CN112783662B - CPU resource sharing system in sensor edge cloud task unloading of integrated block chain - Google Patents
CPU resource sharing system in sensor edge cloud task unloading of integrated block chain Download PDFInfo
- Publication number
- CN112783662B CN112783662B CN202110186724.XA CN202110186724A CN112783662B CN 112783662 B CN112783662 B CN 112783662B CN 202110186724 A CN202110186724 A CN 202110186724A CN 112783662 B CN112783662 B CN 112783662B
- Authority
- CN
- China
- Prior art keywords
- edge
- task
- cpu
- node
- cluster
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004364 calculation method Methods 0.000 claims abstract description 86
- 241000854291 Dianthus carthusianorum Species 0.000 claims abstract description 84
- 238000000034 method Methods 0.000 claims abstract description 42
- 230000009471 action Effects 0.000 claims abstract description 35
- 230000002787 reinforcement Effects 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 37
- 230000006870 function Effects 0.000 claims description 31
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 230000015654 memory Effects 0.000 claims description 13
- 230000006403 short-term memory Effects 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 230000007787 long-term memory Effects 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- 230000001953 sensory effect Effects 0.000 claims description 5
- 238000012546 transfer Methods 0.000 claims description 4
- 230000000306 recurrent effect Effects 0.000 claims 3
- 230000008569 process Effects 0.000 abstract description 18
- 230000006855 networking Effects 0.000 abstract description 16
- 238000005265 energy consumption Methods 0.000 abstract description 14
- 230000007246 mechanism Effects 0.000 description 17
- 238000012545 processing Methods 0.000 description 13
- 230000003993 interaction Effects 0.000 description 9
- 238000012795 verification Methods 0.000 description 7
- 230000008901 benefit Effects 0.000 description 6
- 238000013461 design Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 125000004122 cyclic group Chemical group 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004540 process dynamic Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a CPU resource trusted sharing system in sensor edge cloud task unloading of an integrated block chain, which comprises the following components: the edge node offers the intelligent contract for the cluster head node to achieve the calculation task unloading service and provides the calculation task unloading service; the cluster head node is used for quoting and collecting bidding actions of the sensing equipment and CPU credibility voting rewards fed back by the sensing equipment according to CPU resource unit price provided by the edge node, and selecting credible edge nodes by adopting a reinforcement learning algorithm to achieve intelligent contract and unload calculation tasks; the edge node and the cluster head node have blockchains for storing the smart contracts. The invention provides a novel sensing edge cloud core networking architecture for saving the energy consumption of sensing equipment. In the architecture, the blockchain is designed to realize the transaction process of reliably sharing the CPU resource of the edge node when the computing task of the sensor equipment cluster is unloaded.
Description
Technical Field
The invention belongs to the technical field of the Internet of things, and particularly relates to a CPU resource trusted sharing system in sensor edge cloud task unloading of an integrated block chain.
Background
The CPU of the sensing equipment processes the calculation task and the security protection consumes a large amount of energy resources, in order to save the energy consumption of the sensing equipment, the novel sensing edge cloud architecture offloads the calculation task to the edge node for processing, the CPU resources of the edge node are utilized for processing the calculation task with high energy consumption, and the security verification algorithm with low energy consumption can be processed by the CPU of the sensing equipment. The edge node can directly control the application of the service layer through the wireless access sensing equipment, and on the other hand, the calculation processing and the security verification in the CPU of the sensing equipment are separated, namely the sensing equipment does not need to process calculation tasks, the tasks to be processed are directly unloaded to the edge node for execution, and the CPU of the sensing equipment only collects data and performs the security verification. This greatly reduces the energy consumption of the sensing edge cloud system and improves the reliability and security of the computing task offloading. However, trust management of the nodes is complicated due to the distributed nature of the sensing devices. Secondly, when a large number of sensing devices offload CPU resources of shared edge nodes in the process of computing tasks, malicious attacks enable the offload quantity of the computing tasks to be dynamically changed and difficult to estimate, so that the CPU resources are unevenly distributed, and even the computing tasks are unsuccessfully offloaded. Therefore, in sensing edge clouds, designing an efficient edge CPU sharing mechanism becomes a challenging problem. At the same time, sharing of edge CPUs also faces the problem of unreliability of some nodes, and malicious edge nodes may return inaccurate results or extend computation time.
The sensing edge cloud system provides various computing resource services for the Internet of things and the sensing network. Compared with the traditional sensing cloud system, the sensing edge cloud system enables the sensing equipment and the cloud computing service to realize low-delay and seamless intelligent interaction. The sensing device typically requires a significant amount of energy resources to handle both the computing task and the security verification, which increases the CPU load of the sensing device. Offloading the computing tasks to the edge nodes is an effective solution to reduce the sensor device CPU load, also known as a core networking architecture. In the sensing edge cloud core networking architecture, CPU resources are deployed on edge nodes close to sensing equipment, and computing tasks of the sensing equipment are processed by the CPU of the edge nodes, so that the CPU resources of the edge nodes are shared by a plurality of sensing equipment at the same time. When the sensing device requests CPU resources from the edge node, the control right of the CPU of the edge node is obtained, so that the calculation task can be unloaded to the edge node with low time delay and executed efficiently, and the energy consumption of the sensing device is reduced.
Although the sensor edge cloud core networking architecture is an efficient technology that saves sensor device energy consumption, deploying a sensor edge cloud core networking architecture with CPU sharing capability faces a significant challenge: (1) Due to the limited CPU resources of the edge node, it is not possible for the edge node to provide CPU resources for task processing for all sensing devices. (2) The existence of a large number of sensing devices enables malicious sensing devices to easily initiate DDoS attack, and damages the sensing device clusters formed by the sensing devices to share CPU resources. (3) Since edge nodes are deployed in an untrusted third party environment, some edge nodes may be controlled by malicious attackers, causing the edge nodes to perform untrusted actions, such as: consume CPU resources or prolong the training time of the calculation task, and return the result of lower training precision or incorrect training. (4) The interaction between the sensing device and the edge node does not need supervision of a third party, and an incentive mechanism is not used for standardizing the trusted cooperation process of the two parties. From the above, the CPU processing power constraints of the edge node, the sensor device cluster sharing edge node CPU resource characteristics, and the untrusted security threats from between the edge node and the sensor cloud will greatly impact the performance of the sensor edge system.
