RESEARCH ON 5G NETWORK RESOURCE ORCHESTRATION ALGORITHM BASED ON NETWORK VIRTUALIZATION TECHNOLOGY

 

Y.F. WEI†, Y.N. JIA†, J. LI†   and   X.J. WANG‡

† Beijing Key Laboratory of Work Safety Intelligent Monitoring, Beijing University of Posts and Telecommunications, Beijing, 100876, China. Email: weiyifei@bupt.edu.cn

‡ Dublin City University, Dublin, Dublin 9, Ireland. Email: xiaojun.wang@dcu.ie

Cite this article as:

Wei, Y.F., Jia, Y.N., Wang, X.J. (2022) “Research on 5g network resource orchestration algorithm based on network virtualization technology”, Latin American Applied Research, 52(4) pp 346-352.

Abstract-- With the advancement of wireless network technology to the next generation, network function virtualization (NFV) brings the advantages of centralized scheduling of virtual wireless resources and the ability to orchestrate virtual network functions. This makes it possible to dynamically deploy and manage service function chains (SFCs) in virtualized wireless networks according to changes in network load. Through the network function virtualization technology, the DU/CU separation architecture under 5G-NG-RAN is considered. Aiming at the resource orchestration problem of the 5G access network virtual protocol stack function and SFC after deployment, an energy-aware virtualized network function instance (VNFI) orchestration algorithm is proposed. The algorithm decouples the decision-making process of VNF migration into two specific modules, namely "the VNF selection module" and "the migration destination node selection module". It is verified by simulation that compared with other energy-saving strategies, the algorithm in this paper has better performance in reducing energy consumption and reducing migration loss, and it can take into account both the reduction of the number of open server nodes and the improvement of resource utilization.

Keywords-- NFV; 5G access network; SFC; resource orchestration; energy-aware

I. INTRODUCTION

With the rapid growth of the number of 5G diversified services, various network differentiated performance requirements have also followed, which has led to the emergence of a 5G NG-RAN technical architecture based on network function virtualization. The 5G NG-RAN architecture still retains the characteristics of centralization and collaboration. At the same time, the BBU will sink part of the physical layer to AAU, and the rest will be reconstructed into a two-level architecture of CU and DU, and implemented by high-performance general-purpose server nodes. It has the advantage of flexible arrangement of resources (Kumar et al., 2020).  The rapid growth of network traffic requires higher and higher data processing capabilities of the DU/CU cluster under the NG-RAN architecture, so the number and energy consumption of DU/CU cluster servers have also increased sharply (Gupta et al., 2020). How to reduce the operating energy consumption of the overall general-purpose server in the infrastructure has become a problem that needs to be solved while ensuring the utilization of network resources.

At present, research on energy-saving algorithms for network resource mapping and orchestration based on NFV technology mainly focuses on server resource integration and the reallocation of VNF instances (Keating et al., 2019; Wei et al., 2018; Soualah et al., 2017), without considering more flexible orchestration methods and updated application scenarios such as 5G access networks. The algorithm also lacks consideration of the loss and end-to-end delay requirements caused by the VNF migration process. Therefore, based on the NG-RAN architecture of the 5G access network, this article models the deployment of virtual protocol stack functions in the SFC, and proposes an Energy-Aware VNF Instance orchestration (EVNFIO) algorithm. In this way, the VNF orchestration problem under the dynamic change of business demand in the wireless virtualized network is optimized. The main contributions of this paper are as follows:

(1) Based on the DU/CU separation architecture under the 5G NG-RAN architecture, the energy consumption of the network bottom layer and each node is modeled. Fully consider factors such as resource type, tolerable delay, and migration loss.

(2) Propose a VNF orchestration algorithm based on energy perception, including two core modules: the VNF selection module and the migration destination node selection module.

(3) A series of simulation experiments are used to prove the effectiveness of the proposed algorithm. The simulation results show that energy consumption, VNF migration loss and the number of open nodes all decrease, and resource utilization increases.

