Research on multipriority scheduling technology for real-time interconnection between industrial field data and cloud information

1.1. Application requirements of industrial Internet

3.1. Upline scheduling process of 5G NR

First, when UE needs to send uplink data, it puts the required data into the cache and then submits its buffer state report to the base station through the physical uplink control channel. At the same time, the scheduling request is sent to inform the base station gNB (5G base station) that it needs to send data.¶

Second, the uplink scheduler of gNB receives an uplink scheduling request from UE, and gNB allocates resources to UE on the basis of UE's cache status report and the uplink channel condition of UE. The uplink channel status of UE is obtained by the sounding reference signal that UE periodically sends to gNB. The allocation results are sent to UE via the physical downlink control channel by using the uplink grant.¶

Third, UE uses the resources allocated by the base station to send data to the base station through the physical uplink shared channel.¶

The uplink scheduler of gNB receives the cache status report and upline channel status of UE and completes the dynamic scheduling of time-frequency resources according to the built-in scheduling algorithm. There are three common scheduling algorithms: Round-Robin(RR) algorithm, Max C / Ι algorithm, Proportional Fairness(PF) algorithm.¶

The uplink scheduler of gNB receives the cache status report and the upline channel status of UE and completes the dynamic scheduling of time-frequency resources according to the built-in scheduling algorithm. The three most common scheduling algorithms are the round robin (RR) algorithm, max C/Ι algorithm, and proportional fairness(PF) algorithm.¶

The RR algorithm allocates resources for different users of request scheduling in a circular way. This algorithm only considers the fairness among users and loses the system throughput. The max C/I algorithm always provides resources for the best users of the channel. It can maximize the system throughput, but it cannot guarantee the fairness between cell users. The PF algorithm considers the ratio of instantaneous rate and long-term average rate when selecting users. It adjusts different users by using weight values to achieve the purpose of consideration of the overall throughput of the system and fairness of users. However, it does not consider quality of service (QoS) information.¶

The explosive growth of data rate and capacity demand, as well as the large-scale, high reliability, low delay, and other differentiated demands, have brought about the development of 5G. Therefore, in the face of different industrial scenarios and different QoS requirements of businesses, a highly reasonable multipriority scheduling scheme needs to be designed. Under the condition of limited wireless resources, a reasonable scheme allocates wireless resources for different 5G slices to meet the service requirements of high-priority services in intelligent manufacturing plants and improve the resource utilization rate and fairness among users as much as possible.¶

This section presents a multipriority resource scheduling method for industrial field data to improve resource utilization as much as possible and thereby meet the service quality required by different industrial field services under different 5G slice scenarios.¶

3.2. Uplink scheduling scheme flow

The general flow of the uplink scheduling scheme based on 5G slice is as follows:¶

Step 1: During a scheduling cycle, determine if the task cache queue is empty. If it is null, then wait for the next scheduling cycle; otherwise, proceed to the next step.¶

Step 2: Use the interslice scheduling algorithm to allocate resources to the three network slices according to requirements.¶

Step 3: For the resources obtained from each slice, perform resource scheduling for each user in the slice. Upon completion, wait for the next scheduling cycle.¶

3.3. Interslice scheduling

For Step 2 of the uplink scheduling scheme based on 5G slice, interslice resource scheduling needs to meet the following:¶

1. The resources obtained from different 5G slices are isolated and independent in the frequency domain, and they can be adjusted flexibly. The congestion of one 5G network slice does not affect the other 5G network slices.¶

2. By allocating spectrum resources with good channel conditions to a high-priority slice, the throughput of the system and the service guarantee for high-priority services can be improved.¶

Assume that the number of resource blocks (RBs) that the scheduler can configure is "Q" and that at the time t of scheduling, the total number of users requesting resources is N. In this work, the priority of each slice in each RB is defined in formula (1):¶