The sensing edge cloud core networking architecture provides an efficient solution for the sensing devices to share the CPU resources of the edge nodes and save energy consumption. Based on an edge architecture, songfan Li and the like, a core networking architecture is proposed for a wireless sensing node, in the architecture, in order to save the energy consumption of the sensing node, a CPU for processing tasks on the wireless sensing node is removed, the wireless chip and the edge node are directly connected, the CPU of the edge node is used for completing task processing (Internet-of-Microchips:Direct Radio-to-Bus Communication with SPI Backscatter.Proceedings of the26th Annual International Conference on Mobile Computing and Networking,2020,pp.1-14.). on the sensing node, but a corresponding trusted CPU resource sharing mechanism is not designed in the architecture. Xu and the like establish a trusted edge cache sharing mechanism through historical information of mobile user and edge node interaction, and verify whether cache content is legal or not by using a blockchain so as to maximize the utility of the mobile user and the edge node, and the mechanism, although optimizing cache allocation by using a price mechanism, does not consider that an attacker (Blockchain-Based Trustworthy Edge Caching Scheme for Mobile Cyber-Physical System.IEEE Internet of Things Journal,vol.7,no.2,pp.1098-1110,Feb.2020).J.Lee and the like are restrained by using the price mechanism to meet low-delay requirements of data analysis of the Internet of things equipment, manages computing resources of the mobile edge equipment based on a market model designed by a game theory, and designs the price mechanism to maximize game utility, but the model and the mechanism do not solve the problem of trusted interaction between the Internet of things equipment and the edge node (Market Analysis of Distributed Learning Resource Management for Internet of Things:A Game-Theoretic Approach.IEEE Internet of Things Journal,vol.7,no.9,pp.8430-8439,Sept.2020).
These prior studies have the following disadvantages:
(1) In order to save the energy consumption of the sensing node, although the proposed core network architecture completely removes the CPU module on the sensing node, this would make the trusted authentication mechanism not be implemented on the sensing node, resulting in the sensing node facing security problems.
(2) Although the proposed solution combines blockchain technology to verify the trusted interaction process between the edge node and the mobile user, the dynamic adjustment of the resource unit price of the attacked sensing device computing task offloading process is not performed to adapt to the change of the attack environment, and a large number of computing task dynamic offloading characteristics are not considered.
The proposed solution, while designing a price mechanism to maximize the utility of game participants, does not design a corresponding trusted sharing scheme of computing task processing resources for the dynamic computing task offload execution environment under the sensor edge cloud core networking architecture, which is attacked between the sensor device cluster and the edge nodes.
Disclosure of Invention
The invention designs a trusted edge CPU sharing scheme based on a block chain and a dynamic price mechanism. To make as efficient use of the edge node's CPU resources as possible and to suppress attacks from the sensing devices, the edge CPU scheduling mechanism combines blockchain smart contract conditions and dynamic unit price adjustments to efficiently coordinate and handle the computing tasks offloaded by the sensing devices. Therefore, the calculation task unloaded by the sensing equipment can be rapidly processed by the edge node, and the energy consumption of the sensing equipment is saved. In order to share CPU resources of the edge nodes, a plurality of sensing devices form a sensing device cluster, and tasks are unloaded from the cluster head nodes to the edge nodes, so that energy consumption of cluster member nodes is reduced. In addition, the invention designs a trust management mechanism shared by the edge CPU for the trusted selection of the edge node of the sensing equipment, which not only ensures the unloading safety of the computing task, but also is beneficial to improving the service quality of the unloading of the computing task. In the trust mechanism, the trust degree of the cluster head node and the edge node is evaluated through sensing the unloading history interaction information of the cluster head node and the edge node. Meanwhile, the blockchain is used for distributing and recording the calculation task unloading transaction information between the edge node and the cluster head node of the sensing equipment. During a transaction, the CPU resource unit price of the edge node and the payment cost of the sensing device are recorded in the block by the smart contract, and only transactions meeting the smart contract condition are legal transactions. Therefore, the invention provides a reliable CPU resource sharing method under a sensing edge cloud core networking architecture, which is based on a reliable method for sharing an edge CPU of sensing equipment by long-term memory network (LSTM) and Q-learning design, and improves the service quality of the reliability of the unloading of the computing tasks of the sensing equipment.
According to one aspect of the invention, there is provided a system for trusted sharing of CPU resources in sensory edge cloud task offloading of an integrated blockchain, comprising edge nodes and cluster head nodes;
The edge node is used for providing a calculation task unloading service for the sensing equipment, adjusting the CPU resource unit price according to the principle that the CPU resource unit price is higher as the predicted unloading task amount is larger, and providing the CPU resource unit price to the cluster head node to achieve an intelligent contract of the calculation task unloading service;
The cluster head node is used for quoting and collecting bidding actions { p o,t } of the sensing equipment and edge node CPU credibility voting rewards fed back by the sensing equipment according to CPU resource unit price provided by the edge node, and selecting credible edge nodes by adopting a reinforcement learning algorithm according to the bidding of the sensing equipment and the edge node CPU credibility voting rewards so as to achieve intelligent contract and unload calculation tasks;
the edge node and the cluster head node have blockchains for storing the intelligent contracts; the smart contract includes: the computing task size D j signed via the edge node and cluster head node private keys of both sides of the intelligent contract, the price of the edge node, the training duration, and the contract generation timestamp.
Preferably, the CPU resource trusted sharing system in the sensor edge cloud task unloading of the integrated blockchain, wherein the edge node adopts a circulating neural network to take the first one in the current calculation task unloading amount size sequence provided by the cluster head node as the current unloading amount sizeCalculating the corresponding predicted calculation task load size
Preferably, the CPU resource trusted sharing system in the sensor edge cloud task offloading of the integrated blockchain has the current computing task offloading amount sizeThe method is as follows:
where beta represents the calculated task offloading density for all sensing devices, The number of sensing devices in the sensing device cluster j for offloading calculation tasks is represented, ε= {0,1,2,..mu (t) }, μ (t) is the number of calculation tasks that the sensing devices offload in the sensing device cluster j at time t, and λ is the offloading rate.
Preferably, the memory block of the circulating neural network comprises an input layer, a control layer, a neural network and an output layer.
The input layer is used for inputting the first calculation task unloading amountLong term memory is c t-1 and short term memory h t-1.
The control layer includes three control gates: And The following logic is adopted respectively:
Wherein sigma represents a sigmoid activation function, and the function value is mapped into a range from 0 to 1; Representing the weight; b f,bi,bc denotes offset; tanh represents an activation function, and the mapping data is in the interval-1 to 1.
The output layer is used for outputting the first predicted calculation task unloading amountThe method is as follows:
Wherein, Representing weights, b o representing biases;
The method for updating the long-term memory comprises the following steps:
The short-term memory is obtained as follows: sequence of calculation task load sizes As short-term memory, i.e
Preferably, the CPU resource trusted sharing system in the sensor edge cloud task offloading of the integrated blockchain, wherein the cyclic neural network is trained to obtain network weights in the time interval T according to the following methodAnd bias b f,bi,bc,bo:
initializing network weights and biases; at each time T within the time interval T the following is performed:
Selecting an l-th computing task load size Input, calculate output h t using forward propagation and calculate the loss functionBack-propagation adjustment weights and bias parameters, where L represents the calculated task load sequence length,Indicating the actual computational task load size,Representing the predicted computational task load size.