II. SYSTEM MODEL

In the NFV architecture, the virtualized wireless access network consists of infrastructure provider (InP) and mobile virtual network operator (MVNO). As shown in Fig. 1 below, the InP has physical infrastructure and wireless resources to provide MVNO with the physical resources needed for virtualization services. The virtual resource orchestrator is responsible for performing all related management and orchestration (MANO) operations, and VNF embedding and life cycle management. With the dynamic changes of resource utilization on general server nodes in the infrastructure, if you blindly accept requests and allocate corresponding computing resources, it will lead to the increase of overloaded and underloaded

Figure 1: Schematic diagram of 5G NG-RAN network virtualization architecture.

nodes, which will increase energy consumption. There fore, VNF migration technology is required to manage the VNF on each node (Wei et al., 2018).

A. Underlying Network Model

The underlying infrastructure network is represented by a weighted undirected graph, expressed as: , where  represents the general server node, and  represents the link between nodes. The resources on the general server include CPU resources, memory resources, and hard disk resources. In order to reduce the complexity of the algorithm, these three resources are unified into the computing resources of the node for consideration, and the computing resource capacity of the general server node  is expressed as , the remaining available resources are expressed as , the link between node  and node  is expressed as , the link capacity is expressed as , and the available bandwidth resource is expressed as .

Set  represents the set of VNFs, where  represents the type of VNF. It should be noted that each node may not provide all types of VNFs. Therefore, it is assumed that each VNF type has a set of nodes to be deployed (Wei et al., 2018). Binary variable  indicates whether node n can deploy :

          (1)

The SFC request is composed of multiple VNFs, ingress and egress nodes with order constraints. Figure 1 shows the deployment diagram of the SFC. Let represent the set of SFC requests,  represents the SFC request, where  represent the entry and exit nodes, respectively, and  represents the life cycle of the SFC (the occupancy time of resources).

When the SFC service time exceeds the life cycle, the service ends and the allocated resources are reclaimed.  represents the ordered VNF sequence of SFC, where . Finally, define the binary variable  to represent the mapping relationship between VNF and physical nodes.  If  is deployed on node n, then ; otherwise, .

             (2)

Therefore, the number of VNFs deployed in the k-th SFC can be expressed as:

. (3)

For the SFC , the data processing capacity of the VNF composing the SFC is expressed as , the  computing resource demand and its data processing capacity are defined as a linear relationship, and the correlation coefficient (data processing demand factor) is expressed by , Then the computing resource demand  of the VNF can be expressed as:

                 (4)

Following the above, = can be used to qualitatively express the m-th virtual network function  of the k-th functional service chain, where  is the life cycle of the instance,  is a migration status identifier that records whether the current VNF has deployed node migration, and  is the loss caused by the migration.

At the same time, the end-to-end delay of VNF migration and deployment also needs to be considered. The end-to-end delay of the SFC is composed of processing delay and transmission delay, which can be expressed as:

In the formula,  represents the transmission delay between nodes  and ,  represents the processing delay of the VNF, and  represents the number of hops between nodes  and , so the end-to-end delay of the k-th SFC is expressed as:

   (7)

B. Node Energy Consumption Model

In the infrastructure, the state of general-purpose servers is divided into two types: running state and sleep state. The general-purpose server in the running state is responsible for allocating resources for the VNF deployed on it to ensure its normal operation, and if no VNF is deployed on the general-purpose server, servers will be set to sleep mode to save energy.

When the service load becomes larger, may cause the resource utilization of the server in the running state to exceed the upper limit (Luo et al., 2020). It is necessary to "wake up" a part of the general-purpose servers that are in a sleep state into a running state, and migrate some of the VNFs deployed on overloaded nodes to it to ensure the performance requirements of network services; and when the business load becomes smaller, VNF on nodes with low utilization rate is migrated to other nodes with sufficient resources, and the nodes sleep to reduce energy consumption.

Therefore, the current operating state of the node , that is, sleeping or on, can be represented by the binary variable :

               (8)

Since the focus of NFV deployment in SFC is to implement migration strategies based on changes in network node load, it is particularly important to evaluate the load status. The measurement of server node load level is positively correlated with the size of node resource utilization. Then the computing resource utilization rate of the general server node  can be expressed as:

.                                         (9)

Among them, the utilization rate  is expressed as the ratio of the sum of the resources occupied by all the VNFs mapped on the server node  to the upper limit of the node resource capacity.

