Abstract

  In recent years, there are many work efficiency efforts in particular automation of business processes in various industry. The same trend applies to the network operations of telecommunication carriers; however, although automation of service fulfillment, such as developing network services, is progressing, automation of service assurance such as responding to network failure has not progressed. Currently, network maintenance personnel have independently developed tools for automation of network maintenance (such as troubleshooting) process partially and they are manually performing maintenance tasks by coordinating multiple groups of tools. However, in view of the future decrease in the working population, work efficiency must be further improved. Following this situation, this article focuses on further automation of network maintenance process, by developing technologies to orchestrate existing automation tools.

Aiko Oi, Ryosuke Sato, Yuichi Suto, Kosuke Sakata, Motomu Nakajima, Tsuyoshi Furukawa (NTT Corporation, Japan)

Abstract

  Software-Defined Networking (SDN) and Network Function Virtualization (NFV) help reduce OPEX and CAPEX as well as increase network flexibility and agility. But at the same time, operators have to cope with the increased complexity of managing virtual networks and machines, which are more dynamic and heterogeneous than before. Since this complexity is paired with strict time requirements for making management decisions, traditional mechanisms that rely on, e.g., Integer Linear Programming (ILP) models are no longer feasible. Machine learning has emerged as a possible solution to address network management problems to get near-optimal solutions in a short time. In this paper, we propose a Graph Neural Network (GNN) based algorithm to manage VNFs. The proposed model solves the complex VNF management problem in a short time and gets near-optimal solutions.

Heegon Kim*, Suhyun Park*, Stanislav Lange**, Doyoung Lee*, Dongnyeong Heo***, Heeyoul Choi***, Jae-Hyung Yoo*, James W. Hong*

(*POSTECH, South Korea; **NTNU, Norway; ***Handong Global Univ., South Korea)

Abstract

  Flying networks are resource constraints while the nature of nodes’ mobility is very dynamic and unpredicted. Therefore, these networks are very prone to link failure and performance degradation. By considering the existing limitations, this work proposes a new approach consists of proactive and reactive network failure mitigation techniques that have been named as a hybrid approach. In the proposed architecture, the SDN controllers are distributed where each one controls its local domain nodes. The controller node continuously monitors the network state information and proactively adjusts the near-future changes to the topology. Each local domain also contains a sink node that directly connects to the controller. The sink node is used to forward the network state information to the controller and keep the controller defined flow rules for local domain nodes. The sink node can also request a new path in case of any link failure or any topology updates cause by nodes’ movement. Besides, a distributed routing protocol also runs on domain nodes to establish connectivity toward the sink node.

Muhammad Saqib, Asif Mehmood, Adeel Rafiq, Muhammad Afaq, Wang-Cheol Song

(Jeju National Univ., South Korea)

Abstract

  In recent years, Internet Data Center (IDC) network operation with IPv6 single-stack gathers attention because of IPv4 address pool exhaustion and the high operational load associated with IPv4/IPv6 dual-stack. Since IPv4 traffic still accounts for a large percentage of the total Internet traffic, it is necessary to provide IPv4 services even in IPv6 single-stack networks. A method of providing IPv4 accessibility in an IPv6 single-stack network is known as IPv4 as a Service (IPv4aaS); there are three possible approaches. This research proves that IPv4/IPv6 translation is the most compatible and effective for IDC network operations. The SIIT-DC network design has been standardized as a way to provide IPv4 services with IPv4 / IPv6 Translation. In SIIT-DC, a translation node referred to as Border Relay (BR) is connected between the IPv4 Internet and an IPv6 IDC network in order to translate IPv4 and IPv6 mutually (in order to enable IPv4 service provision in IPv6 server). However, to operate an SIIT-DC network without the lack of redundancy and operational load spikes in multi-homing networks, the Explicit Address Mapping Table (EAMT) must be shared dynamically and consistently with BRs. This paper compares approaches to control EAMT dynamically in SIIT-DC environment and clarifies the benefits and drawbacks of each. The results are expected to contribute to the design of IPv6 single-stack networks in IDC.