Preferably, the system for reliably sharing CPU resource in the task offloading of the sensing edge cloud of the integrated blockchain adjusts the CPU resource unit price according to the principle that the CPU resource unit price is higher when the predicted offloading task amount is larger, and preferably calculates the adjusted CPU resource unit price by adopting the following formula
Wherein p i is the current CPU resource unit price of the edge node i,Representing the CPU maximum capacity of the edge node i, ζ is a weight parameter between the preset CPU price and the maximum capacity.
Preferably, in the system for reliably sharing CPU resources in the task unloading of the sensing edge cloud of the integrated blockchain, the cluster head nodes adopt a sensing equipment cluster learning model based on Q-learning to select trusted edge nodes.
Preferably, the system for reliably sharing CPU resources in the task unloading of the sensing edge cloud of the integrated blockchain comprises a sensing equipment cluster learning model based on Q-learning, wherein the sensing equipment cluster learning model comprises the following components:
s represents a state space of the sensor equipment cluster sharing the CPU of the edge node; a is the space of action and the space of action, A t＝{p1,t,p2,t,...,po,t,...,pM,t }, where p o,t represents a bid for each sensing device o to purchase edge node CPU resources at time slot t; t n represents the maximum bid slot number; y is the probability of transfer,M represents the number of sensing devices in the sensing device cluster, namely the number of Q-learning intelligent agents; representing rewards generated by the sensing device o; Representing rewards generated by the sensor device cluster j;
The cluster head node calculates the Q value of the CPU resource unit price under the state s t by adopting a state-action Q function of the cluster head node according to the action a t of each sensing device in the sensing device cluster in the time slot t when bidding, namely whether each sensing device purchases CPU resource of the edge node of the time slot offer, so that the Q value is maximized to maximize the local average rewards of each sensing device, and selects a trusted edge node to achieve an intelligent contract according to the bidding action strategy pi epsilon R |S|×|A| adopted by the sensing device at the moment, wherein pi represents the probability of the sensing device cluster taking the offer action under the state s t.
The state-action Q function of the cluster head node is:
The Q value update function is as follows:
The optimal strategy of the cluster head node is to select the edge node with the largest Q value, namely Action a t at that time, so the cluster head node's optimal policy pi *(at) is expressed as:
Preferably, the CPU resource trusted sharing system in the task offloading of the sensing edge cloud of the integrated blockchain has rewards generated by sensing devices o of the sensing device cluster S t+1. Epsilon.S is calculated as follows:
Where η 1,η2 is the weight parameter of performance and credibility evaluation of the training task offload execution, r 1 o (n) is the performance reward of the training task offload execution of the sensing device o, Reliability rewards for edge nodes to complete computational task offloading:
Wherein κ i represents the training duration parameter of the edge node i, and the larger r 1 o (n) is smaller, ω i represents the training accuracy parameter of the edge node i, and n represents the number of current computing tasks.
Preferably, the system for reliably sharing the CPU resource in the unloading of the sensing edge cloud task of the integrated blockchain maximizes the local average rewards of each sensing device, specifically:
Where, E π [. Cndot. ] represents the expectations of the state-action pairs of all the sensing devices in the sensing device cluster under the bidding action policy pi.
In general, the above technical solutions conceived by the present invention, compared with the prior art, enable the following beneficial effects to be obtained:
The invention provides a novel sensing edge cloud core networking architecture for saving the energy consumption of sensing equipment. In the architecture, the blockchain is designed to realize the transaction process of reliably sharing the CPU resource of the edge node when the computing task of the sensor equipment cluster is unloaded. The behavior of the dual transaction is constrained using intelligent contract conditions and the computing task processing results are checked.
In order to process dynamic change characteristics of computing task unloading capacity of a sensing edge cloud core networking architecture under attack, the invention provides a low-complexity LSTM algorithm to predict the unloading capacity of a computing task of a sensing device cluster, and dynamically adjusted CPU resource unit price is output through the predicted computing task unloading capacity, so that the utility of edge nodes is maximized.
In order to maximize the utility of the sensor device cluster, the invention formalizes a reward function based on the combination of training precision, duration and credibility voting, and proposes a voting reward Q-learning based sensor device cluster bid optimization algorithm.
Drawings
FIG. 1 is a schematic diagram of a CPU resource trusted sharing system in sensor edge cloud task offloading of an integrated blockchain provided by the invention;
FIG. 2 is a schematic diagram of a sensor edge cloud core networking security architecture;
FIG. 3 is a schematic diagram of a sensor cluster trusted shared CPU resource framework based on smart contract dynamic prices;
Fig. 4 is a schematic diagram of a computing task load capacity prediction network structure based on a three-layer LSTM.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
A system for trusted sharing of CPU resources in sensor edge cloud task offloading of an integrated blockchain, as shown in fig. 1, comprising: edge nodes and cluster head nodes;
The edge node is used for providing a calculation task unloading service for the sensing equipment, adjusting the CPU resource unit price according to the principle that the CPU resource unit price is higher as the predicted unloading task amount is larger, and providing the CPU resource unit price to the cluster head node to achieve an intelligent contract of the calculation task unloading service;
the edge node adopts a circulating neural network to serve as the current unloading amount of the calculation task according to the first one in the current calculation task unloading amount sequence provided by the cluster head node Calculating the corresponding predicted calculation task load size
The current calculation task load capacityThe method is as follows:
where beta represents the calculated task offloading density for all sensing devices, The number of sensing devices for unloading the computing tasks in the sensing device cluster j is represented, and is provided by a cluster head node, epsilon= {0,1,2,.. The sensing device cluster j is represented by mu (t) }, mu (t) is the number of the computing tasks unloaded by the sensing devices in the sensing device cluster j at time t, and lambda is the unloading rate.
The memory block of the cyclic neural network comprises an input layer, a control layer, a neural network and an output layer.
The input layer is used for inputting the first calculation task unloading amountLong term memory is c t-1 and short term memory h t-1.
The control layer includes three control gates: And The following logic is adopted respectively:
Wherein sigma represents a sigmoid activation function, and the function value is mapped into a range from 0 to 1; Representing the weight; b f,bi,bc denotes offset; tanh represents an activation function, and the mapping data is in the interval-1 to 1.
The output layer is used for outputting the first predicted calculation task unloading amountThe method is as follows:
Where W o T represents the weight and b o represents the bias.
The method for updating the long-term memory comprises the following steps:
The short-term memory is obtained as follows: sequence of calculation task load sizes As short-term memory, i.e
The cyclic neural network is trained to obtain network weights within the time interval T according to the following methodAnd bias b f,bi,bc,bo:
initializing network weights and biases; at each time T within the time interval T the following is performed:
Selecting an l-th computing task load size Input, calculate output h t using forward propagation and calculate the loss functionBack-propagation adjustment weights and bias parameters, where L represents the calculated task load sequence length,Indicating the actual computational task load size,Representing the predicted computational task load size.