Define  as the load upper limit of node n. If the computing resource utilization of node  meets , the node is determined to be an overloaded node; define  as the node's lower load limit, if the computing resource utilization of node n meets  , The node is judged to be an underload node; the rest are regular nodes.

A running general-purpose server will generate a certain amount of energy consumption, and the power consumption of a general-purpose server that is turned on can be expressed as:

                                   (10)

 is dynamic energy consumption, which is due to the load energy consumption of the node occupied by computing resources;  is static energy consumption, which is a small amount of energy consumption inevitably generated by nodes in a running or sleeping state. Therefore, the total energy consumption  of each general-purpose server in the infrastructure is expressed as:

    (11)

At the same time, it should be considered that VNF migration is not without cost. The migration of VNF instances may cause jitter and partial data loss, which will bring certain retransmission costs and time loss. The cost paid is also positively related to the amount of data that the node needs to process. Therefore, if the unit price of data loss is assumed to be , and consider the above-mentioned  to record whether the VNF Migration has occurred, and the loss  caused by the migration of a specific VNF. Then the total migration decision loss can be expressed as:

                         (12)

C. Optimization Objective

The goal of this article is to achieve the goal of making full use of network resources and minimizing network energy consumption based on the above-mentioned network underlying physical model and node energy loss model. Therefore, the optimization goal of this paper is modeled as:

s.t.

                                                  (13)

Constraint C1 restricts VNFs that can only be mapped to one node in SFC deployment. C2 means that the computing resources occupied by all VNFs deployed on a node must not exceed the total computing resources provided by this node. And because of the binary variable  Indicate whether the node can deploy , C3 ensures that each node can support at least one type of VNF for deployment, and C4 ensures that each type of  can be deployed. C5 means that the bandwidth resources of each link in the network will not exceed the upper limit. C6 indicates that the computing resource utilization of the general-purpose server node in the running state does not exceed the upper limit of computing resource utilization. C7 indicates that the delay of each SFC does not exceed the upper limit, that is, the end-to-end delay of the SFC must not exceed its delay limit.

III. ALGORITHM DESIGN
A. Algorithm Description

The algorithm first evaluates the load status of each server node in the resource pool by calculating node resource utilization, and uses a preset load threshold to determine whether to trigger the dynamic migration of VNF, and aims to minimize the total energy consumption of the system. The VNF migration loss and SFC end-to-end delay limits are considered.

The execution of the algorithm can be represented as two modules, the VNF selection module and the migration destination node selection module. The former sorts out the set of VNFs to be migrated according to the policy as input and calls the latter to select the destination server node.

B. Algorithm Implementation

The input of the algorithm is the set of physical nodes in the infrastructure, the upper and lower thresholds of resource utilization, the set of SFCs deployed on the infrastructure, and the computing resource requirements of the corresponding VNFs; the final output is the new service functional chain deployment solution after VNF migration. During this period, monitor the information about the successful occupation of node resources, and update the network status in time; monitor the information about the release of node resources in the network, and when the life cycle is reached and the service is terminated, the allocated resource occupied by the SFC is recovered.

Figure 2: VNF selection module in the algorithm

The triggering of the migration algorithm and the operation of the VNF selection module in the algorithm are shown in Fig. 2. First, calculate the node computing resource utilization rate based on Eq. (9) and check if it exceeds the upper threshold  or is less than the lower limit , the node is marked as overloaded node or underload node accordingly.

Maintain an ascending queue for storing overloaded nodes and a descending queue for storing underloaded nodes based on the resource utilization rate. An empty set S is created and maintained to store VNFs that need to be moved out and the source node they belong. Each element in the set exists in the form of key-value pairs, the key stores the source node, value stores the VNFs migrated from the node, like . According to related research (Beloglazov et al., 2012), when the resource utilization of the server node exceeds a certain threshold, the virtual opportunity running on the server node will occur to a certain extent. In order to give priority to nodes with lower service performance, the overload queue is processed first and selected from the end of the queue first. Select one or more VNFs to migrate out from the node to ensure that the computing re-

Figure 3: The migration destination node selection module

source utilization of the node drops below .

