A resource scheduling method fusing security decision and computing acceleration

Abstract

The application discloses a kind of resource scheduling methods of fusing security decision and computing acceleration, method includes the following steps: step 1: the processing time and energy consumption of local task are calculated;Step 2: the delay and energy consumption of communication process are calculated;Step 3: the user QoE demand parameter is calculated;Step 4: determine the security decision of user.Step 5: resource scheduling strategy optimization solution.The application comprehensively considers resource allocation, task offloading, user QoE demand and data security, strengthens the data security of task offloading process, and uses computing acceleration as the index of quantitative QoE demand, so that the method is more in line with the real experience process of user, meets the increasingly sensitive experience quality requirement of user, and then seeks the trade-off between delay, energy consumption, QoE and data security.

Description

TECHNICAL FIELD

[0001] The present application relates to a resource scheduling method fusing security decision and computing acceleration, and belongs to the field of Internet and edge computing. BACKGROUND

[0002] In recent years, under the support of mobile Internet, especially 5G technology, the Internet of Things (IoT) has developed rapidly and has been widely used in various aspects of production and life, such as virtual reality / augmented reality, mobile medical treatment, etc. These business scenarios not only require very low and deterministic network delay, but also require massive, heterogeneous and diverse data access. The centralized computing processing mode of traditional cloud computing faces huge computing and network pressure and cannot meet the application requirements of the Internet of Everything. In this context, the edge computing mode has become an effective solution. In the edge computing mode, in order to better support high-density, large-bandwidth and low-latency business scenarios, a more effective way is to build a business platform (edge data center) near the network edge, provide storage, computing, network and other resources, and sink part of the key business applications to the access network edge to reduce the bandwidth and latency loss caused by network transmission and multi-level forwarding.

[0003] In edge computing, not only are there static terminal devices (such as sensors in smart homes, cameras in public places), but also dynamic terminal devices (such as drones and vehicles, etc.), which make resource management more challenging. And appropriate resource scheduling can alleviate this situation. If resource scheduling cannot effectively utilize scattered resources, it will result in too low utilization of actually available resources, and thus problems such as high delay and high energy consumption. And these problems are unacceptable for new applications that are computationally intensive and delay sensitive. And if an efficient resource scheduling strategy is applied, resources can be effectively combined together to establish a pool of available and economically efficient computing resources, thus solving the resource management problem.

[0004] Therefore, considering the importance of resource scheduling, some works have begun to optimize the weighted sum of latency and energy consumption of edge computing systems, and study the trade-off relationship between the two. However, in actual scenarios, in addition to latency and energy consumption, the resource scheduling process also needs to focus on factors such as user quality of experience (QoE) and data security. QoE requirement is an index that can reflect the real experience of users. More and more users begin to pay attention to their subjective feelings, and their QoE requirements are sensitive, and the requirement of low latency cannot completely cover the QoE requirements of users, so this encourages the present application to focus on the QoE index in the process of resource scheduling.

[0005] In addition, QoE requirements are closely related to data security, attackers can easily insert interference factors into unencrypted task data to achieve the purpose of destroying user experience, and wireless interference between MUs will also unintentionally reduce user QoE indicators due to the shared wireless channel, and encryption of task data can ensure data integrity. Therefore, it is necessary to consider data security while considering QoE. At present, although there have been some works focusing on the four factors of delay, energy consumption, QoE and data security, most of the researches are jointly analyzing two or three factors and they do not fully focus on the close coupling between QoE and data security, and the related researches do not well combine the four factors to analyze. The trade-off problem between the four factors of delay, energy consumption, QoE requirements and data security needs to be solved. SUMMARY

[0006] In order to solve the problems and deficiencies in the prior art, the present application proposes a resource scheduling method combining security decision and computing acceleration, which comprehensively considers factors such as resource allocation, computing offloading, data security and QoE, aims to minimize the cost of the whole system, and seeks the trade-off between delay, energy consumption, QoE and data security.

