloading process: For task i, its computation consumes energy E i The total delay T generated by the calculation i Respectively from the following Formula representation:

2. The energy-efficient offloading method for a collaborative mobile edge computing network with delay constraints according to claim 1, It is characterized in that The directed acyclic graph DAG is G=(J,ε) (1) Where J represents the set of tasks generated by an application request, and the ε matrix represents the interdependence between the task sets J, where e ij ∈{0,1},e ij =1 means that the execution of the jth task requires the result of the ith task, e ij = 0 means that the execution of the jth task does not need the result of the ith task, e ij =0(i≥j), The data processing frequency of IoT devices is f d , which characterizes the hardware capabilities of IoT devices; the i-th task attribute is initialized as: in Represents the input and output data size, T j ,C i Represent the latency requirement and the number of CPU execution cycles required respectively; The total performance of the MEC server is initialized as: K MECS (S U ,S d ,f n ), which represent the uplink and downlink rates and CPU operating frequency that can be freely allocated by the MEC server group; the transmission rate and data processing capacity of the cloud server are represented by S C ,f C To characterize.

3. The energy-efficient offloading method for a collaborative mobile edge computing network with delay constraints according to claim 1, It is characterized in that The step 3 is as follows: For IoT devices, when the overall computing energy consumption when offloading to the edge of the network is less than that generated by their own computing, or when they are unable to complete the task computing within a given time due to their own hardware conditions, the offloading process occurs, and ν i Characterizes the CPU operating cycle C required to calculate the task i i and task input data size The ratio of , we get the following formula: when When using ν ie Characterize the complexity of task i as follows: The IoT device initially considers the data size and required computing cycle of task i, and obtains v by calculation. i >v ie When the task is offloaded to the edge of the network for computing requests.

Images