Software defined network (SDN) and network function virtualization (NFV) have become a new paradigm of a new generation of network architecture. SDN and NFV can effectively improve the flexibility of deploying and managing service function chains (SFCs). By combining SDN and NFV and applying them to the resource orchestration problem of SFC deployment, the three-tier architecture consisting of SDN controller, network function virtualization and physical underlying computing resource layer in the process of heterogeneous network resource mapping is considered. And an optimization algorithm for active control resources based on SDN and NFV is proposed. Firstly, the user’s utility is modeled by the multi-standard aggregated multi-criteria utility algorithm, and the optimization goal is transformed into the problem of maximizing the user’s utility. Then the controller, based on the algorithm’s prediction of the future state and real-time monitoring of the network utilization, makes decisions and issues control commands for the arriving SFC requests, based on which it occupies the underlying resources held by the virtualized network function (VNF). The simulation results show that, compared with the static timing resource allocation algorithm, the active control resource deployment algorithm proposed in the article has better performance in terms of resource utilization, acceptance rate, and user creation utility.
In order to adapt to the explosive growth of mobile data traffic and a large number of new applications and business needs, operators have deployed large-scale network infrastructure to provide users with ubiquitous network services. At the same time, different access networks such as wired, cellular, and WLAN coexist in the network space around us. On the one hand, huge traffic and heterogeneous characteristics bring difficulties to network management. On the other hand, as the available bandwidth of the network increases, the performance requirements of users for network services are becoming more and more acute. How to design a solution that can satisfy users’ needs for network-perceived network quality [
Nowadays, SDN and NFV have been widely used in the research of network resource orchestration. With the help of SDN technology, network equipment has changed from a closed mode to an open general equipment mode, which separates the control plane and the data plane, which allows the network to be programmed through open interfaces [
At present, the research on network resource mapping and orchestration algorithms for the integration of SDN and NFV is mainly focused on a certain network resource, and the optimization direction is mostly to optimize the overall network resource utilization efficiency and reduce network costs [
This paper combines SDN and NFV technologies to further abstract the network, which can form an end-to-end and logically isolated network, and realize a user-centric framework. A user agent corresponding to each user can be generated on a virtual device with connection information retention and user situation awareness capabilities. Through the user agent, the ubiquitous and undifferentiated network connection of users in a heterogeneous network can be realized. User preference feature perception, through flexible resource scheduling management to ensure the maximum utility provided to users. The main contributions of this paper are summarized as follows:
Based on the three-tier architecture composed of SDN controller, network function virtualization, and physical underlying computing resource layer in the process of heterogeneous network resource mapping [ Propose a dynamic active control resource deployment algorithm based on request priority and user preference, including two stages: network state prediction and control based on load utilization. The controller makes corresponding decisions on the SFC request and issues control commands based on the real-time monitoring of the utilization rate predicted by the future state. Based on this, the VNF occupies the resources held by the physical nodes to achieve the purpose of active control and maximum utility. A series of simulation experiments are used to prove the effectiveness of the proposed algorithm. The simulation results show that in a congested network scenario, the acceptance rate of high-priority requests and the overall user utility benefit are significantly improved. At the same time, this method does not reduce the utilization rate of the network, and ensures the efficiency of resource utilization.
Under the business of virtual network resource allocation problem, different users access corresponding service providers (SP) to obtain network services according to business needs. SP constructs corresponding virtual network requests according to business needs, and then based on infrastructure providers (InP) physical network resource constraints, using resource allocation algorithms to allocate suitable physical networks to carry the virtual network [
The underlying infrastructure network is represented by an undirected graph G = (N, L), where
After the underlying physical network is virtualized, it is constructed by multiple physical nodes that can provide physical resources. These physical nodes provide VNFs with various types of resources such as CPU, memory, and disk. At this time, each VNF in the service function chain can be deployed on any physical node, and one physical node can provide services for multiple VNFs located in the same service function chain [
The amount of data processed by each VNF determines its own resource requirements. In order to reduce the complexity of the algorithm, this model unifies the multiple resources (CPU, memory, disk) that can be provided by general-purpose processor nodes into computing resources. The virtual nodes in different virtual requests can be mapped to the same physical node. For each node n(n ∈ N), its available resources are:
The underlying network can provide VNFs of different protocol layer types. Let
The SFC request is composed of multiple VNFs, ingress and egress nodes with order constraints [
When the SFC service time exceeds the life cycle, the service ends and the allocated resources are reclaimed [
Therefore, the number of VNFs deployed in the k-th service function chain can be expressed as:
For the service function chain
In microeconomics, utility refers to the ability of goods or services to meet human needs. A related term is utility function, which relates to the utility that consumers obtain from goods or services. Different consumers with different user preferences will have different utility values for the same product. The basic idea of the utility theory is that the decision maker will always choose the decision with the greatest utility value.