[0007] In order to achieve the above-mentioned purpose, the technical scheme of the present application is as follows: a resource scheduling method combining security decision and computing acceleration, comprising the following steps:

[0008] Step 1: Calculate the processing time and energy consumption of local tasks. Obtain the CPU core number n of different mobile users MU i i and the processing capacity f of each core i l , then calculate the power of each core , then obtain the calculation time of the serialized part based on Amdahl's law and the calculation time of the parallel part , so as to obtain the total local calculation time and the total local calculation energy consumption, that is , wherein T i represents the task generated by each MU i, c i represents the processing density of T i , D i represents the task data size generated by MU i, a i represents the parallel fraction of T i , and λ i represents the offloading decision variable of the task Ti generated by MU i, and k1 is a coefficient reflecting the relationship between the processing capacity and power consumption of the user mobile device.

[0009] Step 2: Calculate the delay and energy consumption of the communication process. A wireless transmission model based on Shannon's formula is used, and the data transmission rate r is characterized by channel power gain and channel noise power density. i Among them, let P i It is the transmission power of the mobile user equipment (MUi), h i This represents the channel power gain, ω0 represents the channel noise power density, and B represents the wireless bandwidth. Next, the transmission delay is calculated. and transmission energy consumption Then calculate the edge processing time. and edge processing energy consumption Where m i f represents the number of cores allocated by the edge server for processing tasks generated by different MU i. i e This represents the processing power (CPU frequency in cycles per second) allocated to each core on the edge server of MU i, and k2 is a coefficient reflecting the relationship between edge-side processing power and power consumption. Finally, the transmission latency is given in the returned results. and corresponding energy consumption Where γ i This indicates the size of the returned result relative to T. i The ratio of the initial task data size to the edge server's transmission power.

[0010] Step 3: Calculate user QoE requirement parameters. First, the mobile user sends a small reconnaissance packet to the base station. Then, the base station returns the result information. After these two operations, the number of CPU cores in the mobile user device and the number of CPU cores in the edge server are obtained. Then, based on Amdahl's Law, the local computing acceleration is calculated separately. and edge computing acceleration value Where α i T represents i Parallelizable fractions, m i n represents the number of cores allocated by the edge server for processing tasks generated by different MU i. i This indicates the number of CPU cores for different mobile users MU i.

[0011] Step 4: Determine the user's security decisions. First, set S... i ∈{0,1} represents each mobile user MU i The binary security decision is made by each MU based on the privacy requirements of the application data. Wherein S... i =0 indicates that the mobile terminal device MU i The computational tasks will be unencrypted; and S i=1 indicates that the mobile terminal device MU i The computation task and its data will be encrypted using ChaCha20 encryption technology before being transmitted to the edge server. After receiving the task and data, the edge server will further decrypt the data, execute the computation task, and send the processed result back to the MU. i .

[0012] Step 5: Solve the resource scheduling strategy. Specifically, based on the first four steps, calculate the total cost for the MU to execute all tasks locally. And the total overhead of performing all tasks on the edge server. Next, weighted parameters w1 and w2 are set for execution time and energy consumption, where w1, w2 ∈ [0, 1]. w1 + w2 = 1. The weighted values ​​are adjusted accordingly when the objective changes. Then, the problem is defined: the objective is to minimize the system cost. Then, the branch and bound method is used to solve the optimal solution of the above objective equation. The optimal solution is the unloading decision set O(λ). I )={λ1,...,λ N The result is represented in the form of}. Finally, after multiple rounds of testing, the sums are averaged to obtain the final result.

[0013] Compared with the prior art, the advantages of the present invention are as follows:

[0014] 1) For the multi-user resource scheduling problem in edge computing systems, a mixed-integer linear programming method is adopted to comprehensively consider factors such as resource allocation, computation offloading, security and QoE. The solution is obtained by integrating a resource scheduling method that combines security decision-making and computation acceleration, which can minimize the implementation cost of the system.