The CPU resource unit price is regulated according to the principle that the CPU resource unit price is higher when the predicted task unloading amount is larger, and the regulated CPU resource unit price is preferably calculated by adopting the following formula
Wherein p i is the current CPU resource unit price of the edge node i,Representing the CPU maximum capacity of the edge node i, ζ is a weight parameter between the preset CPU price and the maximum capacity.
The cluster head node is used for quoting and collecting bidding actions { p o,t } of the sensing equipment and edge node CPU credibility voting rewards fed back by the sensing equipment according to CPU resource unit price provided by the edge node, and selecting credible edge nodes by adopting a reinforcement learning algorithm according to the bidding of the sensing equipment and the edge node CPU credibility voting rewards so as to achieve intelligent contract and unload calculation tasks;
the cluster head node adopts a sensing equipment cluster learning model based on Q-learning to select a trusted edge node, and the specific method is as follows:
The sensing equipment cluster learning model based on Q-learning is as follows:
s represents a state space of the sensor equipment cluster sharing the CPU of the edge node; a is the space of action and the space of action, A t＝{p1,t,p2,t,...,po,t,...,pM,t }, where p o,t represents a bid for a sensor device o in the sensor device cluster to purchase edge node CPU resources at time interval t; t n represents the maximum bid slot number; y is the probability of transfer,M represents the number of sensing devices in the sensing device cluster, namely the number of Q-learning intelligent agents; representing rewards generated by sensor device clusters j, wherein rewards generated by sensor devices o in sensor device clusters The method is as follows:
Where η 1,η2 is the weight parameter of performance and credibility evaluation of the training task offload execution, r 1 o (n) is the performance reward of the training task offload execution of the sensing device o, Reliability rewards for edge nodes to complete computational task offloading:
Wherein κ i represents the training duration parameter of the edge node i, and the larger r 1 o (n) is smaller, ω i represents the training accuracy parameter of the edge node i, and n represents the number of current computing tasks.
The cluster head node calculates the Q value of the CPU resource unit price under the state s t by adopting a state-action Q function of the cluster head node according to the action a t of the gap t of the sensor equipment cluster when bidding, namely whether each sensor equipment purchases CPU resource of the edge node of the time slot offer, wherein the weighted average of the bids is taken as the CPU resource unit price, so that the Q value is maximized to maximize the local average rewards of each sensor equipment, and a trusted edge node is selected according to the bidding action strategy pi E R |S|×|A| adopted by the sensor equipment at the moment to achieve an intelligent contract, and pi represents the probability of the sensor equipment cluster taking the offer action under the state s t.
The method comprises the steps of maximizing local average rewards of each sensing device, specifically:
Where, E π [. Cndot. ] represents the expectations of the state-action pairs of all the sensing devices in the sensing device cluster under the bidding action policy pi.
The state-action Q function of the cluster head node is:
The Q value update function is as follows:
The optimal strategy of the cluster head node is to select the edge node with the largest Q value, namely Action a t at that time, so the cluster head node's optimal policy pi *(at) is expressed as:
the edge node and the cluster head node have blockchains for storing the intelligent contracts; the smart contract includes: the computing task size D j signed via the edge node and cluster head node private keys of both sides of the intelligent contract, the price of the edge node, the training duration, and the contract generation timestamp.
The following are examples:
The sensing edge cloud core networking security architecture consists of two parts, namely sensing equipment and edge nodes, as shown in fig. 2: three CPUs are deployed in the edge node, and the edge node is connected with the sensing device through the wireless chip and provides a computing task unloading service for the sensing device. A CPU is deployed in the sensing device for running lightweight security verification. The sensing equipment only performs data collection and lightweight security verification, and the computing task is processed by the edge node, so that the energy consumption of the sensing equipment is effectively saved. On this architecture, a plurality of sensing devices form a sensing device cluster sharing the CPU resources of the edge node.
A CPU resource trusted sharing system in the task unloading of a sensing edge cloud integrating a blockchain is shown in fig. 1, and the sensing edge cloud cluster topology based on a core networking architecture consists of sensing equipment, sensing equipment clusters and edge nodes. The edge node is provided with a plurality of CPUs for providing task processing services for the sensing equipment, and when the sensing equipment unloads calculation tasks to the edge node to share CPU processing resources, the processing time of the edge node is proportional to the calculation task unloading capacity of the sensing equipment cluster. In the sensing edge cloud, in order to increase the reliability and security of a network, generally, sensing devices form a cluster topology, share data and CPU resources of edge nodes, a sensing device cluster member selects a cluster head node to initiate a shared CPU request to the edge nodes, and the number of interactions between the sensing devices and the edge nodes is obviously reduced by the sharing mode.
In the sensing edge cloud, the set of edge nodes is e= {1,2, I, each edge node CPU processing capacity isIn order to achieve trusted offloading of computing tasks, the invention considers malicious edge nodes, which can launch attacks on the offloaded computing tasks, such as: consume CPU resources, extend training time for tasks, and the like. The sensor device cluster may be represented as v= {1,2, once. At time t, the number of sensing devices in the sensing device cluster is M, and the sensing device cluster has high CPU requirements in order to process the computing task.
The sensing device cluster shares the CPU resources of the edge node with three advantages: (1) In the same sensor device cluster, the sensor devices share the cost of purchasing edge CPU resources, which enables sensor device cluster members to obtain more shared CPU resources with minimal cost. (2) In the cluster topology environment, the sensor equipment cluster head node interacts with the edge node to request CPU resources, so that the interaction times of the sensor equipment and the edge node are reduced, and meanwhile, the message quantity in the cluster topology is effectively saved. (3) The sensor equipment cluster head node unloads the calculation tasks of the cluster members to the edge node, the sensor equipment cluster members are not required to be directly unloaded to the edge node, the trust state of the edge node is only required to be verified through the sensor equipment cluster head node, and the CPU resource consumption of the cluster members for safety verification is reduced.
In order to suppress malicious attacks of the sensor device cluster and evaluate the trusted condition of the edge node and update the trust, the invention provides a sensor device cluster trusted shared CPU resource framework based on intelligent contract dynamic price, as shown in figure 3. In the framework, the LSTM network and the Q-learning are integrated into the framework. The LSTM-based sensing device calculates a task load sensing method to process the task load dynamically unloaded when attacked. The trusted status of the edge node is processed by a Q-learning-based edge node trusted status learning method. In order to realize the dynamic calculation task unloading capacity sensing of the sensor device cluster, the calculation task unloading capacity of the sensor device is processed regularly through an LSTM network, and then the LSTM weight is updated according to a prediction result. After training is finished, the calculation task unloading capacity of the sensor equipment cluster is obtained in a prediction reasoning stage, and the dynamic unit price of the CPU resource of the edge node is obtained, so that malicious attack behaviors of the sensor equipment cluster head node are restrained. The sensor cluster head node uses Q-learning to evaluate and update the credibility of the edge node, and uses whether the intelligent contract condition is met to evaluate the credibility of the edge node, and meanwhile obtains the dynamic unit price of purchasing CPU resources, so as to achieve the purpose of maximizing the utility of the sensor cluster.