Yasunobu Toyota, Osamu Nakamura (Keio Univ., Japan)

Abstract

  5G claims the service provisioning for a variety of new applications increasing orchestration complexity. Additionally, many different orchestrators and platforms have been developed over recent years to handle and control the network infrastructures. However, each of the orchestration platforms requires different expertise for handling them. The diversity and dynamicity in terms of requirements made it very complex for the operators to handle runtime operations. As with the increasing number of devices, the amount of traffic streamed over network deviates drastically making it impossible for network operators to handle the network operations manually. Hence automated platforms are a dire need, this work proposed a generic intent-based system that can automatically orchestrate and manage network lifecycle over multiple domains, sites and orchestrators. This system will include the orchestration over multiple domains access, transport and core and it also includes the use of Machine Learning for prediction of resource status for proactive decision making to manage the slice lifecycle automatically.

Talha Ahmed Khan, Khizar Abbas, Adeel Rafiq, Muhammad Afaq, Wang-Cheol Song (Jeju National Univ., South Korea )

Abstract

  We introduce a structural deterioration determination system that automatically measures the deflection of concrete utility poles from a 3D model created by using 3D point cloud data. As pole deflection is an accurate precursor of cracking, our automated deflection assessment system improves the efficiency of utility pole inspection, by replacing the conventional visual inspections needed to identify cracking. We show that the time and effort of inspection can be reduced by about 52% compared with conventional visual inspection work.

Yukihiro Goto, Shigehiro Matsuda, Masaki Waki, Hiroyuki Oshida (NTT Corporation, Japan )

Abstract

  The Internet of Things is now part of everyday life and there has been a wide range of novel IoT applications collecting cyber-physical data and providing information on the environment. As it is expected that the IoT traffic will count for a major part of the Internet traffic, it is essential to characterize the IoT traffic and to identify each device, and especially in the case of cyberattacks. In this paper, we present a new method to identify IoT devices based on traffic entropy. We compute the entropy values of traffic features and we rely on Machine Learning algorithms to classify the traffic. Our method succeeds in identifying devices under various network conditions with performances up to 94% in all cases. Our method is also robust to unpredictable network behavior with anomalies spreading into the network.

Hung Nguyen-An*/**, Thomas Silverston*, Taku Yamazaki*, Takumi Miyoshi*

(*Shibaura Institute of Technology, Japan; **Hanoi Univ. of Science and Technology, Vietnam )

Abstract

  The importance of real-time notification has been growing for social services and Intelligent Transporting System (ITS). As an advanced version of Pub/Sub systems, publishprocess-subscribe systems, where published messages are spooled and processed on edge servers, have been proposed to achieve data-driven intelligent notifications. In this paper, we present a system that allows a topic to be managed on multiple edge servers so that messages are processed near publishers and subscribers, even when publishers are spread over a wide area. However, permitting a topic to use multiple edge servers can cause exhaustion of resources as the number of topics and consumed resources by each topic increase. We formulate the delay caused by the depletion of edge server resources to find the optimal allocation of topics on limited edge server resources. As the optimization problem is NP-hard, we propose heuristics leveraging the given locality and the pub/sub relationships observed between clients to use the edge server resources efficiently while delivering realtime notifications. The experimental results demonstrate that the proposed method reduces the delay to deliver notifications and the effectiveness of the strategy exploiting the relationships between clients.