[0015] 2) Introducing ChaCha20 encryption technology as a security decision can prevent task data from being maliciously stolen by attackers during the unloading process.

[0016] 3) Using compute acceleration as a metric to quantify QoE requirements can more accurately reflect users' service experience needs. Furthermore, this metric can be well coupled with security decisions, enabling resource scheduling methods that integrate compute acceleration and security decisions to achieve a good balance between latency, energy consumption, QoE requirements, and data security. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of an embodiment of the present invention.

[0018] Figure 2 This is an architecture diagram of an embodiment of the present invention.

[0019] Figure 3A computational graph for implementing an optimization objective Obj of the present application. DETAILED DESCRIPTION

[0020] In order to deepen the understanding and understanding of the present application, the following specific embodiments, further illustrate the present application.

[0021] Example 1: a resource scheduling method fusing security decision and computing acceleration, the method first task whether to be handled locally; then, see Figure 2 , the three-layer model is used as the overall architecture, then the multi-user resource scheduling problem is abstracted into a mixed integer linear programming problem, according to the branch and bound method to solve the defined optimization problem Obj, the process of calculating Obj see Figure 3 , finally combined with the constraint conditions set to obtain the optimal solution of the optimization problem. The detailed implementation steps are as follows:

[0022] Step 1: calculate the processing time and energy consumption of local task, obtain the CPU core number n i and the processing capacity f i l of each core, then calculate the power of each core , then based on Amdahl's law to obtain the calculation time of the serialized part and the calculation time of the parallel part , so as to obtain the total local calculation time and the total local calculation energy consumption, that is , wherein T i represents the task generated by each MU i, c i represents the processing density of T i , D i represents the task data size generated by MU i, a i represents the parallel fraction of T i , and l i represents the offload decision variable of MU i generated task Ti, k1 is the coefficient reflecting the relationship between the processing capacity and power consumption of the user mobile device.

[0023] Step 2: calculate the delay and energy consumption of the communication process, based on the wireless transmission model of Shannon formula, and according to the channel power gain and channel noise power density to represent the data transmission rate r i ; then calculate the transmission delay and transmission energy consumption. Then calculate the edge processing time and edge processing energy consumption; finally give the transmission delay and the corresponding energy consumption returned by the result. The specific implementation is as follows:

[0024] Sub-step 2-1: Data transmission process construction, the data transmission process is usually constructed based on Shannon formula, the data transmission rate needs wireless bandwidth parameter B, transmission power P of mobile user equipment MU i i and channel power gain h i and channel noise power density ω0, through the above parameters, the data transmission rate can be expressed as:

[0025]

[0026] Formula (1) represents the transmission rate of data when the data is unloaded from the mobile user equipment to the edge in the specified wireless bandwidth B.

[0027] Sub-step 2-2: Task T generated by MU i i Unloading transmission delay: based on the above analysis, λ i D i The transmission delay of D-bit data unloaded to the edge can be obtained by the following way:

[0028]

[0029] Where λ i represents the unloading decision variable of the task Ti generated by the MU i, D i represents the data size of the task generated by the MU i.

[0030] Transmission energy consumption of task unloading process: the energy consumption of the terminal equipment of the λ i D i bit data of transmission unloading is expressed as:

[0031]

[0032] Where λ i represents the unloading decision variable of the task T i generated by the MU i, D i represents the data size of the task generated by the MU i, P i represents the transmission power of the mobile user equipment MU i.