The edge node is used for providing a calculation task unloading service for the sensing equipment, adjusting the CPU resource unit price according to the principle that the CPU resource unit price is higher as the predicted unloading task amount is larger, and providing the CPU resource unit price to the cluster head node to achieve an intelligent contract of the calculation task unloading service;
Thus, each edge node decides the unit price of its CPU resource sharing services and maximizes its utility. The utility of an edge node is expressed as:
Where θ i is the computational cost of consuming the CPU per computational task. Where φ represents the Joule coefficient corresponding to one energy unit, E i is the energy consumed by the CPU to process one computational task, and f i is the number of CPU cycles.Is the capacity requirement of the sensing device for CPU resources in the edge node. p i is the unit price of the current sensing device to calculate task usage CPU resources.
When the demand of all the sensing devices on the shared CPU resource capacity is high, in order to prevent the calculation task from failing, malicious nodes are restrained from sharing the CPU resource, meanwhile, a trusted calculation result is ensured, and the unit price of the edge nodes is raised. When the demand of the sensing equipment on the shared CPU resource capacity is low, the edge node reduces the unit price of the shared CPU resource, so that the utilization rate of the CPU resource by the edge node is improved. In order to optimize the unit price of CPU resources, the edge node estimates the total calculation task load of each sensor device cluster, but because the calculation task load of the edge node is dynamically changed along with time, in order to estimate the calculation task load of the sensor device cluster in the future, the invention considers that the calculation task load v j,o of each sensor device o in the sensor device cluster j is poisson distribution, and the load rate is lambda, so that the probability that each sensor device loads calculation tasks to a single edge node is as follows:
Where ε= {0,1,2,..mu (t) }, μ (t) is the number of computational tasks that the sensing device offloaded in time t sensing device cluster j. Beta represents the calculated task offload density for all sensing devices. Then, the current calculation task load capacity can be calculated according to the above formula as follows:
Wherein, And the number of the sensing devices for unloading the calculation tasks in the sensing device cluster j is represented.
The edge node adopts a cyclic neural network to calculate the current unloading amount of the task according to the current computing task unloading amount provided by the cluster head node in the sequence of the current computing task unloading amountCalculating a calculation task load size for the first calculation task prediction
Since LSTM can memorize longer sequence of unloading calculation tasks, it can model sequence information of unloading calculation tasks in sensor equipment cluster. At the same time, LSTM adjusts the flow of information in the memory block by controlling the state of the gate to remove or add information to the memory block. The invention designs a three-layer LSTM memory block to predict the load-shedding amount of the calculation task. As shown in fig. 4, the white block diagram shows an LSTM memory structure, and ce1=ce2=ce3 shows that the three memory blocks have the same structure. The current input computing task load is as followsThe output predicted calculation task unloading amount sequence is thatThe long term memory is c t, the short term memory is h t, and the three control gates are shown below:
where σ represents a sigmoid activation function, mapping the function value to a 0 to 1 interval. Representing the weights. b f,bi,bc denotes offset. tanh represents an activation function, and the mapping data is in the interval-1 to 1.
The update process of the memory block states from c t-1 to c t is as follows:
(1) From the following components Filtering the information of the unloading capacity of the computing task which is not wanted to be reserved; (2) And adding the current newly added calculation task unloading amount information. The updated task load memory information in the memory block is obtained by the method:
Finally, obtaining output information of the calculated task unloading amount is as follows:
Wherein, B o is the bias, which is the weight. The predicted output value is thus obtained asFor the edge node i, the input calculation task unloading amount size sequence is as followsObtaining the predicted value of the calculated task unloading capacity asThe three-layer LSTM based predictive model can be described as:
The loss function of the LSTM network is:
Where L represents the calculation task load sequence length. Y t l represents the actual calculation task load size. Predicted computational task load size The weight and bias parameters in the LSTM network are updated using a random gradient descent method on the loss function. After training LSTM network, inputting current calculation task unloading capacityObtaining a predicted computational task load sizeAnd dynamically adjusting the unit price of the CPU resource of the edge node according to the predicted calculation task unloading amount.
LSTM-based computing task load prediction and CPU resource dynamic price optimization (Algorithm 1), comprising:
The first stage: training LSTM networks
(1) Initializing: calculating task unloading time interval T, wherein training data is a sequence of calculating task unloading capacity
(2) Random initialization of network weightsAnd bias b f,bi,bc,bo
(3) For every training interval do
(4) Selection in computing task load size sequences
(5) Calculating output h using forward propagation t
(6) Calculating a loss function
(7) Counter-propagation adjustment weights and bias parameters
(8)End For
(9) Outputting network weightsAnd bias b f,bi,bc,bo.
And a second stage: predicting computational task load size
(1) And (3) loading: number of prediction intervals, time interval T, weightAnd bias b f,bi,bc,bo
(2) For each prediction interval do
(3) Fetching from a current computing task load size set
(4) Forward propagation, calculate output h t
(5) Calculating a loss function
(6)End For
(7) Output the calculated task load size of the next time slot
(8)IfThen
(9)Else
(10) Output of
The algorithm can obtain the optimal CPU resource price when the edge node is maximizedWherein,Representing the CPU maximum capacity of the edge node i. ζ is a weight parameter between the CPU price and the maximum capacity.
Since the sensing device offloads a large number of calculation tasks to the edge node through the cluster head node in the continuous time slot, an attacker initiates a DDoS attack to the edge node through the offloaded time slot, and CPU time and resources of the edge node are wasted. At this time, the attacked edge node uses the calculation task unloading capacity prediction and the CPU dynamic price mechanism in the algorithm 1 to inhibit the attack of malicious attackers, so as to reduce the waste of CPU resources. When the edge node predicts that the calculated task unloading amount is higher, the edge node modifies CPU resource unit price in the intelligent contract, dynamically increases unit price, effectively reduces the load of the edge node CPU, and simultaneously relieves the influence of an attacker on the shared edge node CPU resource of the sensor cluster.