For the nodes whose computing resource utilization is lower than the lower threshold , the algorithm takes out the nodes in sequence from the descending queue storing the underloaded nodes and puts them into the set S. Different from overloaded nodes, in the subsequent invocation process, it is necessary to determine whether all the VNFs running on these underloaded nodes can be migrated to other nodes with abundant resources. If it can be achieved, the status of the idle node where the deployed VNF has been migrated is changed from on to sleep. And migrating the VNFs to be migrated from the source server node to the target server node, and updating the computing resource usage of the source server node and target server node will be the target of the subsequent modules, as shown in Fig. 3.

In the migration destination node selection module, first traverse the VNFs in value, nodes whose utilization is less than  and whose SFC end-to-end delay does not exceed the limit threshold after migration is what we are looking for. In Fig. 3, if the target cannot be found, the underload node resources will not be released,

Table 1. Simulation data table

Parameter

Value

Number of nodes

100

Node static energy consumption

50

Node dynamic energy consumption

300

Transmission delay between nodes

VNF processing delay

SFC end-to-end delay limitation

Node load upper threshold and lower threshold

Figure 4: Comparison of total power consumption of server nodes.

the node is judged as the first choice of the "destination node to be migrated", so that the utilization rate is greater than the threshold , and the underloaded state is changed to the normal operating state.

IV. SIMULATION RESULTS

To evaluate the method proposed in this article, the simulation platform is built by Python 3.6.9, which runs on an Ubuntu 18.04 virtual machine with 4.0GB RAM and a 2-core processor. At the same time, the simulation results are compared and analyzed with a Least Used Placement Engines (LUPE) algorithm (Laghrissi and Taled, 2018). This energy-saving algorithm always selects the target node with sufficient computing resources and the lowest resource utilization for VNF redeployment. In this paper, the server node parameters and SFC parameters in infrastructure are set by referring to the selection of network size, node power consumption and random data in relevant studies (Qu et al., 2020; Keating et al., 2019; Wei et al., 2018), specific simulation parameters are shown below:

In order to verify the performance of the proposed energy-aware VNF instance orchestration algorithm, the following four performance indicators are considered: power consumption of server nodes, migration loss, computing resource utilization and the proportion of the number of servers in the running state (Sun et al., 2021; Qu et al., 2019)

Figure 4 shows the comparison between the EVNFIO algorithm and the LUPE algorithm in terms of total network power consumption. As the number of deployed SFCs increases, the proportion of nodes in the sleep state decreases, and the total power consumption of the system starts to rise. As the number of SFCs increases, the slope of the curve changes from large to small. This is because when the number of SFCs grad-

Figure 5: Comparison of VNF migration loss.

Figure 6: Comparison of resource utilization.

ally approaches the upper limit of the resources that the cluster can provide, each server node is forced to be used efficiently. As a result, the VNFs that need to be migrated begins to decrease. However, in engineering, the computing resources provided by clusters are often far more than the actual demand (Guo et al., 2022), so it is difficult to approach the upper limit of energy consumption. This shows that when solving practical problems, EVNFIO algorithm has obvious advantages because it will not approach the full load of the cluster.

Figure 5 compares the loss values caused by the migration of virtual network functions using the EVNFIO algorithm and the LUPE algorithm under the conditions of different numbers of SFC deployments. The loss caused by the EVNFIO algorithm is generally less than the comparison algorithm. When the number of SFCs increases, the gap becomes more obvious; when the number is small, the gap is not obvious. This is because the EVNFIO algorithm will migrate the VNFs on the underloaded nodes, which increases a small part of the loss.

As shown in Fig. 6, the changes in the utilization of network resources along with the increase in the number of deployed SFCs are measured. The use of EVNFIO algorithm is higher than the comparison algorithm in terms of resource utilization efficiency, and when the resource demand of SFC is smaller than the sum of resources provided by all nodes, the gap is particularly obvious. This is the advantage of EVNFIO's ability to adjust VNF deployments on underload nodes in a timely manner.