In a heterogeneous network, for the user side, the utility function is defined as
For the single-criteria utility function, select the Sigmoid function:
The utility function used in the early research work is the logarithmic utility function, which is a good approximation of the utility value of the wired communication network. In recent years, the demand for wireless adaptive real-time applications has continued to increase, and the utility function that approximates real-time applications is the sigmoid function [
Obviously, the value of
It can be defined as follows:
The upper limit
As for the total benefits provided by the allocation of SFC computing resources for a single user to the user, it is necessary to consider the priority of the user’s request and the pickiness of resources to calculate the benefits. Intuitively, high-priority has a larger slope and lower fault tolerance, and a low-priority has a smaller slope and higher fault tolerance. The criteria can be grouped into two types. The timeliness of acceptance of VNF requests and resource allocation in the service function chain can be regarded as a criterion. Whether the user’s criticality of the accessed network resources is met in a timely manner can also be regarded as another criterion. Therefore, it is stipulated that if the request is accepted within the time unit, or the resource criticality of a single user’s SFC access is satisfied (for example, the preference of cellular network access for security considerations, or the preference for WIFI resources based on cost performance), then the corresponding criterion obtains the maximum benefit, that is, the independent variable of
Otherwise, the corresponding benefit will be lost according to the amount of elapsed unit time, that is,
The above
Therefore, if the long-term realized total user benefit is calculated, it is necessary to perform cumulative calculation to allocate the total benefit of k SFC computing resources.
The goal of this article is to fully consider user preferences and the priority of requests for SFC resource allocation. Calculate user request priority and resource criticality according to user characteristics to obtain utility benefits, thereby establishing a user-centric architecture. Therefore, the optimization goal of this paper is modeled as:
C1 is the user utility calculation coefficient constraint. The calculation of the multi-standard aggregate utility
The algorithm in this paper fully considers the decision factors including priority and resource preference when the controller receives the SFC request to allocate the actual resources for each VNF component, and determines the resource allocation plan within the time slice. In this way, a control instruction is issued to the user agent generated at the target node. The whole process is mainly composed of three parts: status collection, decision-making and issuing instructions.
Before making any decision, the controller will collect the underlying network status information, calculate and update the available resources of each node, and collect user preferences. Then, the controller will execute the decision-making component, which determines the virtual resource allocation plan within each time slice. Once the allocation plan is determined, the controller will issue a control instruction to the user agent, and the user agent will occupy the underlying virtual resources held by the VNF according to the control instruction of the controller.
The input of the algorithm includes: the received VNFR, user characteristics (high and low priority) and resource preference; thus output: whether to accept the request and allocate the actual computing resources, including the selected physical nodes and the resources provided by them. During this period, monitor the information on the successful occupation of node resources, and update the network status in time; monitor the information on the release of node resources in the network, and when the service is terminated, the allocated resources occupied by the service function chain are recovered.
Active control resource deployment based on request priority and user preference is the main focus of this algorithm. Assuming that the controller knows the average arrival rate of SFC requests, it can predict the next state of the network by considering the arrival probability of the newly added SFC in the next time unit. Then, according to the prediction of the next state, the resource utilization rate of the next time unit is calculated. Based on the comparison between this value and the network utilization threshold, the algorithm selects and decides whether to temporarily not process a low-priority request. Therefore, the active control resource deployment algorithm proposed in this paper includes two stages: network state prediction and control based on load utilization. The flow chart of the SFC deployment algorithm is shown in Network state prediction: At this stage, the controller estimates the next state of the network according to the arrival probability of the upcoming request in the next time unit [ Control based on load utilization: Calculate the resource
where ρ represents the probability of arriving at a new request in the next time unit, and the first term on the right side of the equation represents the arrival of a new request, and the needed resources are allocated according to the average amount of requested computing resources. The second term on the right side of the equation represents that the network status does not change: the probability that there is no request in the next time unit (1-ρ). The left side represents the predicted next state (next time unit) of the underlying network.
