Virtual Network Embedding with Dynamic Speed Switching Orchestration in Fog/Edge Network
In the future 5G networks, network deployment flexibility and low network latency are two of the most critical requirements and issues. Recently, network virtualization and Fog/Edge computing have been proposed as two potential solutions to enable the desired future network environment. This paper investigates the Virtual Network Embedding (VNE) problem in a Multi-access Edge Computing (MEC) architecture, according to the standards proposed by European Telecommunications Standards Institute
... dards Institute (ETSI). We propose an embedding algorithm, called PSO-CSNR, to optimize end-to-end latency constraints in an MEC network. In addition, we adopt Activity on Vertex (AOV) network as our Virtual Network Request (VNR), which is more realistic to real applications. Moreover, we consider the latest processor technologies for substrate nodes, where the CPUs are deployed with asymmetric core frequencies, and propose the second algorithm, called DSS. The DSS can dynamically orchestrate the processing speed of each virtual function, in order to decrease the processing time of virtual functions on virtual nodes, so that the Infrastructure Providers (InPs) can gain more profit in the same amount of time. We then combine the PSO-CSNR with DSS, and refer to it as VNE-DSSO. The simulation results show that the VNE-DSSO algorithm outperforms the other existing algorithms in terms of revenue, acceptance ratio and embedding cost. INDEX TERMS Network function virtualization, virtual network embedding, multi-access edge computing, asymmetric frequency core, dynamic speed switching. VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ 84753 Y. Chiang et al.: VNE With Dynamic Speed Switching Orchestration in Fog/Edge Network (VNs), and each VN offers a customized service in a specific scenario . The VNs consist of several virtualized network functions, called virtual nodes. These virtual nodes are connected by virtual links, which require bandwidth resources. SPs hand these VNs to InPs in a VNR format. The VNRs are then mapped on virtualized substrate resources (e.g. CPU, bandwidth), deployed and maintained by one or more InPs. The network resources are virtualized and isolated from other users on the same physical infrastructure. This increases the network flexibility, and mitigate the ossification of the physical network . Mapping multiple virtual networks onto a given Substrate Network (SN) is a major resource allocation challenge in network virtualization and is usually referred to as VNE problem. Although some previous efforts have been made to design algorithms for the VNE problem with different objectives, most of the studies of the works considered VNE problems in a signal data center environment -, or focused on the offline solutions  . Those VNE methods cannot be carried out in an MEC network environment, which consists of distributed MEC servers in different edge areas. Besides, the end-to-end latency is the most important issue in an MEC network; however, solutions considering the cloud environment or offline cases fail to apply in real-time MEC applications. The VNRs in most of the previous researches are arbitrary network requests -  , which consist of nodes and links in a random method. Here, we consider Activity on Vertex (AOV) networks as our VNRs ,  , which the nodes are represented by network functions, and they will process one after another. We believe that AOV network VNRs are more realistic to most of the MEC applications, such as video stream and facial detection. Take facial detection application for example, the photo goes through decompressing before the face recognition and then proceeds to the next step if the identity is verified. Additionally, the substrate nodes in most VNE problems are resources, such as CPU or storage, with fixed capacities and symmetric processing speed -  . VNRs with different service priorities or different latency requirements will be ignored and treated equally in terms of processing time. The fixed capacities and symmetric processing speed substrate nodes are not suitable for serving heterogeneous virtual requests simultaneously in that they are comprised of latency-sensitive MEC applications and non-latency-sensitive applications. With the emerge of new processor technologies, such as Intel Speed Select Technology (SST) ,  , high priority or latencysensitive workloads are able to be powered up, while lower the processing speed of the other workloads, leading to higher software performance. This kind of concept perfectly meets the characteristic of MEC applications in a virtualized 5G environment or even 6G networks in the future. If InPs can dynamically adjust the processing speed of the substrate nodes for each network request, the overall processing delay will decrease, and the revenue in a long run will increase at the same time.