The cluster head node is used for quoting and collecting bidding actions { p o,t } of the sensing equipment and edge node CPU credibility voting rewards fed back by the sensing equipment according to CPU resource unit price provided by the edge node, and selecting credible edge nodes by adopting a reinforcement learning algorithm according to the bidding of the sensing equipment and the edge node CPU credibility voting rewards so as to achieve intelligent contract and unload calculation tasks;
In the sensing edge cloud core networking architecture, part of edge nodes can be malicious, and the unloading process of the computing task and the result attack can be carried out, and the method mainly prolongs the training time, returns inaccurate training precision and the like. In order to avoid the reduction of the quality of service of the unloading of the computing tasks of the sensor device cluster caused by malicious edge nodes, a trusted mechanism is required to be designed to provide reliable and safe shared CPU resource service for the sensor device cluster and ensure the reliability and training precision of the execution of the computing tasks of the unloading of the sensor device cluster. According to interaction between the cluster head node and the edge node of the sensing equipment, the sensing equipment feeds back and evaluates CPU resource service provided by the edge node through voting. If the CPU shared resource service provided by the edge node is trusted, the cluster of sensing devices records and updates the benefit obtained from the edge node. The CPU resource price dynamically adjusted by the edge node has a certain influence on the CPU resource of the edge node purchased by the sensor equipment cluster bid. For those edge nodes with lower CPU prices, the sensor device cluster can easily obtain CPU resources, but an attacker can use the price advantage to attack the sensor device cluster, delay the calculation task time and return inaccurate results. Thus, to maximize the utility of the cluster of sensing devices, ensure a reliable computational task offload, improve the reliability of the computational results, the sensing devices bid on the CPU resources of the edge nodes by voting on the edge nodes and evaluate the benefits obtained by the bidding, thereby selecting a reliable cluster of sensing devices to offload the computational tasks, the utility of the cluster of sensing devices j being determined by its benefits and costs, which can be expressed as follows:
uj＝Γ(Dj)-Ω(pj,Dj)
Wherein Γ (D j) represents satisfaction of training results returned by the edge node when the calculation task size of the sensing device cluster is D j; Ω (p j,Dj) represents the cost function of the sensing device. To maximize the utility of the sensor device cluster, the sensor device cluster takes an optimal price to obtain trusted CPU resources.
And the cluster head node adopts a sensing equipment cluster learning model based on Q-learning to select a trusted edge node.
The sensing device cluster learning model based on Q-learning is defined as follows:
S represents a state space of a sensor device cluster shared edge node CPU, and the state space comprises a trusted state and an untrusted state; a is the action space, a= { a t }, Wherein p o,t represents the bid for each sensing device o to purchase edge node CPU resources at time interval t in the bid; t n represents the maximum bid slot number; y is the probability of transfer,M represents the number of sensing devices in the sensing device cluster, namely the number of Q-learning intelligent agents; representing rewards generated by sensor device clusters j, wherein rewards generated by sensor devices o in sensor device clusters The method is as follows:
Where η 1,η2 is the weight parameter of performance and credibility evaluation of the training task offload execution, r 1 o (n) is the performance reward of the training task offload execution of the sensing device o, Reliability rewards for edge nodes to complete computational task offloading:
Wherein κ i represents the training duration parameter of the edge node i, and the larger r 1 o (n) is smaller, ω i represents the training accuracy parameter of the edge node i, and n represents the number of current computing tasks.
In the system, each sensing device decides its own bid, and its working process is as follows: (1) All sensing devices observe state S t e S. (2) Each sensing device takes bidding action a t in state s t based on the historical votes. (3) The system state space passes to the next state S t+1 e S with a probability of passingIf the cluster of sensing devices does not acquire CPU resources of the edge node, bidding continues. If the sensor equipment cluster obtains CPU resources of the edge node, after a period of time, obtaining training precision and duration provided by the CPU of the edge node and generating credibility voting rewards asThus, this reward function is divided into two parts, one part is the performance of the training task offload execution, and the other part is the credibility evaluation of the edge node to complete the computing task offload, which is represented by the voting value.
The cluster head node calculates the Q value of the CPU resource unit price under the state s t by adopting a state-action Q function of the cluster head node according to the action a t of each sensing device in the sensing device cluster in the time slot t when bidding, namely whether each sensing device purchases CPU resource of the edge node of the time slot offer, so that the Q value is maximized to maximize the local average rewards of each sensing device, and selects a trusted edge node to achieve an intelligent contract according to the bidding action strategy pi epsilon R |S|×|A| adopted by the sensing device at the moment, wherein pi represents the probability of the sensing device cluster taking the offer action under the state s t.
The method comprises the steps of maximizing local average rewards of each sensing device, specifically:
Where, E π [. Cndot. ] represents the expectations of the state-action pairs of all the sensing devices in the sensing device cluster under the bidding action policy pi.
The state-action Q function of the cluster head node is:
The Q value update function is as follows:
The optimal strategy of the cluster head node is to select the edge node with the largest Q value, namely Action a t at that time, so the cluster head node's optimal policy pi *(at) is expressed as:
The sensing equipment cluster sharing edge node CPU resource optimal bid optimization (algorithm 2) based on Q-learning comprises the following steps:
(1) Initializing tuples
(2)For t＝1,2,...,Tn do
(3) Based on historical voting informationSensing device bid a t
(4) If the sensor equipment cluster does not acquire CPU resources of the edge node, turning to the step (3), otherwise, executing the step (5)
(5) The system state space passes to the next state s t+1 and generates a local rewards
(6)For j＝1,2...,M do
(7) Each sensing device updates its local Q value
(8)End For
(9) The sensor cluster head node searches for rewards for each sensor, and updates the global Q value using equation (17)
(10)End For
(11) Output of
The optimal strategy pi *(at) and actions of the sensor device cluster can be obtained by the algorithm 2And optimal biddingAt this time, the sensing device cluster obtains the maximum utility, and the sensing device is ensured to obtain the CPU resource of the edge node credibility.
In the sensor device cluster, the cluster member node firstly unloads the calculation task to the cluster head node, and then the cluster head node unloads the calculation task to the edge node, and the calculation task unloading set of the cluster member node is DT j＝{DTj,1,DTj,2,...,DTj,o,...,DTv,M. Wherein DT j,o represents the calculation task load capacity of the sensing device o in the sensing device cluster j, and the total calculation task size is:
In the unloading of the computing task, the trusted transaction information of the edge node and the cluster head node of the sensing equipment needs to be determined respectively. In order to record and restrict the transaction process of the edge node and the sensing equipment cluster head node, the invention deploys a blockchain between the edge node and the sensing equipment cluster head node, stores transaction information into the sensing equipment blockchain, and the sensing equipment cluster head node is responsible for creating new block storage calculation task unloading transaction information, and also receives information returned by the edge node, verifies a calculation result and then sends the calculation result to sensing equipment cluster members. The process of verifying the calculation result is as follows: and triggering an intelligent contract to judge whether the calculation task unloading transaction meets the contract condition or not when the edge node returns, and recording in a block of the cluster head node of the sensing equipment if the contract condition is met.
The edge node and the cluster head node have blockchains for storing the intelligent contracts; the smart contract includes: the computing task size D j signed via the edge node and cluster head node private keys of both sides of the intelligent contract, the price of the edge node, the training duration, and the contract generation timestamp.
