Recent reinforcement learning approaches for continuous control in wireless mobile networks have shown impressive results. But due to the lack of open and compatible simulators, authors typically create their own simulation environments for training and evaluation. This is cumbersome and time-consuming for authors and limits reproducibility and comparability, ultimately impeding progress in the field. To this end, we propose mobile-env, a simple and open platform for training, evaluating, and comparing reinforcement learning and conventional approaches for continuous control in mobile wireless networks. mobile-env is lightweight and implements the common OpenAI Gym interface and additional wrappers, which allows connecting virtually any single-agent or multi-agent reinforcement learning framework to the environment. While mobile-env provides sensible default values and can be used out of the box, it also has many configuration options and is easy to extend. We therefore believe mobile-env to be a valuable platform for driving meaningful progress in autonomous coordination of wireless mobile networks.

The decentralized nature of multi-agent systems requires continuous data exchange to achieve global objectives. In such scenarios, Age of Information (AoI) has become an important metric of the freshness of exchanged data due to the error-proneness and delays of communication systems. Communication systems usually possess dependencies: the process describing the success or failure of communication is highly correlated when these attempts are ``close'' in some domain (e.g. in time, frequency, space or code as in wireless communication) and is, in general, non-stationary. To study AoI in such scenarios, we consider an abstract event-based AoI process $\Delta(n)$, expressing time since the last update: If, at time $n$, a monitoring node receives a status update from a source node (event $A(n-1)$ occurs), then $\Delta(n)$ is reset to one; otherwise, $\Delta(n)$ grows linearly in time. This AoI process can thus be viewed as a special random walk with resets. The event process $A(n)$ may be nonstationary and we merely assume that its temporal dependencies decay sufficiently, described by $\alpha$-mixing. We calculate moment bounds for the resulting AoI process as a function of the mixing rate of $A(n)$. Furthermore, we prove that the AoI process $\Delta(n)$ is itself $\alpha$-mixing from which we conclude a strong law of large numbers for $\Delta(n)$. These results are new, since AoI processes have not been studied so far in this general strongly mixing setting. This opens up future work on renewal processes with non-independent interarrival times.

Modern services often comprise several components, such as chained virtual network functions, microservices, or machine learning functions. Providing such services requires to decide how often to instantiate each component, where to place these instances in the network, how to chain them and route traffic through them. To overcome limitations of conventional, hardwired heuristics, deep reinforcement learning (DRL) approaches for self-learning network and service management have emerged recently. These model-free DRL approaches are more flexible but typically learn tabula rasa, i.e., disregard existing understanding of networks, services, and their coordination. Instead, we propose FutureCoord, a novel model-based AI approach that leverages existing understanding of networks and services for more efficient and effective coordination without time-intensive training. FutureCoord combines Monte Carlo Tree Search with a stochastic traffic model. This allows FutureCoord to estimate the impact of future incoming traffic and effectively optimize long-term effects, taking fluctuating demand and Quality of Service (QoS) requirements into account. Our extensive evaluation based on real-world network topologies, services, and traffic traces indicates that FutureCoord clearly outperforms state-of-the-art model-free and model-based approaches with up to 51% higher flow success ratios.

Datacenter applications have different resource requirements from network and developing flow scheduling heuristics for every workload is practically infeasible. In this paper, we show that deep reinforcement learning (RL) can be used to efficiently learn flow scheduling policies for different workloads without manual feature engineering. Specifically, we present LFS, which learns to optimize a high-level performance objective, e.g., maximize the number of flow admissions while meeting the deadlines. The LFS scheduler is trained through deep RL to learn a scheduling policy on continuous online flow arrivals. The evaluation results show that the trained LFS scheduler admits 1.05x more flows than the greedy flow scheduling heuristics under varying network load.

Data-parallel applications are developed using different data programming models, e.g., MapReduce, partition/aggregate. These models represent diverse resource requirements of application in a datacenter network, which can be represented by the coflow abstraction. The conventional method of creating hand-crafted coflow heuristics for admission or scheduling for different workloads is practically infeasible. In this paper, we propose a deep reinforcement learning (DRL)-based coflow admission scheme -- LCS -- that can learn an admission policy for a higher-level performance objective, i.e., maximize successful coflow admissions, without manual feature engineering. LCS is trained on a production trace, which has online coflow arrivals. The evaluation results show that LCS is able to learn a reasonable admission policy that admits more coflows than state-of-the-art Varys heuristic while meeting their deadlines.