          P(i,j) = R(i,j)                                (1)

where P(i,j) represents the priority of the i-th slice at the j-th RB in one scheduling cycle. The greater P(i,j) is, the higher the scheduling priority of the i-th user in the j-th RB will be. In improving the system throughput, the priority is based on the rate and calculation of all users in the j-th RB in the i-th slice. R(i,j) represents the rate sum of all users in the j-th RB in the i-th slice. According to the calculation, we can obtain the priority matrix.¶

For the service requirements of different businesses using industrial field data, the uRLLC slice needs low delay and high reliability, such as those exhibited by a real-time remote cooperative robot, which needs to prioritize the allocation of resources and reduce queuing delay. The eMMB slice has high data volume business requirements. Hence, the priority allocation of resources will improve the overall network throughput. However, its delay requirements are not as high as those of the uRLLC slice. The mMTC slice has the lowest scheduling priority, and in most cases, the amount of uplink data is not large, and the delay requirement is low. Therefore, given the order of the uRLLC slice, eMMB slice, and mMTC slice, RB resources are configured according to the priority matrix.¶

According to the resource scheduling requirements of slices, the final interslice resource scheduling scheme is as follows:¶

Step 1: Calculate the priority matrix according to formula (1).¶

Step 2: According to the priority order of slices, select the i-th slice for resource scheduling, and initialize i to 1.¶

Step 3: Select the i-th slice and the j-th RB with priority ranking, and initialize j to 1.¶

Step 4: Determine whether the currently scheduled RB is adjacent to the RB allocated by the slice. If yes, then perform step 5. Otherwise, set j = j + 1 and repeat step 3.¶

Step 5: Assign priority j-th RB to the i-th slice and remove the RB from the RB sequence.¶

Step 6: Determine whether the resource request for the i-th slice has obtained enough resources. If so, then perform step 7. Otherwise, set j = j + 1 and repeat step 3.¶

Step 7: Determine whether the RB sequence is empty or whether all slices have obtained sufficient resources. If so, then end slice resource scheduling. Otherwise, execute i = i + 1 and repeat step 2.¶

3.4. Intraslice user scheduling

After resource scheduling among slices, the slices obtain their respective continuous and isolated RB groups. It does not interfere with the intraslice user scheduling.¶

The user scheduling in the slice can be understood as a logical cell, and the scheduler conducts resource scheduling on the users belonging to this cell through the resources obtained by intraslice scheduling. Given the different QoS requirements of data in the 5G industrial field, the performance indicators that need to be comprehensively considered in the process of user scheduling in the slice include transmission rate, delay demand, packet loss rate, and the amount of data to be transmitted.¶

The priority of user i in the j-th RB group at time t is calculated by formula (2):¶

        P(i, j) = -log(p(i))/Td(i) * r(i, j)/R(i) * d(i)/D  (2)

p(i) represents the maximum rate of packet loss of user i, i.e., p(i) in (0, 1). Therefore, -log(p(i)) indicates that the lower the maximum packet loss rate is, the higher the priority of user will be. Td(i) represents the maximum wait delay for user i. The smaller Td(i) is, the higher the user priority will be. r(i, j) represents the instantaneous transfer rate of the i-th user in the j-th RB group during the t scheduling cycle. R(i) represents the average transmission rate before the i-th user. The higher the instantaneous transmission rate is, the better the channel quality condition is, and the higher the priority is. d(i) represents the amount of data that user i is waiting to send at time t. The higher the value of d(i)/D is, the higher the proportion of the pending business volume of user i in the total business volume of all requesting users at time tis, and the higher the priority is.¶

Therefore, the intraslice user scheduling scheme is as follows:¶

Step 1: Complete the interslice scheduling.¶

Step 2: For all RBs of a single slice, they are divided into RB groups of the same size according to the number of slice users.¶

Step 3: Calculate the priority of each user in the slice on each RB group according to formula 2.¶

Step 4: Assign the RB group with the highest priority to each user in turn according to user priority.¶