[0033] Sub-step 2-3: Calculation time required for edge server to process task: λ i D i After D-bit data is unloaded to the edge, the edge will process the data, let m i represents the number of kernels allocated by the edge server for processing the tasks generated by different MU i, f i e represents the processing capacity (CPU frequency in cycles / second) of each kernel of the edge server allocated to the MU i,

[0034] f i e >f i l (16)

[0035] The power consumption of each core processing data can be expressed as:

[0036]

[0037] where k2 is a coefficient reflecting the relationship between the edge side processing capacity and power consumption, λ i D i The calculation time of the bit data includes the calculation time of the serialized part and the calculation time of the parallel part, and the calculation time of the serialized part is expressed as:

[0038]

[0039] c i The processing density of T i D i The size of the task data generated by MU i, α i The parallel fraction of T i , λ i The offloading decision variable of the task T i generated by MU i.

[0040] The calculation time of the parallel part can be expressed as:

[0041]

[0042] m i The number of kernels allocated by the edge server for processing tasks generated by different MU i, f i e The processing capacity (CPU frequency in cycles / second) of each kernel of the edge server allocated to MU i. The calculation time required by the edge server to process the task can be calculated as:

[0043]

[0044] The calculation time of the serialized part and The calculation time of the parallel part.

[0045] Substep 2-4: Energy consumption of the edge server processing the task: calculation data λ i D i The edge energy consumption formula of the bit data is as follows:

[0046]

[0047] where k2 is a coefficient reflecting the relationship between edge side processing capability and power consumption, c i represents the processing density of T i , α i represents the parallelism fraction of T i , λ i represents the offloading decision variable of the task T i generated by MU i. f i e represents the processing capability of each core of the edge server assigned to MU i.

[0048] Sub-step 2-5: Result return: After the processing of task T i is completed, the result will be returned to the mobile terminal device, denoted as r i ' represents the data transmission rate during the result return process, P0 is the transmission power of the edge server, and the data transmission rate needs the wireless bandwidth parameter B, the channel power gain h i , and the channel noise power density ω0. Similar to the data transmission rate of offloading, r i ' can also be expressed as:

[0049]

[0050] Transmission delay of result return: Based on the above analysis, γ i D i the transmission delay of the r i -bit result return can be expressed as:

[0051]

[0052] where γ i represents the ratio of the data size of the returned result to the initial task data size of T i i Therefore, the edge server energy consumption of transmitting the γ i D i -bit processing result to the mobile terminal device is expressed as:

[0054]

[0055] Many research works ignore the return process because the processing result value is very small. Unlike these works, this patent pays more attention to the result return process, and is more concerned about the information carried by the result return value, which can better reflect the real wireless communication process.

[0056] Step 3: Calculate the user QoE demand parameters, first the mobile user sends a small amount of data to the base station, and then the base station returns the result information, after the above two operations, the CPU core number of the mobile user equipment and the CPU core number of the edge server are obtained, and then the local computing acceleration and edge computing acceleration values are calculated according to Amdahl's law where α i represents the parallel fraction of T i , m i represents the number of cores allocated by the edge server for different MU i task processing, and n i represents the CPU core number of different mobile users MU i.

[0057] Step 4: Determine the user's security decision, first S i ∈{0,1} represents the binary security decision for each mobile user MUi, which is personalized by each MU according to the privacy requirements of the application data. Where S i = 0 means that the mobile terminal device MU i will unload the computing task in an unencrypted manner; and S i = 1 means that the mobile terminal device MU i will use ChaCha20 encryption technology to encrypt the computing task and its data before transmission to the edge server. After receiving the task and data, the edge server further decrypts the data, then executes the computing task and sends the processed result back to MU i.

[0058] Step 5: Resource scheduling strategy optimization solution, this embodiment uses a PC equipped with AMD Ryzen 5 4600H CPU, 3.0GHz frequency and 16GB RAM capacity, and the edge computing system has 60 MUs, and runs the EdgeCloudSim system for multiple rounds of testing and weighted average. More specifically, according to the first four steps, the total overhead of MU executing all tasks locally and the total overhead of executing all tasks on the edge server are solved. Then set the weighted parameters w1 and w2 of execution time and energy consumption, where w1, w2 ∈ [0,1]. And let w1+w2=1. When the target is different, the weight value set changes accordingly. Then establish the problem definition, the goal is to minimize the cost of the system After that, the branch and bound method is used to solve the optimal solution of the above objective equation, and the optimal solution is represented in the form of the unloading decision set O(λ I )={λ1,...,λ N}. Finally, through multiple rounds of testing, the sum is taken as the average value as the final return result.