Figure 7 shows the ratio of the number of server nodes in the running state to the total number. The algorithm in this paper aims to reduce energy consumption as much as possible, and implement measures to migrate

Figure 7: Comparison of the proportion of running nodes

and sleep for nodes with low utilization. This controls the proportion of nodes in the running state in the total number of nodes, and the contrast gap in Fig. 7 is indeed very significant. The more resources are occupied by each SFC, the running nodes gradually increase, and the gap between the two is gradually narrowing, which is in line with expectations.

V. CONCLUSIONS

Based on the network function virtualization technology and the current DU/CU separation architecture under 5G-NG-RAN, this paper solves the problem of resource scheduling after the deployment of the SFC. The energy consumption of the bottom layer of the network and each node is modeled, and an energy-aware VNF instance orchestration algorithm is proposed. The algorithm fully considers decision factors such as node utilization, resource type, tolerable delay, and migration loss. The decision-making process consists of two core modules: the VNF selection module and the migration destination node selection module. Through flexible resource orchestration to ensure that the overall power consumption is minimized. The simulation results show that, compared with other energy-saving strategies, proposed solution performs better in terms of reducing energy consumption and migrating losses, and can reduce the number of server nodes that are turned on, use network resources more effectively and conveniently.

ACKNOWLEDGE

This work was supported by the National Natural Science Foundation of China (61871058).

REFERENCES

Beloglazov, A., Abawajy, J. and Buyya, R. (2012) Energy-aware resource allocation heuristics for efficient management of data centers for Cloud computing. Future Generation Computer Systems. 28, 755-768.

Guo, M., Guan, Q., Chen, W., Ji, F. and Peng, Z. (2022) Delay-Optimal Scheduling of VMs in a Queueing Cloud Computing System with Heterogeneous Workloads. IEEE Trans. Services Computing. 15, 110–123.

Gupta, H., Sharma, M., Franklin, A. and Tamma, B.R. (2020) Apt-RAN: A Flexible Split-Based 5G RAN to Minimize Energy Consumption and Handovers. IEEE Transactions on Network and Service Management. 17, 473-487.

Keating, R., Saily, M., Hulkkonen, J. and Karjalainen, J. (2019) Overview of Positioning in 5G New Radio. 16th International Symposium on Wireless Communication Systems. Oulu. Finland, 320-324.

Kumar, N., Rawat, K. and Ghannouchi, F.M. (2020) Multi-Band All-Digital Transmission for 5G NG-RAN Communication. IEEE 3rd 5G World Forum (5GWF) Bangalore. India, 497-501.

Laghrissi, A. and Taleb, T. (2018) A survey on the placement of virtual resources and virtual network function. IEEE Communications Surveys & Tutorials. 21, 1409-1434.

Luo, C., Ji, J., Wang, Q., Chen, X. and Li, P. (2020) Channel State Information Prediction for 5G Wireless Communications: A Deep Learning Approach. IEEE Transactions on Network Science and Engineering. 7, 227-236.

Qu, Z., Huang, Y. and Zheng, M. (2020) A novel coherence-based quantum steganalysis protocol. Quantum Information Processing. 19, 1-19.

Qu, Z., Sun, H. and Zheng, M. (2021) An efficient quantum image steganography protocol based on improved EMD algorithm. Quantum Information Processing. 20, 1-29.

Soualah, O., Mechtri, M., Ghribi, C. and Zeghlache, D. (2017) Energy Efficient Algorithm for VNF Placement and Chaining. 17th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID). Madrid. Spain, 579-588.

Sun, L., Wang, Y., Qu, Z. and Xiong, N.N. (2021) BeatClass: A Sustainable ECG Classification System in IoT-based eHealth. IEEE Internet of Things Journal. August.

Wei, Y., Yu, F.R., Song, M. and Han, Z. (2018) User Scheduling and Resource Allocation in HetNets with Hybrid Energy Supply: An Actor-Critic Reinforcement Learning Approach. IEEE Transactions on Wireless communications. 17, 680-692.

 

Received: January 21, 2022

Sent to Subject Editor:  January 27, 2022

Accepted:  March 28, 2022

Recommended by Subject Editor David Zumoffen