In management of computer network systems, a service function chaining (SFC) module plays a vital role in generating an efficient network traffic path that connects virtualized network functions (VNF) on network topology to serve a user request. The SFC module needs to generate a complete path quickly even in various network situations, including dynamic VNF resources, various types of requests, and various network topologies to provide the best quality of service. The previous supervised learning method demonstrated that graph neural networks (GNN) could represent network features for the SFC task. However, the supervised learning method works properly only in the network situation in which the model was trained with labels. Due to the limitation, it showed poor performance on new unseen network situations. In this paper, we apply a reinforcement learning algorithm to train GNN based models in various network situations even without label information. In the experiments, compared to the previous supervised learning method, the proposed methods demonstrate remarkable generalization effects, showing that the proposed methods could work successfully on unseen network situations without re-designing and re-training.

Typing on wearables while situationally impaired, such as while walking, is challenging. However, while HCI research on wearable typing is diverse, existing work focuses on stationary scenarios and fine-grained input that will likely perform poorly when users are on-the-go. To address this issue we explore single-handed wearable typing using intra-hand touches between the thumb and fingers, a modality we argue will be robust to the physical disturbances inherent to input while mobile. We first examine the impact of walking on performance of these touches, noting no significant differences in accuracy or speed, then feed our study data into a multi-objective optimization process in order to design keyboard layouts (for both five and ten keys) capable of supporting rapid, accurate, comfortable, and unambiguous typing. A final study tests these layouts against QWERTY baselines and reports performance improvements of up to 10.45% WPM and 39.44% WER when users type while walking.

Video conferencing systems have been developed for a long time. However, due to COVID-19, people came to realize the importance of such systems that their widespread usage increased drastically. With the growing number of users of video conferencing systems, it is essential to be able to manage the system better. Even though Network Function Virtualization (NFV) has helped in handling the dynamic nature of video conferencing systems by providing the ability to deploy and/or remove virtual instances, being able to manage the virtual instances efficiently and optimally is a necessity. One of the main problems to be solved to achieve this goal is the auto-scaling problem. In this problem, an auto-scaling agent has to decide whether the number of a virtual instance type has to be increased, decreased, or maintained. In this paper, we designed and experimented a new auto-scaling algorithm by using Reinforcement Learning (RL), specifically the Deep Q-Network (DQN) algorithm. The experiment was done in both emulated and simulated environments. We then evaluated our proposed approach by comparing it with a threshold-based auto-scaler, the baseline algorithm. The results showed that the DQN-based auto-scaler performed better than the baseline algorithm, using fewer virtual instances with a better quality of service overall, as well as being more robust to shifting data distribution. Overall, the applications of DQN on auto-scaling algorithms for cost-effective and high-performance video conferencing systems are promising.

Network softwarization paradigms such as SDN and NFV provide network operators with advantages in terms of scalability, cost and resource efficiency, as well as flexibility. However, in order to fully reap these benefits and cope with new challenges regarding the heterogeneity of user demands and an ever-growing service landscape, management and operation of such networks requires a high degree of automation that ensures fast and proactive decision making. With the recent success of machine learning (ML) across numerous domains, a shift from traditional rule-based policies towards ML-based approaches in the context of network management is taking place. Although many individual contributions cover use cases such as predicting various network characteristics or optimizing the configuration of components, a fully integrated architecture for achievingNetwork Intelligenceis still missing. Hence, in this work, we propose such an architecture that combines the capabilities of softwarized networks with ML-based management. The contribution of this article is threefold: first, we present the proposed architecture alongside its components. Second, we implement a proof-of-concept version of all components in our OpenStack-based testbed. Finally, we demonstrate in a case study regarding VNF resource prediction how the proposed architecture can be used to generate realistic data sets to train and evaluate ML-based models for this task.

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.