Step 5: Determine whether the RB sequence is empty. If so, then end the resource allocation. Otherwise, repeat step 4.¶

3.5. Summary of this chapter

The scheme proposed in this section has the following advantages:¶

1. It analyzes the different service requirements for industrial field data in a 5G environment. The interslice scheduling scheme is used to complete the resource allocation of three 5G slices to ensure the flexible scheduling and isolation of resources between slices. The scheme also ensures that the required resources for high-priority businesses, such as the uRLLC slice business, are allocated and improve the throughput of the system.¶

2. The intraslice scheduling comprehensively considers the data service transmission rate, delay demand, packet loss rate, data volume to be transmitted, and other performance indicators. It also gives priority to the scheduling of users with good channel conditions, high delay requirements, high-reliability requirements, and large data volume to be sent.¶

4.1. Flow of collaborative task scheduling algorithm

On the basis of the complex industrial site environment, task requirements, and the different amounts of calculation, we take on a new perspective as we comprehensively consider computing resources and user satisfaction through edge computing technology. The goal is to perform tasks between the terminal and edge server resource scheduling problem. In the case of meeting the minimum resource requirements, user satisfaction can also be guaranteed to meet the real-time task processing generated in a variety of industrial production processes.¶

According to the actual needs of industrial field task data, we considered and provided a collaborative scheduling algorithm for industrial field big data tasks based on edge computing. The steps are as follows: the terminal publishes the service, and the scheduler obtains the service information and calculates the number of tasks and delay requirements. According to the task delay requirements, a task's urgency and priority are evaluated. Subsequently, whether the task should be offloaded to the edge server is determined according to the current bandwidth resources. Meanwhile, the task cache status of the scheduler and the computing information of each edge server are obtained. On the basis of the system resource status, number of queued processes, current business task volume, delay requirements, etc., the reasonable task scheduling of the server and terminal is performed. The process is repeated until all tasks are allocated and executed.¶

The specific steps are as follows:¶

Step 1: The terminal publishes the service, and the scheduler obtains the service information, such as the number of calculation tasks N(i) and the delay requirements T(i) (i with (1,N)).¶

Step 2: The task priority LEVEL(i) (i with (1,N)) is evaluated according to the time delay requirements of the task. It exerts an impact on the subsequent task scheduling and further reflects user satisfaction.¶

Step 3: The current network link information, that is, the remaining bandwidth of the network R, is obtained. Assuming that the upload rate of the task is α, the task upload needs to meet. If the current bandwidth is not enough, then the task needs to be unloaded, and the waiting delay is recorded.¶

Step 4: The task scheduling threshold of the scheduler is obtained, along with the calculation information of each edge server, the number of tasks queued, and the queue's waiting delay.¶

Step 5: The reasonable task scheduling of the server and terminal is performed according to the state of system resources, current business task volume, and delay requirements.¶

4.2. Scheduling strategy process

The following is the processing flow of the scheduling strategy.¶

We consider the task delay requirements, the remaining capacity of the terminal, the remaining computing capacity of the edge server, and the total delay of the task assigned to the terminal as the input of the machine learning algorithm. The results of the former calculation are then fed to the fully connected network layer, and the output layer is maximized through the fully connected layer. The softmax layer estimates the probability of assigning to the terminal or the edge server. Therefore, the internal parameters of the network are learnable parameters so that it can perform adaptive adjustment to provide a basis for subsequent optimization on the basis of the customized loss, system resource conditions, network load, and optimization goals. The number of iterations can be determined accordingly. The updated parameters are used to allocate real resources to the tasks in the current scheduler.¶

4.3. Summary of this chapter

Relative to the existing task resource collaborative scheduling algorithm in the industrial field, the algorithm proposed in this work has the following advantages:¶