[0059] Based on the same inventive concept, the application also discloses a resource scheduling method combining safety decision and calculation acceleration, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor.

[0060] Those skilled in the art will appreciate that the embodiments described herein are for the purpose of helping the reader understand the principles of the application and it should be understood that the embodiments are only used to illustrate the application and not to limit the scope of the application, and after reading the application, those skilled in the art can make various equivalent modifications to the application, which all fall within the scope defined by the claims of the application.

Claims

1. A resource scheduling method integrating security decision-making and computational acceleration, characterized in that, The method includes the following steps: Step 1: Calculate the processing time and energy consumption of the local task. Step 2: Calculate the latency and energy consumption of the communication process. Step 3: Calculate user QoE requirement parameters. Step 4: Determine the user's security decision. Step 5: Solve the resource scheduling strategy optimization; Step 3 involves calculating the user's QoE requirements. First, the mobile user sends a small reconnaissance packet to the base station. Then, the base station returns the result. After these two operations, the number of CPU cores in the mobile user's device and the number of CPU cores in the edge server are obtained. Finally, based on Amdahl's Law, the local computing acceleration is calculated. and edge computing acceleration value ,in T represents i Parallelizable fractions, m i This represents the number of cores allocated by the edge server for processing tasks generated by different MU i. This indicates the number of CPU cores for different mobile users MU i; Step 4: Determine the user's security decision, first of all... Represented as MU for each mobile user The binary security decision is made by each MU based on the privacy requirements of the application data. Indicates mobile terminal device MU The computing tasks will be unencrypted; Indicates mobile terminal device MU The computation task and its data will be encrypted using ChaCha20 encryption technology before being transmitted to the edge server. After receiving the task and data, the edge server will further decrypt the data, execute the computation task, and send the processed result back to MU. ; Step 5: Solve the resource scheduling strategy. Based on the previous four steps, calculate the total cost for MU to execute all tasks locally. And the total overhead of performing all tasks on the edge server. Next, set the weighted parameters for execution time and energy consumption. and ,in and order When the objectives differ, the weight values ​​are set accordingly. Then, the problem definition is established, with the objective being to minimize the system cost. Then, the branch and bound method is used to find the optimal solution to the objective equation, and the optimal solution is obtained by unloading the decision set. The result is expressed in the form of , and after multiple rounds of testing, the sums are averaged to obtain the final result.

2. The resource scheduling method integrating security decision-making and computational acceleration according to claim 1, characterized in that, Step 1: Calculate the processing time and energy consumption of local tasks to obtain the number of CPU cores for different mobile user MUi. and the processing power of each core Then calculate the power of each core. The computation time for the serialization part is then obtained based on Amdahl's law. Computation time of the parallelization part This is used to derive the total local computation time. And the total local computing power consumption, i.e. T i This represents the task generated by each MUi. T represents i The processing density, This indicates the size of the task data generated by MU i. T represents i Parallelizable fractions, Let represent the unloading decision variable for task Ti generated by MU i, and k1 is a coefficient reflecting the relationship between the processing power and power consumption of the user's mobile device.