With the advent of the 5G network era, it is required to flexibly build and manage networks to meet rapidly changing service requirements. Software-defined networking (SDN) and network function virtualization (NFV) are key technologies that enable flexible network management by transforming networks into software-based networks. Besides, NFV has the advantage of virtualizing network functions, operating those on commercial (COTS) servers, and managing the network functions dynamically. However, the numerous virtual networks and resources created by NFV can cause problems complicating network management. To solve the problems, research on managing complex NFV environments using artificial intelligence (AI) has recently attracted attention. In particular, service function chaining (SFC) is one of the essential NFV technologies, and it is required to create an efficient SFC path in dynamic networks. In this paper, we propose a method of finding optimal SFC path considering the resource utilization of virtual network function (VNF) and VNF placement by using Q-learning, one of the reinforcement learning (RL) algorithms.

Touches between the fingers of an unencumbered hand represent a ready-to-use, eyes-free and expressive input space suitable for interacting with wearable devices such as smart glasses or watches. While prior work has focused on touches to the inner surface of the hand, touches to the nails, a practical site for mounting sensing hardware, have been comparatively overlooked. We extend prior implementations of single touch sensing nails to a full set of five and explore their potential for wearable input. We present design ideas and an input space of 144 touches (taps, flicks and swipes) derived from an ideation workshop. We complement this with data from two studies characterizing the subjective comfort and objective characteristics (task time, accuracy) of each touch. We conclude by synthesizing this material into a set of 29 viable nail touches, assessing their performance in a final study and illustrating how they could be used by presenting, and qualitatively evaluating, two example applications.

Network function virtualization (NFV) is a key technology of the 5G network era. NFV decouples a network function from proprietary hardware so that the network function can operate on commercial off-the-shelf (COTS) servers as a form of virtual network functions (VNFs). Owing to the advantage of NFV, network functions can be applied dynamically to the networks. However, NFV complicates network management because this technology creates numerous virtual resources that should be managed. To solve the problem of complicated network management, studies on applying artificial intelligence (AI) to the NFV-enabled networks, i.e., VNF life cycle management, have attracted attention. In particular, autoscaling, which is one of the essential functions of VNF life cycle management, adds or removes VNF instances to meet service requirements. It is a challenging task to determine the optimal number of VNF instances in dynamic networks, satisfying service requirements. In this paper, we propose a novel auto-scaling method using reinforcement learning (RL) for scale-in/out of multi-tier VNF instances, i.e., service function chaining (SFC) in NFV environments. The proposed approach defines RL's states using a status of SFC composed of multi-tier VNF instances and uses service level objectives (SLO) to make a reward model. We validate the proposed approach in an OpenStack environment, and it shows that our proposed auto-scaling method provides the optimal number of VNF instances in each tier while minimizing SLO violation.

Software-Defined Networking (SDN) and Network Function Virtualization (NFV) greatly facilitate network service management. Specifically, these new network paradigms help manage the network environment dynamically and cost-efficiently. Virtual Network Function (VNF) and Service Function Chaining (SFC) are important aspects of the NFV environment. In terms of NFV management, resource demand of VNFs can be predicted at a future time to handle Quality of Service (QoS) and resource allocation problems efficiently. Hence, researchers study and build a management system where machine-learning-based predictions of VNF information are used to handle auto-scaling, deployment and migration of VNFs. In addition, in recent studies, these systems have involved SFC to obtain useful information, not just a lone VNF. However, not many of studies explain clearly how chaining dependency among VNFs in a SFC can be used to predict future resource demand of a VNF. In this paper, we introduce VNF resource prediction machine learning model that maximizes the benefits of using SFC. Then, we compare several machine learning models and analyze how SFC data can help predict resource usage patterns of VNFs. We also show benefits of Attention model to improve prediction accuracy and convergence time through experiments.