Tomoya Tanaka, Tomio Kamada, Chikara Ohta (Kobe Univ., Japan )

Abstract

  In recent years, LTE system with unlicensed bands is proposed. It can achieve not only to boost the capacity of cellular networks but also to use unlicensed bands effectively since LTE can achieve better spectrum efficiency than Wi-Fi. However, when LTE is introduced into an environment where WiFi access points are already deployed, significant interference may be caused to Wi-Fi. Therefore, the coexistence of LTE and Wi-Fi with unlicensed bands is the problem. In this paper, we assume an environment where LTE base stations and Wi-Fi access points are deployed high-densely with unlicensed bands and co-channel interference frequently is caused. Then, we aim to enhance the average user throughput and its fairness through a centralized control of LTE and Wi-Fi. Specifically, we propose a channel assignment method and a users’ access system selection method using GA. The proposed method enhances both the average user throughput and the fairness of user throughput compared with conventional static channel assignment methods.

Kazuhiko Kinoshita*, Kazuki Ginnan**, Keita Kawano***, Hiroki Nakayama****, Tsunemasa Hayash****, Takashi Watanabe**

(*Tokushima Univ., Japan; **Osaka Univ., Japan; ***Okayama Univ., Japan; ****Bosco Technologies Inc., Japan )

Abstract

  Task offloading scheme provides opportunistic energy saving for computation-intensive on-vehicle applications. The evolution of the Vehicular Edge Computing (VEC) paradigm has contributed a vast potential that can enhance the performance of such vehicles with energy-hungry and delay-sensitive services. However, determining how much workload to compute locally or offload to the VEC server is still quite challenging. Moreover, when all the vehicles try to offload their computation tasks to the same VEC server, it leads to deterioration in the performance gain due to overburden. Recently, unmanned aerial vehicle (UAV) as the edge server has gained huge attraction due to its well maneuverability and cost efficiency. In this paper, we study the energy-efficient offloading as well as association of the vehicles between the road side unit (RSU) and UAV. First, we formulate the joint offloading and association problem. Next, we decompose the formulated mixed integer linear (MIL) problem into two subproblems and then solve them by using standard convex optimization. Finally, we compare our proposed algorithm with benchmark schemes and the numerical results demonstrate that our algorithm outperforms the benchmark solutions.

Pyae Sone Aung, Yan Kyaw Tun, Nway Nway Ei, Choong Seon Hong (Kyung Hee Univ., South Korea )

Abstract

  As the number of network devices increases rapidly, it becomes more and more difficult to defend network attacks. Large-scaled attacks, such as SYN flooding, may lead to heavy burden to the switches as well as the controller in a software defined network (SDN). In this paper, we investigate the SYN flooding and Address Resolution Protocol (ARP) spoofing attacks in SDN, and then propose mechanisms to address these two attacks. We also present a new scheme to detect SYN flooding by using only a few forwarding rules. Moreover, we utilize the Programming Protocol-independent Packet Processors (P4) technique to mitigate the burden of the controller

Ting-Yu Lin, Jhen-Ping Wu, Pei-Hsuan Hung, Ching-Hsuan Shao, Yu-Ting Wang, Yun-Zhan Cai, Meng-Hsun Tsai

(National Cheng Kung Univ., Taiwan)

Abstract

  Port scanning is a well-known behavior when a botnet searches target devices. To detect port scanning accurately, data with high discriminatory power are indispensable. Most related works, however, focus on data analysis methods but neglect storage limitations of switches, which makes their methods impractical. Therefore, we propose a new data collection method for collecting network information of port scanning in P4-based SDN named 0-replacement. Through simulations, we compare the 0-replacement method with two classic data collection methods. Results show that the 0-replacement method improves the true positive ratio by at least 25 percentage points but only consumes 0.36% memory space.

Yun-Zhan Cai, Chih-Hao Lai, Yu-Ting Wang, Meng-Hsun Tsai (National Cheng Kung Univ., Taiwan)

Abstract

  By enabling the deployment of softwarelized network functions on commodity servers, Network Function Virtualization (NFV) brings many benefits such as rapid development and deployment, simplicity and flexibility in network operations and management. Monitoring the performance characteristics of Virtual Network Functions (VNFs), such as packet processing time, is crucial to achieving maximum benefit from NFV. In this paper, we present Packer Processing Time Monitoring (PPTMon) - a solution for real-time and lightweight VNF packet processing time monitoring. PPTMon embeds timestamp information directly into the packets. PPTMon is implemented using extended Berkeley Packet Filter (eBPF) - a new Linux framework that allows high-speed packet processing. Our experiments showed that PPTMon can monitor VNFs with high accuracy and low performance overhead.