In the formula, U is the ratio of network allocation resources to total resources. If the percentage of network utilization is greater than the pre-selected threshold, it is possible that in the next state, the upcoming high-priority VN request will be rejected. Therefore, as will be mentioned below, in this case, the proposed active control resource deployment algorithm will actively not process a low-priority VNFR temporarily.
Since the algorithm solves the problem of VNF deployment and orchestration under dynamic changes in business requirements, the value will only change when the request is accepted and the actual computing resources are allocated. The essence of dynamic resource changes is the arrival of new requests and the release of allocated resources. When new requests arrive in each time slot, the deployment decision of the algorithm will be triggered. The dynamic changes of available resources can be captured by the controller in real time, based on which the next decision can be made.
After receiving a new request in this time slot, the algorithm will sequentially traverse the SFC request queue within the time unit and traverse the VNFs in the service chain in order. By evaluating the types of resources required by each VNF, a series of candidate nodes that can be deployed are selected. If the remaining available resources of the candidate node are sufficient to provide the mapping requirements of the VNF, the calculation resource pre-mapping is performed, that is, the process of assuming that the request is accepted and calculating the possible network resource utilization rate in the future. If the prediction result shows that even if all the requests are accepted, the network resource utilization rate of the next time slot will not exceed the threshold, then the actual resource allocation is performed. If it is found that the future utilization rate may exceed the preset threshold, select a low-priority SFC to temporarily not allocate available resources for it, and mark the request so that it will be prioritized in the next competition of the same level. This is to prevent the same or several requests from being discarded consecutively, which will not meet expectations. After the resource allocation is over, the algorithm updates the network state and starts to process the SFC request of the next time slot, and starts to traverse the newly arrived request queue.
In order to evaluate the method proposed in this article, we chose to use Python 3.6.9 for simulation and run on an Ubuntu 18.04 virtual machine with 4.0GB RAM and a 2-core processor. As shown in formula 17, the simulation assumes that the arrival rate of VNF requests follows the Poisson distribution
And all the simulation data are shown in
Parameter | Value |
---|---|
Number of nodes | 12 |
Types of VNFs that can be deployed | 2 |
Computing resources provided by the node | |
Number of VNF types | 3 |
VNF resource requirements | |
Data processing demand factor | |
SFC life cycle | 180 |
High priority request arrival rate | |
Low priority request arrival rate | |
Total time | 1500 |
In order to verify the performance of the proposed active control resource deployment algorithm, the following four performance indicators are considered:
Acceptance rate of high-priority VNF requests: According to the purpose of resource management in this article, in order to increase the number of high-priority VN requests accepted in the network, we calculated the ratio of accepted high-priority VN requests to total high-priority VN requests. Resource utilization: The average ratio of the sum of the used resources of the network to the total available resources is measured. User utility: The algorithm in this paper aims to meet the needs and preferences of users as much as possible, so as to maximize the network’s service quality for users. The measurement of user utility can intuitively reflect the improvement of the algorithm to the user’s perception quality and the advantages of the algorithm. Utility load ratio: In order to better evaluate the utilization value of the algorithm to the unit load, the ratio between utility and load intensity is calculated and measured in real time.
As shown in
Based on SDN and NFV technology, this paper abstracts the network to form an end-to-end and logically isolated network, realizes a user-centric framework, and proposes an active control resource deployment algorithm that maximizes user utility. In the process from receiving the SFC request to allocating actual resources for each VNF component, the controller takes the lead to consider the SFC demand on a user basis, and fully considers decision factors including priority, resource preference, and network prediction. Through flexible resource scheduling management to ensure the maximum utility provided to users. The simulation results show that, compared with the strategy of passively accepting VNF requests and allocating resources, our proposed solution performs better, can use network resources more effectively, and create greater value for network users.
This work was supported by the National Natural Science Foundation of China (61871058).