Software-Defined Networking (SDN) and Network Function Virtualization (NFV) are paradigms that help administrators to manage dynamic networks. While SDN allows centralized network control, NFV provides flexible and scalable Virtual Network Functions (VNFs). These paradigms are also enablers for concepts such as Service Function Chaining (SFC) where chains are composed of several VNFs to provide a specific service. However, in order to maximize the benefits from the above-mentioned flexibility, new research questions need to be addressed, e.g., regarding effective management processes for dynamic networks. We proposed a novel learning model based on the flexibility of softwarization and abundant volume of monitoring data in NFV environments to predict VNF resource demands using SFC data. Our model is based on Context and Aspect Embedded Attentive Target Dependent Long Short Term Memory (CAT-LSTM) that consists of Target-Dependent LSTM (TD-LSTM), context embedding, aspect embedding, and attention. We developed this model to obtain high accuracy for the prediction of VNF resources such as the CPU. Our model uses two labeling systems: the qualitative resource state and the quantitative resource usage, both of which are used to evaluate its performance. This assists the administrator in understanding the network conditions, improves prediction performance, and provides practically useful information. Our learning model for predicting VNF resource demands can be utilized to solve essential SFC problems such as auto-scaling and optimal placement, which in turn prevent service interruption and provide high reliability.

Currently, server (host) virtualization technology that brings effective use of server resources to a data center is promising as cloud services are being prevalent with increasing traffic volumes and requirements for higher service quality. Proposed network TAP, named vTAP (Virtual Test Access Port), overcomes the problem that existing hardware TAP devices cannot be utilized for virtual network links to monitor traffic among virtual machines (VMs) at a packet level. vTAP can be implemented by a virtual switch that gives network connectivity to VMs by switching packets over the virtual network links. The port mirroring feature of a virtual switch can be a naive solution to provide packet level monitoring among VMs. However, using the feature in an environment that needs to treat large volume of network traffic with low delay such as NFV (Network Function Virtualization) incurs performance degradation in packet switching capability of the switch and error-prone manual configurations. This paper provides design and implementation approaches to vTAP using Open vSwitch with DPDK (Data Plane Development Kit) and an OpenFlow SDN (Software-Defined Networking) controller to overcome the problems. DPDK can accelerate overall packet processing operations needed in vTAP, and OpenFlow controller can provide a centralized and flexible way to apply and manage TAP policies in an SDN network. This paper also provides performance comparisons of the proposed vTAP and the naive method, port mirroring.

Wearable head-mounted displays combine rich graphical output with an impoverished input space. Hand-to-face gestures have been proposed as a way to add input expressivity while keeping control movements unobtrusive. To better understand how to design such techniques, we describe an elicitation study conducted in a busy public space in which pairs of users were asked to generate unobtrusive, socially acceptable hand-to-face input actions. Based on the results, we describe five design strategies: miniaturizing, obfuscating, screening, camouflaging and re-purposing. We instantiate these strategies in two hand-to-face input prototypes, one based on touches to the ear and the other based on touches of the thumbnail to the chin or cheek. Performance assessments characterize time and error rates with these devices. The paper closes with a validation study in which pairs of users experience the prototypes in a public setting and we gather data on the social acceptability of the designs and reflect on the effectiveness of the different strategies.

Software-defined Networking (SDN) is a network paradigm to resolve the challenges of traditional networks. SDN can utilize its concept of control and data plane separation to improve Quality of Service (QoS) of certain network traffic efficiently. Because traffic engineering in SDN is a promising way to satisfy the requirement well, we propose an application-aware traffic engineering system that cooperates with Deep Packet Inspection (DPI). In the system, port number and DPI-based traffic classification are used to identify application or service flows. The system improves QoS of the identified flows by distributing them to multiple queues with different priorities in each switch port. A network admin can define a mapping table between an identified flow and its QoS priority (queue) in the system. To demonstrate the feasibility of the system, we designed and implemented the system by constructing an SDN controller application and data plane entities. The results of the experiment shows increased throughput and reduced packet delay for identified application traffic.