In the present invention, the smart contract is a programmatic protocol for restricting collaborative negotiating procedures between edge nodes and sensor device cluster head nodes. Under the condition that no third party node participates, the intelligent contract predefines the trusted conditions of the transaction of the shared CPU resources of the two parties, and automatically and dynamically adjusts the trusted shared CPU prices of the edge node and the cluster head node of the sensing equipment. The negotiating process of the smart contract includes three phases: negotiations, deployments, and transactions.
Negotiating: sensor cluster head nodes interact with edge nodes to negotiate intelligent contracts to restrict sharing of edge nodes
CPU resources. The smart contract contains the following entries: the size of the task D j, the price of the edge node, the training duration, and the contract generation timestamp are calculated. After negotiating the intelligent contract, the edge node and the sensor device cluster head node are signed with private keys, respectively, in order to ensure contract security.
Deployment phase: after the edge node and the sensor equipment cluster head node negotiate intelligent contracts, the edge node and the sensor equipment cluster head node respectively deploy contracts into the blockchain and issue contract addresses, and after the unloading of the computing task is completed, the edge node and the sensor equipment cluster head node respectively access the intelligent contract addresses to verify whether the two parties observe the intelligent contracts. This effectively prevents the untrusted behavior of the two parties.
Shared CPU resource transaction phase: in order to prevent illegal sensor cluster head nodes from repeatedly unloading calculation tasks, the edge nodes generate tokens for sharing CPU resources, the edge nodes send the serial numbers of the CPUs containing signature information and the token information to the sensor cluster head nodes, and after the sensor cluster head nodes verify the information, the sensor cluster head nodes generate a calculation task unloading transaction and send the transaction to an intelligent contract address and bind the intelligent contract. At the same time, the edge node and the sensor device cluster head node record this transaction into the blockchain. And finally, starting an intelligent contract agent to monitor the transaction process of the edge node and the cluster head node of the sensing equipment.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. The CPU resource trusted sharing system in the sensor edge cloud task unloading of the integrated block chain is characterized by comprising edge nodes and cluster head nodes;
The edge node is used for providing a calculation task unloading service for the sensing equipment, adjusting the CPU resource unit price according to the principle that the CPU resource unit price is higher as the predicted unloading task amount is larger, and providing the CPU resource unit price to the cluster head node to achieve an intelligent contract of the calculation task unloading service;
The cluster head node is used for quoting and collecting bidding actions { p o,t } of the sensing equipment and edge node CPU credibility voting rewards fed back by the sensing equipment according to CPU resource unit price provided by the edge node, and selecting credible edge nodes by adopting a reinforcement learning algorithm according to the bidding of the sensing equipment and the edge node CPU credibility voting rewards so as to achieve intelligent contract and unload calculation tasks;
the edge node and the cluster head node have blockchains for storing the intelligent contracts; the smart contract includes: the computing task size D j signed via the edge node and cluster head node private keys of both sides of the intelligent contract, the price of the edge node, the training duration, and the contract generation timestamp.
2. The system for reliably sharing CPU resources in sensor edge cloud task offloading of an integrated blockchain as in claim 1, wherein the edge node calculates its corresponding predicted computational task offload size using a recurrent neural network based on the first of a current computational task offload size sequence provided by a cluster head node as a current offload size X t l
3. The system for trusted sharing of CPU resources in sensory edge cloud task offloading of integrated blockchains of claim 2, wherein the current computing task offload sizeThe method is as follows:
where beta represents the calculated task offloading density for all sensing devices, The number of sensing devices in the sensing device cluster j for offloading calculation tasks is represented, ε= {0,1,2,..mu (t) }, μ (t) is the number of calculation tasks that the sensing devices offload in the sensing device cluster j at time t, and λ is the offloading rate.
4. The system for trusted sharing of CPU resources in sensor edge cloud task offloading of integrated blockchains of claim 2, wherein the memory blocks of the recurrent neural network include an input layer, a control layer, a neural network, and an output layer;
The input layer is used for inputting the first calculation task unloading amount X t l, long-term memory of c t-1 and short-term memory of h t-1;
The control layer includes three control gates: g ft、git and g ct, respectively, employ the following logic:
Wherein sigma represents a sigmoid activation function, and the function value is mapped into a range from 0 to 1; Representing the weight; b f,bi,bc denotes offset; tanh represents an activation function, and the mapping data is in a range from-1 to 1;
The output layer is used for outputting the first predicted calculation task unloading amount The method is as follows:
Wherein, Representing weights, b o representing biases;
The method for updating the long-term memory comprises the following steps:
The short-term memory is obtained as follows: calculating task load size to be predicted As short-term memory, i.e
5. The system for reliably sharing CPU resources in sensor edge cloud task offloading of integrated blockchain of claim 4, wherein the recurrent neural network is trained to obtain network weights during time interval T as followsAnd bias b f,bi,bc,bo:
initializing network weights and biases; at each time T within the time interval T the following is performed:
Selecting an l-th computing task load size Input, calculate output h t using forward propagation and calculate the loss functionBack-propagation adjustment weights and bias parameters, where L represents the calculated task load sequence length,Indicating the actual computational task load size,Representing the predicted computational task load size.
6. The system for trusted sharing of CPU resources in sensor edge cloud task offloading of an integrated blockchain of claim 1, wherein the CPU resource unit price is adjusted according to a principle that the CPU resource unit price is higher as the predicted offload task amount is larger, and the adjusted CPU resource unit price p i * is calculated by adopting the following formula:
Wherein p i is the current CPU resource unit price of the edge node i, Representing the CPU maximum capacity of the edge node i, ζ is a weight parameter between the preset CPU price and the maximum capacity,Representing the predicted computational task load size.
7. The system for trusted sharing of CPU resources in sensor edge cloud task offloading of an integrated blockchain of claim 1, wherein the cluster head node selects trusted edge nodes using a Q-learning based sensor device cluster learning model.