Services often consist of multiple chained components such as microservices in a service mesh, or machine learning functions in a pipeline. Providing these services requires online coordination including scaling the service, placing instance of all components in the network, scheduling traffic to these instances, and routing traffic through the network. Optimized service coordination is still a hard problem due to many influencing factors such as rapidly arriving user demands and limited node and link capacity. Existing approaches to solve the problem are often built on rigid models and assumptions, tailored to specific scenarios. If the scenario changes and the assumptions no longer hold, they easily break and require manual adjustments by experts. Novel self-learning approaches using deep reinforcement learning (DRL) are promising but still have limitations as they only address simplified versions of the problem and are typically centralized and thus do not scale to practical large-scale networks. To address these issues, we propose a distributed self-learning service coordination approach using DRL. After centralized training, we deploy a distributed DRL agent at each node in the network, making fast coordination decisions locally in parallel with the other nodes. Each agent only observes its direct neighbors and does not need global knowledge. Hence, our approach scales independently from the size of the network. In our extensive evaluation using real-world network topologies and traffic traces, we show that our proposed approach outperforms a state-of-the-art conventional heuristic as well as a centralized DRL approach (60% higher throughput on average) while requiring less time per online decision (1 ms).

In practical, large-scale networks, services are requested by users across the globe, e.g., for video streaming. Services consist of multiple interconnected components such as microservices in a service mesh. Coordinating these services requires scaling them according to continuously changing user demand, deploying instances at the edge close to their users, and routing traffic efficiently between users and connected instances. Network and service coordination is commonly addressed through centralized approaches, where a single coordinator knows everything and coordinates the entire network globally. While such centralized approaches can reach global optima, they do not scale to large, realistic networks. In contrast, distributed approaches scale well, but sacrifice solution quality due to their limited scope of knowledge and coordination decisions. To this end, we propose a hierarchical coordination approach that combines the good solution quality of centralized approaches with the scalability of distributed approaches. In doing so, we divide the network into multiple hierarchical domains and optimize coordination in a top-down manner. We compare our hierarchical with a centralized approach in an extensive evaluation on a real-world network topology. Our results indicate that hierarchical coordination can find close-to-optimal solutions in a fraction of the runtime of centralized approaches.

Modern services consist of modular, interconnected components, e.g., microservices forming a service mesh. To dynamically adjust to ever-changing service demands, service components have to be instantiated on nodes across the network. Incoming flows requesting a service then need to be routed through the deployed instances while considering node and link capacities. Ultimately, the goal is to maximize the successfully served flows and Quality of Service (QoS) through online service coordination. Current approaches for service coordination are usually centralized, assuming up-to-date global knowledge and making global decisions for all nodes in the network. Such global knowledge and centralized decisions are not realistic in practical large-scale networks. To solve this problem, we propose two algorithms for fully distributed service coordination. The proposed algorithms can be executed individually at each node in parallel and require only very limited global knowledge. We compare and evaluate both algorithms with a state-of-the-art centralized approach in extensive simulations on a large-scale, real-world network topology. Our results indicate that the two algorithms can compete with centralized approaches in terms of solution quality but require less global knowledge and are magnitudes faster (more than 100x).

Modern services comprise interconnected components, e.g., microservices in a service mesh, that can scale and run on multiple nodes across the network on demand. To process incoming traffic, service components have to be instantiated and traffic assigned to these instances, taking capacities and changing demands into account. This challenge is usually solved with custom approaches designed by experts. While this typically works well for the considered scenario, the models often rely on unrealistic assumptions or on knowledge that is not available in practice (e.g., a priori knowledge). We propose a novel deep reinforcement learning approach that learns how to best coordinate services and is geared towards realistic assumptions. It interacts with the network and relies on available, possibly delayed monitoring information. Rather than defining a complex model or an algorithm how to achieve an objective, our model-free approach adapts to various objectives and traffic patterns. An agent is trained offline without expert knowledge and then applied online with minimal overhead. Compared to a state-of-the-art heuristic, it significantly improves flow throughput and overall network utility on real-world network topologies and traffic traces. It also learns to optimize different objectives, generalizes to scenarios with unseen, stochastic traffic patterns, and scales to large real-world networks.

Data-parallel applications run on cluster of servers in a datacenter and their communication triggers correlated resource demand on multiple links that can be abstracted as coflow. They often desire predictable network performance, which can be passed to network via coflow abstraction for application-aware network scheduling. In this paper, we propose a heuristic and an optimization algorithm for predictable network performance such that they guarantee coflows completion within their deadlines. The algorithms also ensure high network utilization, i.e., it's work-conserving, and avoids starvation of coflows. We evaluate both algorithms via trace-driven simulation and show that they admit 1.1x more coflows than the Varys scheme while meeting their deadlines.

Network function virtualization (NFV) proposes to replace physical middleboxes with more flexible virtual network functions (VNFs). To dynamically adjust to everchanging traffic demands, VNFs have to be instantiated and their allocated resources have to be adjusted on demand. Deciding the amount of allocated resources is non-trivial. Existing optimization approaches often assume fixed resource requirements for each VNF instance. However, this can easily lead to either waste of resources or bad service quality if too many or too few resources are allocated. To solve this problem, we train machine learning models on real VNF data, containing measurements of performance and resource requirements. For each VNF, the trained models can then accurately predict the required resources to handle a certain traffic load. We integrate these machine learning models into an algorithm for joint VNF scaling and placement and evaluate their impact on resulting VNF placements. Our evaluation based on real-world data shows that using suitable machine learning models effectively avoids over- and underallocation of resources, leading to up to 12 times lower resource consumption and better service quality with up to 4.5 times lower total delay than using standard fixed resource allocation.