Owing to the development of information technology, many people can access various services through the Internet. In addition, with the emergence of the Web 2.0 concept of opening, participating, and sharing, public institutions and companies have made it possible to use their data and services in a variety of ways. In this environment, service providers have developed new services by linking open data and services to meet the rapidly changing requirements of Internet users. A new service developed through the interworking of data and services is called a mashup service. A mashup service has the advantage of providing useful functions required by users by combining the existing data and services. However, to develop a mashup service, it is necessary to process the data collected from other sources and link the services, which greatly increases the developers burden. In this study, we propose an open application programming interface (openAPI)-based message router for mashup service development to overcome this problem. The message router supports the development of the mashup service by providing useful functions required by the developer in cooperation with various platforms according to the request messages transmitted through openAPI.

A common streaming service is the one typically provided by a Client-Server model, where tasks are partitioned between providers of the service, called servers, and the service requesters, called clients; thus, the quality of the content depends on the network condition existing between the server and the client. However, this streaming service model is changing as streamers, such as private Internet broadcasters, are replacing the role of content providers and allowing content to be sent to the end user in real time. Because streamers transmit the content in real time, efficient models are needed on the streamer, the server and the client side for users to receive reliable content. In this paper, we measure and analyze the quality of the content according to the streamers network condition on the streamer side. On the server side, we add flexibility to the existing content delivery network structure and verify our model by assessing metrics such as throughput, delay and jitter. On the client side, we propose a grid-based P2P model that allows users to receive content reliably.

The small screens of smartwatches provide limited space for input tasks. Finger identification is a promising technique to address this problem by associating different functions with different fingers. However, current technologies for finger identification are unavailable or unsuitable for smartwatches. To address this problem, this paper observes that normal smartwatch use takes places with a relatively static pose between the two hands. In this situation, we argue that the touch and angle profiles generated by different fingers on a standard smartwatch touch screen will differ sufficiently to support reliable identification. The viability of this idea is explored in two studies that capture touches in natural and exaggerated poses during tapping and swiping tasks. Machine learning models report accuracies of up to 93% and 98% respectively, figures that are sufficient for many common interaction tasks. Furthermore, the exaggerated poses show modest costs (in terms of time/errors) compared to the natural touches. We conclude by presenting examples and discussing how interaction designs using finger identification can be adapted to the smartwatch form factor.

Network virtualization is a technique that abstracts the underlying physical infrastructures into multiple isolated networks. Currently, network virtualization based on Software-Defined Networking (SDN) has attracted interests from industry and academia to utilize limited network resources by using benefits of SDN. SDN has useful features such as programmability, flexibility, and agility. In order to virtualize networks in SDN, a network hypervisor intercepts and modifies OpenFlow messages so that it provisions multiple virtual networks, virtual Software-Defined Networks (vSDNs). However, existing SDN-based network hypervisors do not provide an easy-to-use method to connect a created vSDN with external networks. It limits the usefulness of vSDNs. To resolve this problem, we propose a virtual gateway for external connectivity in vSDN. The proposed virtual gateway is implemented using ONOS virtualization subsystem. The virtual gateway is able to provide external connectivity and other useful network functions such as firewall, traffic shaping, and load-balancing. To demonstrate the feasibility of virtual gateway, we evaluate round trip time and deployment time to show a connectivity and overhead of the virtual gateway deployment.

Software-Defined Networking (SDN) is an emerging networking paradigm aims to improve network management flexibility and efficiency. OpenFlow is the popular SDN de-facto standard, which has been prevalently adopted by both academia and industry for research and development purpose. OpenFlow provides rich programmable interface to network administrator to ease traffic monitoring and control. Because OpenFlow supports L4 network stack, it is feasible to provide application level traffic control by specifying TCP/UDP port number in flow rules. The major deficiency of the port-based traffic control is that it only provides the ability to control traffic from applications which have well-known TCP/UDP port numbers. In the case of port number change or dynamic (ephemeral) port allocation to an application, it is difficult to accurately control the application traffic. To be a solution, we propose an application-aware traffic management method by integrating Deep Packet Inspection (DPI) function with SDN controller. To show the feasibility, we design and implement Firewall and Bandwidth Manager applications based on the proposed management method. The applications perform on top of ONOS [1] controller, and FTP rate control example is shown to prove the feasibility of the proposed flow management method.

With the advancement of the Internet technologies, cloud services and various open data, there have been many active attempts to develop mashup services. To develop a mashup service, many data sources and service platforms should be interlinked. Therefore, efficient ways of interconnecting various service platforms and managing data are essential to provide easy service development environment. However, current mashup service development environments do not provide any of them, which forces each developer to fully understand every platforms, interfaces and data format to develop a mahsup service. So, it gives a big burden to mashup service developers and also hinders growth of mashup service industry. In this paper, we propose ICBMS Smart Mediator which connects IoT, cloud, big data, mobile, and security platforms and helps developers to easily utilize them with less overhead and provides easy access to data. We have designed and implemented the proposed ICBMS Smart Mediator and created a new mahsup service with the ICBMS Smart Mediator.

Smartwatches are emerging device category that feature highly limited input and display surfaces. We explore how touch contact areas, such as lines generated by flat fingers, can be used to increase input expressivity in these diminutive systems in three ways. Firstly, we present four design themes that emerged from an ideation workshop in which five designers proposed concepts for smartwatch touch area interaction. Secondly, we describe a sensor unit and study that captured user performance with 31 area touches and contrasted this against standard targeting performance. Finally, we describe three demonstration applications that instantiate ideas from the workshop and deploy the most reliably and rapidly produced area touches. We report generally positive user reactions to these demonstrators: the area touch interactions were perceived as quick, convenient and easy to learn and remember. Together this work characterizes how designers can use area touches in watch UIs, which area touches are most appropriate and how users respond to this interaction style.

Interacting with smartwatches poses new challenges. Although capable of displaying complex content, their extremely small screens poorly match many of the touchscreen interaction techniques dominant on larger mobile devices. Addressing this problem, this paper presents beating gestures, a novel form of input based on pairs of simultaneous or rapidly sequential and overlapping screen taps made by the index and middle finger of one hand. Distinguished simply by their temporal sequence and relative left/right position these gestures are designed explicitly for the very small screens (approx. 40mm square) of smartwatches and to operate without interfering with regular single touch input. This paper presents the design of beating gestures and a rigorous empirical study that characterizes how users perform them -- in a mean of 355ms and with an error rate of 5.5%. We also derive thresholds for reliably distinguishing between simultaneous (under 30ms) and sequential (under 400ms) pairs of screen touches or releases. We then present five interface designs and evaluate them in a qualitative study in which users report valuing the speed and ready availability of beating gestures.

The touch screen interaction paradigm, currently dominant in mobile devices, begins to fail when very small systems are considered. Specifically, "fat fingers", a term referring to the fact that users' extremities physically obstruct their view of screen content and feedback, become particularly problematic. This paper presents a novel solution for this issue based on sensing touches to the perpendicular edges of a device featuring a front-mounted screen. The use of such offset contact points ensures that both a user's fingers and the device screen remain clearly in view throughout a targeting operation. The configuration also supports a range of novel interaction scenarios based on the touch, grip and grasp patterns it affords. To explore the viability of this concept, this paper describes EdgeTouch, a small (6 cm) hardware prototype instantiating this multi-touch functionality. User studies characterizing targeting performance, typical user grasps and exploring input affordances are presented. The results show that targets of 7.5-22.5 degrees in angular size are acquired in 1.25-1.75 seconds and with accuracy rates of 3%-18%, promising results considering the small form factor of the device. Furthermore, grasps made with between two and five fingers are robustly identifiable. Finally, we characterize the types of input users envisage performing with EdgeTouch, and report occurrence rates for key interactions such as taps, holds, strokes and multi-touch and compound input. The paper concludes with a discussion of the interaction scenarios enabled by offset sensing.