Tu Van Nguyen, Jae-Hyoung Yoo and James W. Hong (POSTECH, South Korea )

Abstract

  Federated learning is a distributed learning method that trains a deep network on user devices without collecting data from central server. It is useful when the central server can’t collect data. However, the absence of data on central server means that deep network compression using data is not possible. Deep network compression is very important because it enables inference even on device with low capacity. In this paper, we proposed a new quantization method that significantly reduces FPROPS(floatingpoint operations per second) in deep networks without leaking user data in federated learning. Quantization parameters are trained by general learning loss, and updated simultaneously with weight. We call this method as OQFL(Optimized Quantization in Federated Learning). OQFL is a method of learning deep networks and quantization while maintaining security in a distributed network environment including edge computing. We introduce the OQFL method and simulate it in various Convolutional deep neural networks. We shows that OQFL is possible in most representative convolutional deep neural network. Surprisingly, OQFL(4bits) can preserve the accuracy of conventional federated learning(32bits) in test dataset.

YouJun Kim, Choong Seon Hong (Kyung Hee Univ., South Korea )

Abstract

  Large-scale web services are increasingly adopting the microservice architecture that mainly utilizes container technologies. Microservices are operated on complex configured infrastructures, such as containers, virtual machines, and physical machines. To ensure service quality of microservices, it is important to monitor not only the quality of services but also the quality of the infrastructures utilized by the services. Therefore, the metrics of the infrastructure related with the services should be traced. An extended Berkeley Packet Filter (eBPF) is a relatively new Linux’s function, which is effectively used as a sensor of container-network metrics. There are two key challenges in realizing the service-linked monitoring system. One challenge is making the full-stack topology between microservices, containers, and machines visible to set the sensor related with the services. Another challenge is dynamic sensor management that can relocate the sensor quickly after the topology’s change. In this paper, we propose a real-time monitoring system that creates a full-stack topology and relocates the sensor in conjunction with events from a container orchestrator. The system enables a dynamic deployment of the sensors related with the monitored services.

Takashi Shiraishi, Masaaki Noro, Reiko Kondo, Yousuke Takano, Naoki Oguchi (Fujitsu Laboratories Ltd, Japan )

Abstract

  The advent of GPGPU (General-Purpose Graphic Processing Unit) containers enlarges opportunities of acceleration and easy-to-use in clouds. However, there is still lack of research on utilizing efficiently GPU resource and managing multiple applications at the same time. Co-execution of applications without understanding applications’ execution characteristics may result in low performance caused by their interference problems. To solve the problem, this paper defines resource metrics that causes performance degradation when sharing resource. We calculate the degree of interference during concurrent execution of multi applications using a ML (Machine Learning) method with the metrics. The experiments show that the execution of interference aware groups improves 7% in execution time compared to non-interference aware group in overall. For a workload consisting of several applications, the overall performance was improved by 18% and 25%, respectively, when compared to SJF and random

Sejin Kim, Yoonhee Kim (Sookmyung Women's Univ., South Korea )