Softwarization facilitates the introduction of smart manufacturing applications in the industry. Manifold devices such as machine computers, Industrial IoT devices, tablets, smartphones and smart glasses are integrated into factory networks to enable shop floor digitalization and big data analysis. To handle the increasing number of devices and the resulting traffic, a flexible and scalable factory network is necessary which can be realized using softwarization technologies like Network Function Virtualization (NFV). However, the security risks increase with the increasing number of new devices, so that cyber security must also be considered in NFV-based networks. Therefore, extending our previous work, we showcase threat detection using a cloud-native NFV-driven intrusion detection system (IDS) that is integrated in our industrial-specific network services. As a result of the threat detection, the affected network service is put into quarantine via automatic network reconfiguration. We use the 5GTANGO service platform to deploy our developed network services on Kubernetes and to initiate the network reconfiguration.

For optimal placement and orchestration of network services, it is crucial that their structure and semantics are specified clearly and comprehensively and are available to an orchestrator. Existing specification approaches are either ambiguous or miss important aspects regarding the behavior of virtual network functions (VNFs) forming a service. We propose to formally and unambiguously specify the behavior of these functions and services using Queuing Petri Nets (QPNs). QPNs are an established method that allows to express queuing, synchronization, stochastically distributed processing delays, and changing traffic volume and characteristics at each VNF. With QPNs, multiple VNFs can be connected to complete network services in any structure, even specifying bidirectional network services containing loops. We discuss how management and orchestration systems can benefit from our clear and comprehensive specification approach, leading to better placement of VNFs and improved Quality of Service. Another benefit of formally specifying network services with QPNs are diverse analysis options, which allow valuable insights such as the distribution of end-to-end delay. We propose a tool-based workflow that supports the specification of network services and the automatic generation of corresponding simulation code to enable an in-depth analysis of their behavior and performance.

As 5G and network function virtualization (NFV) are maturing, it becomes crucial to demonstrate their feasibility and benefits by means of vertical scenarios. While 5GPPP has identified smart manufacturing as one of the most important vertical industries, there is still a lack of specific, practical use cases. Using the experience from a large-scale manufacturing company, Weidm{\"u}ller Group, we present a detailed use case that reflects the needs of real-world manufacturers. We also propose an architecture with specific network services and virtual network functions (VNFs) that realize the use case in practice. As a proof of concept, we implement the required services and deploy them on an emulation-based prototyping platform. Our experimental results indicate that a fully virtualized smart manufacturing use case is not only feasible but also reduces machine interconnection and configuration time and thus improves productivity by orders of magnitude.

5G together with software defined networking (SDN) and network function virtualisation (NFV) will enable a wide variety of vertical use cases. One of them is the smart man- ufacturing case which utilises 5G networks to interconnect production machines, machine parks, and factory sites to enable new possibilities in terms of flexibility, automation, and novel applications (industry 4.0). However, the availability of realistic and practical proof-of-concepts for those smart manufacturing scenarios is still limited. This demo fills this gap by not only showing a real-world smart manufacturing application entirely implemented using NFV concepts, but also a lightweight prototyping framework that simplifies the realisation of vertical NFV proof-of-concepts. Dur- ing the demo, we show how an NFV-based smart manufacturing scenario can be specified, on-boarded, and instantiated before we demonstrate how the presented NFV services simplify machine data collection, aggregation, and analysis.

By distributing the computational load over the nodes of a Wireless Acoustic Sensor Network (WASN), the real-time capability of the TRINICON (TRIple-N-Independent component analysis for CONvolutive mixtures) framework for Blind Source Separation (BSS) can be ensured, even if the individual network nodes are not powerful enough to run TRINICON in real-time by themselves. To optimally utilize the limited computing power and data rate in WASNs, the MARVELO (Multicast-Aware Routing for Virtual network Embedding with Loops in Overlays) framework is expanded for use with TRINICON, while a feature-based selection scheme is proposed to exploit the most beneficial parts of the input signal for adapting the demixing system. The simulation results of realistic scenarios show only a minor degradation of the separation performance even in heavily resource-limited situations.

Internet of Things (IoT) applications witness an exceptional evolution of traffic demands, while existing protocols, as seen in wireless sensor networks (WSNs), struggle to cope with these demands. Traditional protocols rely on finding a routing path between sensors generating data and sinks acting as gateway or databases. Meanwhile, the network will suffer from high collisions in case of high data rates. In this context, in-network processing solutions are used to leverage the wireless nodes' computations, by distributing processing tasks on the nodes along the routing path. Although in-network processing solutions are very popular in wired networks (e.g., data centers and wide area networks), there are many challenges to adopt these solutions in wireless networks, due to the interference problem. In this paper, we solve the problem of routing and task distribution jointly using a greedy Virtual Network Embedding (VNE) algorithm, and consider power control as well. Through simulations, we compare the proposed algorithm to optimal solutions and show that it achieves good results in terms of delay. Moreover, we discuss its sub-optimality by driving tight lower bounds and loose upper bounds. We also compare our solution with another wireless VNE solution to show the trade-off between delay and symbol error rate.

One of the major challenges in implementing wireless virtualization is the resource discovery. This is particularly important for the embedding-algorithms that are used to distribute the tasks to nodes. MARVELO is a prototype framework for executing different distributed algorithms on the top of a wireless (802.11) ad-hoc network. The aim of MARVELO is to select the nodes for running the algorithms and to define the routing between the nodes. Hence, it also supports monitoring functionalities to collect information about the available resources and to assist in profiling the algorithms. The objective of this demo is to show how MAVRLEO distributes tasks in an ad-hoc network, based on a feedback from our monitoring tool. Additionally, we explain the work-flow, composition and execution of the framework.

More and more management and orchestration approaches for (software) networks are based on machine learning paradigms and solutions. These approaches depend not only on their program code to operate properly, but also require enough input data to train their internal models. However, such training data is barely available for the software networking domain and most presented solutions rely on their own, sometimes not even published, data sets. This makes it hard, or even infeasible, to reproduce and compare many of the existing solutions. As a result, it ultimately slows down the adoption of machine learning approaches in softwarised networks. To this end, we introduce the "softwarised network data zoo" (SNDZoo), an open collection of software networking data sets aiming to streamline and ease machine learning research in the software networking domain. We present a general methodology to collect, archive, and publish those data sets for use by other researches and, as an example, eight initial data sets, focusing on the performance of virtualised network functions.

Offloading packet processing tasks to programmable switches and/or to programmable network interfaces, so called “SmartNICs”, is one of the key concepts to prepare softwarized networks for the high traffic demands of the future. However, implementing network functions that make use of those offload- ing technologies is still challenging and usually requires the availability of specialized hardware. It becomes even harder if heterogeneous services, making use of different offloading and network virtualization technologies, should be developed. In this paper, we introduce FOP4 (Function Offloading Pro- totyping with P4), a novel prototyping platform that allows to prototype heterogeneous software network scenarios, including container-based, P4-switch-based, and SmartNIC-based network functions. The presented work substantially extends our existing Containernet platform with the means to prototype offloading scenarios. Besides presenting the platform’s system design, we evaluate its scalability and show that it can run scenarios with more than 64 P4 switch or SmartNIC nodes on a single laptop. Finally, we presented a case study in which we use the presented platform to prototype an extended in-band network telemetry use case.

Emulation platforms supporting Virtual Network Functions (VNFs) allow developers to rapidly prototype network services. None of the available platforms, however, supports experimenting with programmable data planes to enable VNF offloading. In this demonstration, we show FOP4, a flexible platform that provides support for Docker-based VNFs, and VNF offloading, by means of P4-enabled switches. The platform provides interfaces to program the P4 devices and to deploy network functions. We demonstrate FOP4 with two complex example scenarios, demonstrating how developers can exploit data plane programmability to implement network functions.

Given the recent development in embedded devices, wireless senor nodes are no longer limited to data collection but they can also do processing (e.g., smartphones). Accordingly, new types of applications take an advantage of the processing and flexibility provided by the wireless network. A common property between these applications is that the processing is not running on only one single node, but it is broken-down into smaller tasks that can run over multiple nodes, i.e., exploiting the in-network processing. We study a special variant of in-network processing, where the application is given by a graph; the processing tasks have predefined connections to be executed in a predefined sequence. The problem of embedding an application graph into a network is commonly known as Virtual Network Embedding (VNE). In this paper, we present a Genetic Algorithm (GA) solution to solve this wireless VNE problem, where we take into account the interference and multi-cast properties. We show that the GA has a good performance and fast execution compared to the optimization problem.

Building on 5G and network function virtualization (NFV), smart manufacturing has the potential to drastically increase productivity, reduce cost, and introduce novel, flexible manufacturing services. Current work mostly focuses on high-level scenarios or emulation-based prototype deployments. Extending our previous work, we showcase one of the first cloud-native 5G verticals focusing on the deployment of smart manufacturing use cases on production infrastructure. In particular, we use the 5GTANGO service platform to deploy our developed network services on Kubernetes. For this demo, we implemented a series of cloud-native virtualized network functions (VNFs) and created suitable service descriptors. Their light-weight, stateless deployment on Kubernetes enables quick instantiation, scalability, and robustness.

Understanding the behavior of the components of service function chains (SFCs) in different load situations is important for efficient and automatic management and orches- tration of services. For this purpose and for practical research in network function virtualization in general, there is a great need for benchmarks and experimental data. In this paper, we describe our experiments for characterizing the relationship between resource demands of virtual network functions (VNFs) and the expected performance of the SFC, considering the individual performance of the VNFs as well as the interdependencies among VNFs within the SFC. We have designed our experiments focusing on video streaming, an important application in this context. We present examples of models for predicting the interdependence between resource demands and performance characteristics of SFCs using support vector regression and polynomial regression models. We also show practical evidence from our experiments that VNFs need to be benchmarked in their final chain setup, rather than individually, to capture important interdependencies that affect their performance. The data gathered from our experiments is publicly available.

Network function virtualization requires scaling and placement, deciding the number and the location of function instances. Current approaches are limited in flexibility and practical applicability. Specifically, we study dynamic, single-step, joint scaling and placement of network services with bidirectional flows traversing Physical or Virtual Network Functions (VNFs) and returning to their sources. We develop models to support stateful components and legacy network functions with fixed locations in these network services as well as the possibility of reusing VNFs across network services. We formalize the problem of jointly scaling and placing such network services as a mixed- integer linear program (MILP). We show that this problem is NP-complete and also present a heuristic algorithm to find good solutions in short time. In an extensive evaluation with realistic scenarios, we investigate the capabilities of the two approaches.

Dynamically steering flows through virtualized net- work function instances is a key enabler for elastic, on-demand deployments of virtualized network functions. This becomes par- ticular challenging when stateful functions are involved, necessi- tating state management. The problem with existing solutions is that they typically embrace state migration and flow rerouting jointly, imposing a huge set of requirements on the on-boarded VNFs, e.g., solution-specific state management interfaces. In this paper, we introduce the seamless handover proto- col (SHarP). It provides an easy-to-use, loss-less, and order- preserving flow rerouting mechanism that is not fixed to a single state management approach. This allows VNF vendors to implement or use the state management solution of their choice. SHarP supports these solutions with additional information when flows are migrated. Further, we show how SHarP significantly reduces the buffer usage at a central (SDN) controller, which is a typical bottleneck in existing solutions. Our experiments show that SHarP uses a constant amount of controller buffer, irrespective of the time taken to migrate the VNF state.

Developing a virtualized network service does not only involve the implementation and configuration of the network functions it is composed of but also its integration and test with management solutions that will control the service in its production environment. These integration tasks require testbeds that offer the needed network function virtualization infrastructure~(NFVI), like OpenStack, introducing a lot of management and maintenance overheads. Such testbed setups become even more complicated when the multi point-of-presence~(PoP) case, with multiple infrastructure installations, is considered. In this demo, we showcase an emulation platform that executes containerized network services in user-defined multi-PoP topologies. The platform does not only allow network service developers to locally test their services but also to connect real-world management and orchestration solutions to the emulated PoPs. During our interactive demonstration we focus on the integration between the emulated infrastructure and state-of-the-art orchestration solutions like SONATA or OSM.

Management and orchestration~(MANO) systems are the key components of future large-scale NFV environments. They will manage resources of hundreds or even thousands of NFV infrastructure installations, so called points of presence~(PoP). Such scenarios need to be automatically tested during the development phase of a MANO system. This task becomes very challenging because large-scale NFV testbeds are hard to maintain, too expensive, or simply not available. In this paper, we present a multi-PoP NFV infrastructure emulation platform that enables automated, large-scale testing of MANO stacks. We show that our platform can easily emulate hundreds of PoPs on a single physical machine and reduces the setup time of a test PoP by a factor of 232x compared to a DevStack-based test PoP installation. Further, we present a case study in which we test ETSI's Open Source MANO~(OSM) against our proposed system to gain insights about OSM's behaviour in large-scale NFV deployments.

Signal processing in WASNs is based on a software framework for hosting the algorithms as well as on a set of wireless connected devices representing the hardware. Each of the nodes contributes memory, processing power, communication bandwidth and some sensor information for the tasks to be solved on the network. In this paper we present our MARVELO framework for distributed signal processing. It is intended for transforming existing centralized implementations into distributed versions. To this end, the software only needs a block-oriented implementation, which MARVELO picks-up and distributes on the network. Additionally, our sensor node hardware and the audio interfaces responsible for multi-channel recordings are presented.

Dynamic allocation of resources is a key feature in network function virtualization (NFV), enabling flexible adjustment of slices and contained network services to ever-changing service demands. Considering resource allocation across the entire network, many authors have proposed approaches to optimize the placement and chaining of virtual network function (VNF) instances and the allocation of resources to these VNF instances. In doing so, various optimization objectives are conceivable, e.g., minimizing certain required resources or the end-to-end delay of the placed services. In this paper, we investigate the relationship between four typical optimization objectives when coordinating the placement and resource allocation of chained VNF instances. We observe an interesting trade-off between minimizing the overhead of starting/stopping VNF instances and all other objectives when adapting to changed service demands.

In recent years, a variety of different approaches have been proposed to tackle the problem of scaling and placing network services, consisting of interconnected virtual network functions (VNFs). This paper presents a placement abstraction layer (PAL) that provides a clear and simple northbound interface for using such algorithms while hiding their internal functionality and implementation. Through its southbound interface, PAL can connect to different back ends that evaluate the calculated placements, e.g., using simulations, emulations, or testbed approaches. As an example for such evaluation back ends, we introduce a novel placement emulation framework (PEF) that allows executing calculated placements using real, containerbased VNFs on real-world network topologies. In a case study, we show how PAL and PEF facilitate reusing and evaluating placement algorithms as well as validating their underlying models and performance claims.

A key challenge of network function virtualization (NFV) is the complexity of developing and deploying new network services. Currently, development requires many manual steps that are time-consuming and error-prone (e.g., for creating service descriptors). Furthermore, existing management and orchestration (MANO) platforms only offer limited support of standardized descriptor models or package formats, limiting the re-usability of network services. To this end, we introduce a fully integrated, open-source NFV service development kit (SDK) with multi-MANO platform support. Our SDK simplifies many NFV service development steps by offering initial generation of descriptors, advanced project management, as well as fully automated packaging and submission for on-boarding. To achieve multi-platform support, we present a package format that extends ETSI’s VNF package format. In this demonstration, we present the end-to-end workflow to develop an NFV service that is then packaged for multiple platforms, i.e., 5GTANGO and OSM.

In conventional large-scale networks, creation and management of network services are costly and complex tasks that often consume a lot of resources, including time and manpower. Network softwarization and network function virtualization have been introduced to tackle these problems, aiming at decreasing costs and complexity of implementing new services, maintaining the implemented services, and managing available resources in service provisioning platforms and underlying infrastructures. To experience the full potential of these approaches, innovative development support tools and service provisioning environments are needed. To answer these needs, we introduce the architecture of the open-source SONATA system, a service programming, orchestration, and management framework. We present a development toolchain for virtualized network services, fully integrated with a service platform and orchestration system. We introduce the modular and flexible architecture of our system and discuss its main components and features, such as function- and service-specific managers that allow fine-grained service management, slicing support to facilitate multi-tenancy, recursiveness for improved scalability, and full-featured DevOps support.

Developing a virtualized network service does not only involve the implementation and configuration of the network functions it is composed of but also its integration and test with management solutions that will control the service in its production environment. These integration tasks require testbeds that offer the needed network function virtualization infrastructure~(NFVI), like OpenStack, introducing a lot of management and maintenance overheads. Such testbed setups become even more complicated when the multi point-of-presence~(PoP) case, with multiple infrastructure installations, is considered. In this demo, we showcase an emulation platform that executes containerized network services in user-defined multi-PoP topologies. The platform does not only allow network service developers to locally test their services but also to connect real-world management and orchestration solutions to the emulated PoPs. During our interactive demonstration we focus on the integration between the emulated infrastructure and state-of-the-art orchestration solutions like SONATA or OSM.

Management of complex network services requires flexible and efficient service provisioning as well as optimized handling of continuous changes in the workload of the service.To adapt to changes in the demand, service components need to be replicated (scaling) and allocated to physical resources (placement) dynamically. In this paper, we propose a fullyautomated approach to the joint optimization problem of scaling and placement, enabling quick reaction to changes. We formalize the problem, analyze its complexity, and develop two algorithms to solve it. Extensive empirical results show the applicability andeffectiveness of the proposed approach.

Benchmarking and profiling virtual network functions (VNFs) generates input knowledge for resource management decisions taken by management and orchestration systems. Such VNFs are usually not executed in isolation but are often deployed as part of a service function chain (SFC) that connects single functions into complex structures. To manage such chains, isolated performance profiles of single functions have to be combined to get insights into the overall behavior of an SFC. This becomes particularly challenging in highly agile DevOps environments in which profiling processes need to be fully automated and detailed insights about a chain's internal structures are not always available. In this paper, we introduce a fully automatable, flexible, and platform-agnostic profiling system that allows to profile entire SFCs at once. This obviates manual modeling procedures to combine profiling results from single VNFs to reflect SFC performance. We use a case study with different SFC configurations to show that it is hard to model the resulting SFC performance based on single-VNF measurements and that performance interactions between real, non-trivial functions that are deployed in a chain exist.

Allocating resources to virtualized network functions and services to meet service level agreements is a challenging task for NFV management and orchestration systems. This becomes even more challenging when agile development methodologies, like DevOps, are applied. In such scenarios, management and orchestration systems are continuously facing new versions of functions and services which makes it hard to decide how much resources have to be allocated to them to provide the expected service performance. One solution for this problem is to support resource allocation decisions with performance behavior information obtained by profiling techniques applied to such network functions and services. In this position paper, we analyze and discuss the components needed to generate such performance behavior information within the NFV DevOps workflow. We also outline research questions that identify open issues and missing pieces for a fully integrated NFV profiling solution. Further, we introduce a novel profiling mechanism that is able to profile virtualized network functions and entire network service chains under different resource constraints before they are deployed on production infrastructure.

Traditional cellular networks are forced to remain active regardless of the actual amount of traffic that is currently produced/requested, with a clear waste of energy. Two-layer mobile networks with separated signalling and data layers have been recently proposed for energy savings in future implementations. These networks are able to switch off unneeded data cells completely while maintaining full coverage with their signalling cells, thus saving energy. In this demonstration, we showcase a testbed that uses Wi-Fi access points to emulate small cells of the data layer and a publicly available cellular connection as the signalling layer. We use off-the-shelf Android smartphones with an ad-hoc networking management module and a MultiPath TCP-enabled kernel to manage the Wi-Fi and cellular interfaces simultaneously. The testbed is used to demonstrate the general feasibility of this layered architecture and to facilitate experiments with network-wide resource optimization.