8. The system for trusted sharing of CPU resources in sensor edge cloud task offloading of an integrated blockchain of claim 7, wherein the Q-learning based sensor device cluster learning model is:
s represents a state space of the sensor equipment cluster sharing the CPU of the edge node; a is the space of action and the space of action, A t＝{p1,t,p2,t,...,po,t,...,pM,t }, where p o,t represents a bid for each sensing device o to purchase edge node CPU resources at time slot t; t n represents the maximum bid slot number; y is the probability of transfer,M represents the number of sensing devices in the sensing device cluster, namely the number of Q-learning intelligent agents; representing rewards generated by the sensing device o; Representing rewards generated by the sensor device cluster j;
The cluster head node calculates the Q value of the unit price of the CPU resource under the state s t by adopting a state-action Q function of the cluster head node according to the action a t of the gap t of each sensing device in the sensing device cluster when bidding, namely whether each sensing device purchases the CPU resource of the edge node of the time gap offer, and selects a trusted edge node to achieve an intelligent contract according to the bidding action strategy pi epsilon R |S|×|A| adopted by the sensing device at the moment, wherein pi represents the probability of the sensing device cluster taking the offer action under the state s t;
The state-action Q function of the cluster head node is:
The Q value update function is as follows:
The optimal strategy of the cluster head node is to select the edge node with the largest Q value, namely Action a t at that time, so the cluster head node's optimal policy pi *(at) is expressed as:
9. the system for trusted sharing of CPU resources in sensory edge cloud task offloading of integrated blockchain of claim 8, wherein the rewards generated by sensory devices o of the cluster of sensory devices The method is as follows:
Wherein eta 1,η2 is a weight parameter for performance and credibility evaluation of the unloading execution of the training task, The performance rewards performed for the training tasks of the sensing device o are offloaded,Reliability rewards for edge nodes to complete computational task offloading:
wherein κ i represents the training duration parameter of edge node i, the larger resulting in The smaller ω i represents the edge node i training accuracy parameter and n represents the current calculation task number.
10. The system for trusted sharing of CPU resources in sensor edge cloud task offloading of integrated blockchains of claim 8, wherein the maximizing of the local average rewards for each sensor device is specifically:
Where, E π [. Cndot. ] represents the expectations of the state-action pairs of all the sensing devices in the sensing device cluster under the bidding action policy pi.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110186724.XA CN112783662B (en) | 2021-02-18 | 2021-02-18 | CPU resource sharing system in sensor edge cloud task unloading of integrated block chain |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110186724.XA CN112783662B (en) | 2021-02-18 | 2021-02-18 | CPU resource sharing system in sensor edge cloud task unloading of integrated block chain |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112783662A CN112783662A (en) | 2021-05-11 |
| CN112783662B true CN112783662B (en) | 2024-07-12 |
Family
ID=75761520
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110186724.XA Active CN112783662B (en) | 2021-02-18 | 2021-02-18 | CPU resource sharing system in sensor edge cloud task unloading of integrated block chain |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112783662B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113382073B (en) * | 2021-06-08 | 2022-06-21 | 重庆邮电大学 | Monitoring system and method for edge nodes in cloud edge-side industrial control system |
| CN113114790B (en) * | 2021-06-10 | 2021-09-14 | 武汉研众科技有限公司 | Load balancing method and system based on block chain and edge calculation |
| CN113590328B (en) * | 2021-08-02 | 2023-06-27 | 重庆大学 | Edge computing service interaction method and system based on block chain |
| CN114048578A (en) * | 2021-09-03 | 2022-02-15 | 南京邮电大学 | High-throughput block chain system and performance optimization model for 6G network |
| CN114661378B (en) * | 2022-03-31 | 2024-06-25 | 华南理工大学 | Meta universe multidimensional computing power unloading scheme based on encryption computing block chain |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101156618B1 (en) * | 2008-11-21 | 2012-06-14 | 연세대학교 산학협력단 | Method for allocating resources for wireless network |
| US10326766B2 (en) * | 2017-07-13 | 2019-06-18 | Dell Products, Lp | Method and apparatus for optimizing mobile edge computing for nomadic computing capabilities as a service |
| US10924363B2 (en) * | 2018-04-13 | 2021-02-16 | The Curators Of The University Of Missouri | Method and system for secure resource management utilizing blockchain and smart contracts |
| US11153621B2 (en) * | 2019-05-14 | 2021-10-19 | At&T Intellectual Property I, L.P. | System and method for managing dynamic pricing of media content through blockchain |
| CN110417872B (en) * | 2019-07-08 | 2022-04-29 | 深圳供电局有限公司 | Edge network resource allocation method facing mobile block chain |
| CN110418416B (en) * | 2019-07-26 | 2023-04-18 | 东南大学 | Resource allocation method based on multi-agent reinforcement learning in mobile edge computing system |
| US12041177B2 (en) * | 2020-09-25 | 2024-07-16 | Intel Corporation | Methods, apparatus and systems to share compute resources among edge compute nodes using an overlay manager |
| CN112202928B (en) * | 2020-11-16 | 2022-05-17 | 绍兴文理学院 | Credible unloading cooperative node selection system and method for sensing edge cloud block chain network |
-
2021
- 2021-02-18 CN CN202110186724.XA patent/CN112783662B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112783662A (en) | 2021-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112783662B (en) | CPU resource sharing system in sensor edge cloud task unloading of integrated block chain | |
| Shi et al. | Priority-aware task offloading in vehicular fog computing based on deep reinforcement learning | |
| Heidari et al. | A green, secure, and deep intelligent method for dynamic IoT-edge-cloud offloading scenarios | |
| Li et al. | Blockchain-based data trading in edge-cloud computing environment | |
| Li et al. | Credit-based payments for fast computing resource trading in edge-assisted Internet of Things | |
| CN113992676B (en) | Incentive method and system for layered federal learning under terminal edge cloud architecture and complete information | |
| Mazzucco et al. | Maximizing cloud providers' revenues via energy aware allocation policies | |
| Duggan et al. | A multitime‐steps‐ahead prediction approach for scheduling live migration in cloud data centers | |
| Tang | Large-scale computing systems workload prediction using parallel improved LSTM neural network | |
| CN111262944B (en) | Method and system for hierarchical task offloading in heterogeneous mobile edge computing network | |
| Razaque et al. | Energy-efficient and secure mobile fog-based cloud for the Internet of Things | |
| Mohammadi et al. | A novel optimized approach for resource reservation in cloud computing using producer–consumer theory of microeconomics | |
| Wu et al. | Optimal deploying IoT services on the fog computing: A metaheuristic-based multi-objective approach | |
| Chen et al. | A novel deep policy gradient action quantization for trusted collaborative computation in intelligent vehicle networks | |
| CN116669111A (en) | Mobile edge computing task unloading method based on blockchain | |
| Xiao et al. | Deep reinforcement learning for optimal resource allocation in blockchain-based IoV secure systems | |
| Li et al. | Computation offloading and service allocation in mobile edge computing | |
| Qi et al. | LightPoW: A trust based time-constrained PoW for blockchain in internet of things | |
| Qin et al. | User‐Edge Collaborative Resource Allocation and Offloading Strategy in Edge Computing | |
| Maciel et al. | Resource aware client selection for federated learning in IoT scenarios | |
| Liu et al. | Edge data caching with consumer-centric service prediction in resilient industry 5.0 | |
| Roy et al. | Dynamic pricing for sensor-cloud platform in the presence of dumb nodes | |
| Sermakani et al. | Dynamic provisioning of virtual machine using optimized bit matrix load distribution in federated cloud | |
| Durga et al. | A novel request state aware resource provisioning and intelligent resource capacity prediction in hybrid mobile cloud | |
| Edalat et al. | Auction‐based task allocation with trust management for shared sensor networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |