Constant monitoring and agile management of cloud-native network functions (CNFs) are the essential to prevent serious degradation of quality of services (QoS) in network service infrastructure. To monitor and agilely manage services against dynamically changing situations in complicated cloud-native (CN) network infrastructure, automation of closed-loop service management should be discussed. In this paper, we discuss a unified interface of interworking multiple functions including different AI models to establish autonomic closed-loop service management that adapts not only to the past failure events, but also to the future ones according to the failure prediction. In addition, we developed the prototype of automatic closed-loop service management for CN infrastructure that automatically constructs network services and maintains QoS of the existing services based on the interworking three AI models dedicated for self-monitoring, self-configuration, and self-optimization.

Network function virtualization (NFV) technology enables the deployment of a distinct network slice composed of various virtualized network functions (VNFs) required for offering a communication service. As the user population or network traffic volume fluctuates, the capacity of the network slices must be dynamically adjusted. It is necessary to decide whether to place VNFs in appropriate locations such that the quality of services (QoS) is guaranteed while optimally utilizing the allocated resources. The above challenge can be addressed by solving an optimization problem, such as integer linear programming. However, as solving the optimization problem requires a considerable amount of time, it is not applicable in agile network control mechanisms. In this paper, we propose a network function placement algorithm based on salience, which is one of the characteristic metrics of links in a graph. Our proposal is aimed to provide the adequate placement in short time to achieve the following objectives of keeping the slice reconstruction cost low and avoiding the cases of QoS degradation. The simulation results show that our scheme performs equally well for different sized slices (i.e., the user population). We found that the network reconfiguration costs are low and almost the same for various sized slices when they are reconstructed by the proposed method in a substrate network of 200 nodes. Furthermore, we confirmed that the proposed method keeps the slice reconstruction cost low and avoids blocking of slice reconstruction even in the situations when a resource competition occurs among multiple slices.

The 5th generation (5G) or beyond 5G (B5G) mobile networks offer a range of services over the existing infrastructure. Service function chaining (SFC) provides a platform for flexible resource management by dynamically allocating and adjusting resources to virtual network functions (VNFs). To satisfy the quality of service (QoS) requirements, the system requires the migration of VNFs from the current server to other servers that have sufficient resources. Previously, we discussed an artificial intelligence (AI)-based VNF migration scheduling mechanism for multiple service function (SF) chains. However, our previous approach focused only on SFC management in small-scale networks. To manage largescale networks, we must consider SFC path selection against interference among multiple SFC management Als. To avoid interference, in this study, we propose a supervisory mechanism to establish an SFC management system consisting of multiple Als. The supervisor determines the path recommendation ratio for each chain group governed by the management Als. The recommendation ratio is determined by the topological features and predicted resource demands of the chains. By using the recommendation ratio, SFC management may reduce the load of VNF-providing servers by approximately 20% compared with the case where no supervisor is used.

Service function chaining (SFC) provides the plat-form for flexible resource management by dynamically allocating resources to virtual and/or container network functions (VNFs/CNFs). To meet the quality of service (QoS) requirements while facing increasing resource demands, the sys-tem will require the migration of the VNFs/CNFs from the current server to the others that offer sufficient resources. In this study, we formulate an integer linear programming (ILP) based optimization model to solve the function migration scheduling problem so that it meets QoS requirements of each service function (SF) chain. The remarkable points of this work are the following two points. The one is that we consider latency between VNFs/CNFs belonging to an SF chain, avoiding overhead due to their unnecessary migration and resource shortage. And the other is that we consider the case in which each VNF/CNF must be to be deployed strictly to a designated virtual machine (or container). To reduce complexity, we apply an encoder-decoder recurrent neural network (ED-RNN) as a machine learning model to the function migration scheduling problem. Performance evaluations show that the ED-RNN based approach achieves a similar performance as the ILP, while adding the benefits of very low complexity.

Network function virtualization (NFV) technology enables us to deploy a distinct network slice composed of various virtualized network functions (VNFs) required for a given service, such as video on demand (VoD) and Internet-of-things (IoT). As a result, application service providers (ASPs) can easily deploy and provide their services to end users (EUs) by leasing resources formed as network slices. However, as the user population or the network traffic volume fluctuates, the capacity of network slices has to be adjusted dynamically. It requires to take decision about placing NFs in appropriate locations so that the quality of services (QoS), such as transmission or processing latency is guaranteed while optimally utilizing the reserved resources. One can address the above challenge by solving an optimization problem such as integer linear programming. However, the optimization problem takes much time to solve. In this paper, we propose a network function placement algorithm based on salience, which is one of characteristic metrics of links in graphs. Our objectives are two folds: keeping the slice reconstruction cost low, and avoiding the cases of QoS degradation. The simulation results show that our scheme performs equally well for different sizes of slices (i.e., the number of EUs). When slices are reconstructed with our scheme in a 200-node network, we found that the network reconfiguration costs are low and almost the same value for various sizes of slices.

Network function virtualization (NFV) enables network operators to flexibly provide diverse virtualized functions for services such as Internet of things (IoT) and mobile applications. NFV platforms are required to offer stable and guaranteed quality-of-service (QoS)even during dynamically changing resource demands and traffic volumes. To meet the QoS requirements against time-varying network environments, infrastructure providers must dynamically adjust the amount of computational resources, such as CPU, assigned to virtual network functions (VNFs). To provide agile resource control and adaptiveness, predicting the virtual server load via machine learning technologies is an effective approach for proactive control. In this paper, we propose a traffic prediction framework based on ensemble learning, comprising weak regressors trained by ML models, such as recurrent neural networks (RNNs), random forest, and elastic net. It was observed that the prediction error tends to worsen with time because the gap of trends between the past and future traffics becomes wider. Therefore, to reduce the prediction errors, we propose an adjustment mechanism for regressors based on forgetting and dynamic ensemble. The evaluation result with real traffic data verified that the resource adjustment scheme based on the proposed traffic prediction framework keeps the frequency of over and under provisioning low, which is lesser by 45% in comparison to RNNs and autoregressive moving average (ARMA).

Service function chaining (SFC) enables network operators to flexibly provide diverse services such as Internet-of-Things (IoT) and mobile applications. SFC technologies are required to offer stable and guaranteed quality-of-service (QoS) even in the circumstances of dynamically-changing resource demands and traffic volumes. To meet QoS requirements against time-varying network environment, infrastructure providers need to dynamically adjust the amount of computational resources such as CPU assigned to virtual network functions (VNFs) in each service function chain. If the increased resource demand of a VNF cannot be satisfied from the available resources in the current node, the VNF should be migrated to another node. Related work has been tackling SFC embedding and scheduling issues by applying various techniques such as game theory, integer linear programming and heuristic methods. However, they have limitations that they cannot provide an agile operation of VNF migration as they require a large number of iterations. In order to overcome the limitations, we propose to utilize an encoder-decoder recurrent neural network and train it to solve the VNF migration scheduling problem. Through the simulation, we verify that the proposed method can agilely determine VNF redeployment locations by keeping the frequency of server overload and VNF migration minimum.

Service function (SF) chaining enables network operators and infrastructure providers to flexibly orchestrate virtualized network functions (NFs) at various places on software defined virtual networks. Meanwhile, with the rapid prevalence of Internet-of-Things (IoT) applications and mobile network services, quality-of-service (QoS) requirements are getting increasingly diverse and network traffic volume and pattern are varying rapidly. The time-varying network traffic requires the infrastructure provider to dynamically adjust the appropriate amount of computational resources (e.g. CPU) assigned to an NF to efficiently process the traffic in the SF chaining infrastructure. In this paper, to adequately allocate CPU resources (available in limited amount) to virtual networks (VNs) requiring to satisfy diverse QoS levels in response to traffic variation without manual operations, we propose autonomic arbitration of CPU resources along SF chains which combines two CPU resource adjustment methods: The first method performs resource arbitration among VNs within a node, and the second method migrates NFs among nodes within a VN. These methods autonomically adjust the amount of CPU resource allocated to each NF so that over utilization of CPU can be avoided as much as possible. Moreover, during the autonomic resource adjustments in SF chains, the proposed methods are able to maintain communication paths on the substrate network so that the service is not interrupted at all. We also propose differentiated resource allocation for QoS differentiation and transferable resource searching. Through computer simulation, we verify that our autonomic resource adjustment methods can reduce the occurrence of CPU-saturation by more than 90%.

The rising popularity of big data processing for semantically rich applications such as social networks and IoT (Internet of Things) has made the range queriable cloud storage increasingly important. To support range queries, the data locality is preserved strictly, which makes the load balancing among nodes a challenging task. Currently, most of the range queriable cloud storage systems adopt the combination of neighbor item exchange and neighbor migration methods, which incurs large overhead, and suffers from slow convergence. In this work, we present a novel virtual node based distributed load balancing method for range queriable cloud systems. In our method, each physical node is partitioned into multiple virtual nodes, and all the virtual nodes are organized with range queriable P2P network. Load balancing is conducted in both overlay level (betweenneighboring virtual nodes) without global knowledge and physical level (among physical nodes) with limited global knowledge. Both theoretical analysis and simulations show that our method can significantly reduce the overhead and shorten the convergence time.

The primary challenges with wireless sensor networks are how to construct the shortest spanning tree and how to determine the optimal sink node position in terms of minimizing the data transmission times and their variances for gathering data from all sensor nodes to a sink node. To solve these two problems, we previously proposed a basic algorithm that uses the polygonal affine shortening algorithm with flow aggregation in a fully distributed manner. In this paper, we proposed an algorithm that is more than 25% faster than the previously proposed algorithm and also proves that the proposed algorithm works in wireless sensor networks. The proposed algorithm was implemented on a well-known multi-agent simulator called Net logo.

The rapid development of sensor networks has made it possible to build large-scale sensor overlay networks by integrating separated sensing resources. As the overlay substrate of sensor overlay networks, range queriable P2Ps such as Skip Graph are often applied since they enable retrieving sensing resources whose properties are within the specified range in an effective and scalable way. The range queriable P2P-based sensor overlay networks have proven to be effective in many scenarios, however, the sensor overlay nodes will suffer from flash crowds of queries when disasters, accidents or some events happen. In this paper, we present a virtual node-based approach to alleviate flash crowds for sensor overlay networks. In our approach, a hot-spot node can ask physically spare nodes located in any where in the overlay to generate virtual nodes with the same key as itself, and make the virtual nodes join the overlay network as if they were normal nodes around the hot-spot. With this method, both the query service load of the hot-spot node and the query routing load of the nearby nodes of the hot-spot node can be well distributed with low cost. With theoretical analysis and simulations, we show that our method is efficient, feasible and scalable.

The rapid development of crowd sensing and cloud computing makes it possible to build world wide Cyber Physical Systems (CPS). Range queriable P2P overlay technologies are considered as promising candidate for realizing some important functionalities of CPS, for example, retrieval based on the properties of sensors. However, it is difficult to build world wide systems with existing range queriable P2Ps efficiently as the lack of proximity awareness. In this paper, we propose the hierarchical neighbor selection (HNS) mechanism for range queriable P2P overlays, which can integrate the hierarchy of physical Internet into overlay construction and routing to improve the performance. We also implement HNS based on Skip Graph (SG), an efficient range queriable P2P technology. With extensive simulations, we show that HNS based Skip Graph (HSG) can improve the routing latency and locality significantly with little overhead.

In-network caching is a core technology for saving network resources and improving the response time in Information-/Content-Centric Networking (ICN/CCN). In an ordinary ICN/CCN, content name based routing is used to find cached content in the network. Although there have been several studies on effective and efficient cache retrieval from in-network caches, we propose a novel content hunting scheme, named Local Tree Hunting (LTH) from the point of view to approach the ideal performance. LTH hunts for the closest node to the request node among the nodes with disseminated cache content. It autonomously adjusts the hunting area by changing the size of the hunting tree depending on the degree to which the content has been disseminated through the network. The proposed scheme was verified in a performance analysis; it was shown to have almost the same level of performance as the case of finding the true closest node. System architecture and operations based on the proposed scheme are also described.

Explicit-state model checking which is conducted by state space search has difficulty in exploring satisfactory state space because of its memory requirements. Though bitstate hashing achieves memory efficiency, it cannot guarantee complete verification. Thus, it is desirable to provide a reliability indicator such as a coverage estimate. However, the existing approaches for coverage estimation are not very accurate when a verification run covers a small portion of state space. This mainly stems from the lack of information that reflects characteristics of models. Therefore, we propose coverage estimation methods using a growth curve that approximates an increase in reached states by enlarging a bloom filter. Our approaches improve estimation accuracy by leveraging the statistics from multiple verification runs. Coverage is estimated by fitting the growth curve to these statistics. Experimental results confirm the validity of the proposed growth curve and the applicability of our approaches to practical models. In fact, for practical models, our approaches outperformed the conventional ones when the actual coverage is relatively low.

Several approaches for IP fast rerouting have been proposed to augment resiliency of IP networks. For example, NotVia uses IP tunnels to avoid failure component, and FIFR prepares two routing tables to switch packets according to incoming interfaces. In this paper, for two-table approach such as FIFR, we propose to use 2-bit marking on packet header to overcome one of the problems called route bouncing. By preventing route bouncing, more efficient use of network resources are possible. We further show that our method not only prevents route bouncing but also prevents packet loops in case of multiple failure. Note that, although multiple failure occurs with low probability, once it occurs network performance significantly degrades since size of loops is far larger than microloops.

Most wireless networks consist of heterogeneous nodes with diverse characteristics. These heterogeneous nodes have various moving characteristics such as speed and direction. Since conventional wireless routing schemes are designed for networks with homogeneous mobility, they have to cope with the routing difficulties between heterogeneous nodes by careful tuning of routing parameters. Because optimally tuning to such parameters is difficult, this tuning tends to result in higher control overhead and to degrade communication quality, e.g., lower packet reachability and higher packet delay. This paper proposes efficient extensions of a proactive routing protocol to achieve sufficient communication quality in heterogeneous mobile networks. Simulation results reveal that the proposed scheme can achieve higher packet reachability and lower delay with low control overhead compared with existing routing schemes.