Elastic deployments of virtualized network functions~(VNF) can automatically scale the amount of used resources in relation to their workload. This is often done by starting new VNF instances or stopping old ones. A problem of these scale operations is that most network functions are stateful and their internal state is not automatically migrated when traffic is redistributed in the deployment. As a result, mechanisms are needed to exchange or migrate internal network function state between VNF instances. This paper presents a state management framework that creates a logically distributed state store on top of elastically deployed virtual network functions. We also introduce a novel programming model that provides both a local and a global view of the state to each VNF instance. We discuss the integration of our framework into existing network function virtualization architectures and compare the performance of our prototype to a centralized and a distributed state store solution.

Virtualized network services consisting of multiple individual network functions are already today deployed across multiple sites, so called multi-PoP (points of presence) environments. This allows to improve service performance by optimizing its placement in the network. But prototyping and testing of these complex distributed software systems becomes extremely challenging. The reason is that not only the network service as such has to be tested but also its integration with management and orchestration systems. Existing solutions, like simulators, basic network emulators, or local cloud testbeds, do not support all aspects of these tasks. To this end, we introduce MeDICINE, a novel NFV prototyping platform that is able to execute production-ready network functions, provided as software containers, in an emulated multi-PoP environment. These network functions can be controlled by any third-party management and orchestration system that connects to our platform through standard interfaces. Based on this, a developer can use our platform to prototype and test complex network services in a realistic environment running on his laptop.

Network function virtualization and software-defined networking allow services consisting of virtual network functions to be designed and implemented with great flexibility by facilitating automatic deployments, migrations, and reconfigurations for services and their components. For extended flexibility, we go beyond seeing services as a fixed chain of functions. We present a YANG model for describing the service structure in deployment requests in a flexible way that enables changing the order of functions in case the order of traversing them does not affect the functionality of the service. Upon receiving such requests, the network orchestration system can choose the optimal composition of service components that gives the best results for placement of services in the network. This introduces new complexities to the placement problem by greatly increasing the number of possible ways a service can be composed. In this paper, we describe a heuristic solution that selects a Pareto set of the possible compositions of a service as well as possible combinations of different services, with respect to different resource requirements of the services. Our evaluations show that the selected combinations consist of representative samples of possible structures and requirements and therefore, can result in optimal or close-to-optimal placement results.

Video streaming is in high demand by mobile users. In cellular networks, however, the unreliable wireless channel leads to two major problems. Poor channel states degrade video quality and interrupt the playback when a user cannot sufficiently fill its local playout buffer: buffer underruns occur. In contrast, good channel conditions cause common greedy buffering schemes to buffer too much data. Such over-buffering wastes expensive wireless channel capacity. Assuming that we can anticipate future data rates, we plan the quality and download time of video segments ahead. This anticipatory download scheduling avoids buffer underruns by downloading a large number of segments before a drop in available data rate occurs, without wasting wireless capacity by excessive buffering.We developed a practical anticipatory scheduling algorithm for segmented video streaming protocols (e.g., HLS or MPEG DASH). Simulation results and testbed measurements show that our solution essentially eliminates playback interruptions without significantly decreasing video quality.

The size of modern data centers is constantly increasing. As it is not economic to interconnect all machines in the data center using a full-bisection-bandwidth network, techniques have to be developed to increase the efficiency of data-center networks. The Software-Defined Network paradigm opened the door for centralized traffic engineering (TE) in such environments. Up to now, there were already a number of TE proposals for SDN-controlled data centers that all work very well. However, these techniques either use a high amount of flow table entries or a high flow installation rate that overwhelms available switching hardware, or they require custom or very expensive end-of-line equipment to be usable in practice. We present HybridTE, a TE technique that uses (uncertain) information about large flows. Using this extra information, our technique has very low hardware requirements while maintaining better performance than existing TE techniques. This enables us to build very low-cost, high performance data-center networks.

Multi-rooted trees are becoming the norm for modern data-center networks. In these networks, scalable flow routing is challenging owing to vast number of flows. Current approaches either employ a central controller that can have scalability issues or a scalable decentralized algorithm only considering local information. In this paper we present a new decentralized approach to least-congested path routing in software-defined data center networks that has neither of these issues: By duplicating the initial (or SYN) packet of a flow and estimating the data rate of multiple flows in parallel, we exploit TCP’s habit to fill buffers to find the least congested path. We show that our algorithm significantly improves flow completion time without the need for a central controller or specialized hardware.