Abstract

  Fast and reliable communication plays a major role in the success of smart shopping applications. In a ”Just Walk Out” shopping scenario, a video camera is installed on the cart to monitor shopping activities and transmit images to the cloud for processing so that items in the cart can be tracked and checked out. This paper proposes a prototype of a smart shopping cart based on image-based action recognition. Firstly, deep learning networks such as Faster R-CNN, YOLOv2, and YOLOv2-Tiny are utilized to analyze the content of each video frame. Frames are classified into three classes: No Hand, Empty Hand, and Holding Items. The classification accuracy based on Faster RCNN, YOLOv2, or YOLOv2-Tiny is between 93.0% and 90.3%, and the processing speed of the three networks can be up to 5 fps, 39 fps, and 50 fps, respectively. Secondly, based on the sequence of frame classes, the timeline is divided into No Hand intervals, Empty Hand intervals, and Holding Items intervals. The accuracy of action recognition is 96%, and the time error is 0.119s on average. Finally, we categorize the events into four cases: No Change, placing, Removing, and Swapping. Even including the correctness of the item recognition, the accuracy of shopping event detection is 97.9%, which is higher than the minimal requirement to deploy such a system in a smart shopping environment. A demo of the system and a link to download the data set used in the paper are in Smart Shopping Cart Prototype or found at this URL: https://hackmd.io/abEiC83rQoqxz7zpL4Kh2w.

Hong-Chuan Chi*, Muhmmad Atif Sarwar*, Yousef-Awwad Daraghmi**, Kuan-Wen Liu*, Tsì-Uí İk*, Yih-Lang Li*

(*National Chiao Tung Univ., Taiwan; **Palestine Technical Univ. Kadoorie, Palestine )

Abstract

  Online social networking platforms are used for the diffusion of information, ideas or news for various applications in our day to day life. Given a budget, the information diffusion to different entities of a network mainly depends on key entities that are initially selected as seed nodes for maximizing the influence within a network. Many studies have been done on Influence Maximization but they ignore the effect of community structure in finding influential seed nodes. This paper finds most influential seed nodes by finding smaller non-overlapping communities in a network and picking the best seed nodes in each community based on Degree centrality and Closeness centrality. This paper presents a comparative study of information diffusion through seed nodes selections by using different centrality measures and k-core decomposition method. We have used Linear Threshold (LT) diffusion Model for finding the number of activated nodes from a given set of seed nodes in one step. Our result shows that the proposed method is able to achieve maximum level of activation within a less budget as compared to five other measures. We have performed our analysis on four different datasets.

Kalyanee Devi, Rohit Tripathi (Indian Institute of Information Technology Guwahati, India )

Abstract

  Recent advance of Network Function Virtualization (NFV) enables deployment of NFV-based network services on a cloud-base infrastructure. Meanwhile, edge computing provides cloud infrastructure on the the edge of access network to serve end user. This paper proposes a decentralized market-based framework that matches independent network service providers (NSPs) with independent edge service providers (ESPs) for possible deployments of network services with associated payments. The framework benefits all participants by guaranteeing that the results are profitable for all participants: no NSP or ESP can be better off by dropping out of the match result. Simulation results indicate that the proposed approach had the highest average number of deployed network services compared with counterparts.

Yi-Chia Chen, Li-Hsing Yen (National Chiao Tung Univ., Taiwan )

Abstract

  Reinforcement learning (RL) has been used in various path finding applications including games, robotics and autonomous systems. Deploying Service Function Chain (SFC) with optimal path and resource utilization in edge computing environment is an important and challenging problem to solve in Software Defined Network (SDN) paradigm. In this paper we used RL based Q-Learning algorithm to find an optimal SFC deployment path in edge computing environment with limited computing and storage resources. To achieve this, our deployment scenario uses a hierarchical network structure with local, neighbor and datacenter servers. Our Q-Learning algorithm uses an intuitive reward function which does not only depend on the optimal path but also considers edge computing resource utilization and SFC length. We defined regret and empirical standard deviation as evaluation parameters. We evaluated our results by making 1200 test cases with varying SFC-length, edge resources and Virtual Network Function's (VNF) resource demand. The computation time of our algorithm varies between 0.03~0.6 seconds depending on the SFC length and resource requirement.

Suman Pandey , Jae-Hyoung Yoo, James W. Hong (POSTECH, South Korea )