It can realize the reasonable scheduling of industrial field tasks in terminal equipment and edge servers, fully consider the system's computing resources, improve the system's ability to process tasks, and minimize the resource consumption of the system. It can also avoid calculation delay caused by an unbalanced resource allocation. The algorithm considers the execution status of the system, the task calculation amount, and the delay requirements for optimal scheduling when performing task scheduling. It also comprehensively considers the construction of new optimization goals to achieve system resource utilization efficiency and user satisfaction.¶

5.1. Physical security requirements

For edge computing equipment, security problems are caused by indirect or self-inflicted causes during operation (e.g., energy supply, cooling and dust removal, and equipment loss). Although threats to operations are not as devastating as the damage caused by natural disasters, the lack of a good response will still lead to disastrous consequences, resulting in the performance degradation of edge computing, service interruption, and data loss. Particularly in the Industrial Internet scene, factories conduct sophisticated equipment maintenance and overhaul, but dealing with the operation and maintenance of IT equipment timely is difficult.¶

5.2. Network security requirements

Relative to cloud computing data centers, edge nodes have limited capabilities and are highly vulnerable to hackers. The damage of a single edge node is not extensive, and the network can quickly find alternative nodes nearby. However, if hackers use the compromised edge nodes as "broilers" to attack other servers, then they could affect the entire network. Most existing security protection technologies have complex computational protection processes, which are not suitable for edge computing scenarios. Therefore, an important network security requirement is to design lightweight security technology suitable for edge computing architecture in the Industrial Internet scene.¶

5.3. Data security requirements

In edge computing, users outsource data to edge nodes and transfer the control of data to them. The process introduces the same security threats as cloud computing. First, ensuring the confidentiality and integrity of data is difficult because the outsourced data may be lost or modified incorrectly. Second, unauthorized parties may misuse the uploaded data to seek other benefits. Relative to the cloud, edge computing avoids the long-distance transmission of multiple routes and greatly reduces the outsourcing risk. Therefore, the security problem of data belonging to edge computing is increasingly prominent. For example, in such a complex and changeable environment, the safe and rapid migration of data after the collapse of an edge node should be realized.¶

5.4. Application of security requirements

Application security, as the name implies, guarantees the security of application processes and results. In the era of marginal big data processing, applications can be guaranteed to get short response times and high reliability by moving application services from cloud computing centers to network edge nodes. Meanwhile, network transmission bandwidth and intelligent terminal power consumption can be greatly reduced. However, edge computing suffers from common application security problems in information systems, such as the denial of service attack, unauthorized access, software vulnerability, abuse of authority, and identity impersonation. Moreover, it has other application security requirements because of its characteristics. In the scenario in which multiple security domains and access networks coexist at the edge, managing user identity and realizing authorized access to resources become important in ensuring application security.¶

8.1. Normative References

[RFC2119]

Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, <https://www.rfc-editor.org/info/rfc2119>.

8.2. Informative References

[Delay-optimal]

Liu, J. and Y. Zhang, "Delay-optimal computation task scheduling for mobile-edge computing systems", IEEE International Symposium on Information Theory (ISIT) pp.1451-1455, July 2016.

[Energy-Efficient]

Wang, J. and J. Tang, "Towards energy-efficient task scheduling on smartphones in mobile crowd sensing systems", Computer Networks pp.115,100-109, 2017.

[QoSBased-Distribution]

Song, Y. and S.S. Yau, "An approach to QoS-based task distribution in edge computing networks for IoT applications", IEEE international conference on edge computing (EDGE) pp.32-39, July 2016.

[Scheduling-Latency-Sensitive]

Scoca, V. and A. Brandic, "Scheduling Latency-Sensitive Applications in Edge Computing", In Closer. pp.158-168, Scheduling Latency-Sensitive , 2018.

[Spanedge]

Sajjad, H. P. and K. Danniswara, "Spanedge: Towards unifying stream processing over central and near-the-edge data centers", IEEE/ACM Symposium on Edge Computing (SEC) pp. 168-178, October 2016.