The increasing amount of mobile traffic leads to a significantly higher energy consumption of mobile networks that is mainly caused by the high number of required base stations. One recent solution for this is based on a two-layered network that uses long-range macro cells to provide a full coverage signaling overlay and short-range small cells for fast data transmissions. These small cells can be switched off when they are not needed and allow network-wide energy optimizations. This paper presents an architecture that extends existing mobile networks to integrate a small cell layer that supports on-demand cell activation. We discuss how additional small cells can be interconnected with existing core components and how they can be controlled by a resource management component. Finally, a Wi-Fi based proof of concept testbed implementation is presented that demonstrates the feasibility of the approach.

Network emulations are widely used for testing novel network protocols and routing algorithms in realistic scenarios. Up to now, there is no emulation tool that is able to emulate large software-defined data center networks that consist of several thousand nodes. Mininet is the most common tool to emulate Software-Defined Networks of several hundred nodes. We extend Mininet to span an emulated network over several physical machines, making it possible to emulate networks of several thousand nodes on just a handful of physical machines. This enables us to emulate, e.g., large data center networks. To test this approach, we additionally introduce a traffic generator for data center traffic. Since there are no data center traffic traces publicly available we use the results of two recent traffic studies to create synthetic traffic. We show the design and discuss some challenges we had in building our traffic generator. As a showcase for our work we emulated a data center consisting of 3200 hosts on a cluster of only 12 physical machines. We show the resulting workloads and the trade-offs involved.

Preemptive Routing and Wavelength Assignment (RWA) algorithms preempt established lightpaths in case notenough resources are available to set up a new lightpath in aWavelength Division Multiplexing (WDM) network. The selectionof lightpaths to be preempted relies on internal decisions of theRWA algorithm. Thus, if dedicated properties of the networktopology are required by the applications running on the network,these requirements have to be known to the RWA algorithm.We present a family of preemptive RWA algorithms for WDMnetworks. These algorithms have two distinguishing features: a)they can handle dynamic traffic by on-the-fly reconfiguration,and b) users can give feedback for reconfiguration decisions andthus influence the preemption decision of the RWA algorithm,leading to networks which adapt directly to application needs.This is different from traffic engineering where the network is(slowly) adapted to observed traffic patterns.Our algorithms handle various WDM network configurationsincluding networks consisting of heterogeneous WDM hardware.To this end, we are using the layered graph approach togetherwith a newly developed graph model that is used to determineconflicting lightpaths.

In distributed cloud computing, application deployment across multiple sites can improve quality of service. Recent research developed algorithms to find optimal locations for virtual machines. However, those algorithms assume to have either single-tier applications or a fixed number of virtual machines – a strong simplification of reality. This paper investigates the placement and scaling of complex application architectures. An application is dynamically scaled to fit both the current demand situation and the currently available infrastructure resources. We compare two approaches: The first one is based on virtual network embedding. The second approach is a novel method called Template Embedding. It is based on a hierarchical 1-allocation hub flow problem and combines applica- tion scaling and embedding in one step. Extensive experiments on 43200 network configurations showed that Template Embedding outperforms virtual network embedding in all cases in three metrics: success rate, solution quality, and runtime. This positive result shows that template embedding is a promising approach for distributed cloud resource allocation.

Network appliances perform different functions on network flows and constitute an important part of an operator’s network. Normally, a set of chained network functions process network flows. Following the trend of virtualization of networks, virtualization of the network functions has also become a topic of interest. We define a model for formalizing the chaining of network functions using a context-free language. We process deployment requests and construct virtual network function graphs that can be mapped to the network. We describe the mapping as a Mixed Integer Quadratically Constrained Program (MIQCP) for finding the placement of the network functions and chaining them together considering the limited network resources and requirements of the functions. We have performed a Pareto set analysis to investigate the possible trade-offs between different optimization objectives.

In the near future many more compute resources will be available at different geographical locations. To minimize the response time of requests, application servers closer to the user can hence be used to shorten network round trip times. However, this advantage is neutralized if the used data centre is highly loaded as the processing time of re- quests is important as well. We model the request response time as the network round trip time plus the processing time at a data centre.We present a capacitated facility location problem formal- ization where the processing time is modelled as the sojourn time of a queueing model. We discuss the Pareto trade-off between the number of used data centres and the resulting response time. For example, using fewer data centres could cut expenses but results in high utilization, high response time, and smaller revenues.Previous work presented a non-linear cost function. We prove its convexity and exploit this property in two ways: First, we transform the convex model into a linear model while controlling the maximum approximation error. Sec- ond, we used a convex solver instead of a slower non-linear solver. Numerical results on network topologies exemplify our work.

This paper explores how cloud provider competition influences instance pricing in an IaaS (Infrastructure-as-a-Service) market. When reserved instance pricing includes an on-demand price component in addition to a reservation fee (two-part tariffs), different providers might offer different price combinations, where the client’s choice depends on its load profile. We investigate a duopoly of providers and analyze stable market prices in two-part tariffs. Further, we study offers that allow a specified amount of included usage (three-part tariffs). Neither two-part nor three-part tariffs produce an equilibrium market outcome other than a service pricing that equals production cost, i.e., complex price structures do not significantly affect the results from ordinary Bertrand competition.