3. The resource scheduling method integrating security decision-making and computational acceleration according to claim 2, characterized in that, Step 2: Calculate the latency and energy consumption of the communication process, based on the Shannon formula wireless transmission model, and characterize the data transmission rate according to channel power gain and channel noise power density. Among them, let P i It is the transmission power of the mobile user equipment (MUi), h i The channel power gain is given by ω0, where ω0 represents the channel noise power density and B represents the wireless bandwidth. The transmission delay is then calculated. and transmission energy consumption Then calculate the edge processing time. and edge processing energy consumption , where m i f represents the number of cores allocated by the edge server for processing tasks generated by different MU i. i e This represents the processing power allocated to each core of the edge server in MU i, the CPU frequency in cycles per second, k2 is a coefficient reflecting the relationship between edge-side processing power and power consumption, and finally, the transmission latency is returned as the result. and corresponding energy consumption ,in This indicates the size of the returned result relative to T. i The ratio of the initial task data size to the edge server's transmission power.

4. The resource scheduling method integrating security decision-making and computational acceleration according to claim 3, characterized in that, Step 2: Calculate the latency and energy consumption of the communication process, as follows: Its implementation consists of the following sub-steps: Sub-step 2-1: Data transmission process construction. The data transmission process is constructed based on Shannon's formula, and the data transmission rate requires wireless bandwidth parameters. Mobile User Equipment (MU) Transmission power and channel power gain and channel noise power density The data transmission rate is expressed as: Formula (1) represents the situation when data is within a specified wireless bandwidth. The data transmission rate when data is offloaded from mobile user equipment to the edge is represented by... Sub-step 2-2: MU Generated tasks The offloading transmission delay will The transmission delay of bit data offloading to the edge is obtained in the following way: in This represents the unloading decision variable for task Ti generated by MU i. This indicates the size of the task data generated by MU i. Transmission energy consumption during task unloading: transmission unloading The energy consumption of a terminal device that handles bit data is expressed as follows: in Indicates the task T generated by MU i i The unloading decision variable, This indicates the size of the task data generated by MU i. This indicates the transmission power of the mobile user equipment MU i. Sub-steps 2-3: Computation time required for the edge server to process the task: After the bit data is unloaded to the edge, the edge will process the data, assuming... This is represented as an edge server for different MUs. The number of cores allocated for processing the generated tasks. Indicates allocation to MU The processing power of each core on the edge server, measured in cycles per second (CPU frequency). The power consumption of each core for edge data processing is expressed as follows: in It is a coefficient reflecting the relationship between edge-side processing capability and power consumption, and is unloaded. The computation time for bit data includes the computation time of the serialization part and the computation time of the parallelizable part. The computation time of the serialization part is expressed as: T represents i The processing density, This indicates the size of the task data generated by MU i. T represents i Parallelizable fractions, Indicates the task T generated by MU i i The unloading decision variable, The computation time for the parallel portion is expressed as: m i This represents the number of cores allocated by the edge server for processing tasks generated by different MU i. This represents the processing power allocated to each core of the edge server in MUi, expressed as CPU frequency in cycles per second. The computation time required for the edge server to process a task is calculated as follows: Indicates the computation time of the serialization part This represents the computation time of the parallelizable portion. Sub-steps 2-4: Energy consumption of edge servers in processing tasks: computational data The edge energy consumption formula for a given bit is as follows: in It is a coefficient that reflects the relationship between edge-side processing capability and power consumption. T represents i The processing density, T represents i Parallelizable fractions, Indicates the task T generated by MU i i The unloading decision variable, This represents the processing power allocated to each core of the edge server in MU i. Sub-steps 2-5: Result return: Task After processing, the result will be returned to the mobile terminal device. This indicates the data transfer rate during the result return process. This refers to the transmission power of the edge server; the data transmission rate requires wireless bandwidth parameters. Channel power gain and channel noise power density Similar to the data transfer rate of unloading, The formal expression is as follows: The transmission delay returned in the result. The sending delay returned by the bit result is represented as follows: in This indicates the size of the returned result relative to T. i The ratio of the initial task data size, Therefore, the processing results The energy consumption of the edge server for transmitting bits to mobile terminal devices is expressed as: in This refers to the transmission power of the edge server.

Images