GPUs are increasingly used to accelerate ML applications. Often, they are used to offload inference tasks, which are usually latency-critical. But the arrival patterns of inference requests are often bursty and include periods without any load. Furthermore, inference tasks may not be able to fully utilize the compute resources of a GPU, even with larger batch sizes. Consequently, the average utilization of a GPU that is exclusively used for an inference service is low. Industry has recognized the problem of underutilization and offers solutions to co-locate applications, improving utilization and cost-efficiency. We show, however, that these state-of-the-art solutions only maintain low inference latency when their compute resources are significantly overprovisioned. We propose Draco, a system to co-locate latency-critical inference tasks with a batch job, e.g., training an ML model, without violating the inference latency requirements. Draco autonomously estimates the resource requirements of inference tasks and detects periods with low load by perodically sampling the GPU's performance monitoring unit (PMU). Furthermore, Draco manages a pool of streaming multiprocessor (SM) partitions and dynamically assigns them to inference tasks to meet latency requirements. Leftover resources are granted to the co-located batch job. Depending on the workload, Draco increases throughput of the batch job by up to$4 x$compared to industrial solutions while keeping inference latency low.

In resource-constrained cloud systems, e.g., at the network edge or in private clouds, serverless computing is increasingly adopted to deploy microservices-based applications, leveraging its promised high resource efficiency. Provisioning resources to serverless services, however, poses several challenges, due to the high cold-start latency of containers and stringent Service Level Agreement (SLA) requirements of the microservices. In response, we investigate the behavior of containers in different states (i.e., running, warm, or cold) and exploit our experimental observations to formulate an optimization problem that minimizes the energy consumption of the active servers while reducing SLA violations. In light of the problem complexity, we propose a low-complexity algorithm, named AiW, which utilizes a multi-queueing approach to balance energy consumption and system performance by reusing containers effectively and invoking cold-starts only when necessary. To further minimize the energy consumption of data centers, we introduce the two-timescale COmputing resource Management at the Edge (COME) framework, comprising an orchestrator running our proposed AiW algorithm for container provisioning and Dynamic Server Provisioner (DSP) for dynamically activating/deactivating servers in response to AiW’s decisions on request scheduling. COME addresses the mismatch in timescales for resource provisioning decisions at the container and server levels. Extensive performance evaluation through simulation shows AiW’s close match to the optimum and COME’s significant reduction in power consumption by 22–64% compared state-of-the-art alternatives.

6G networks will be highly dynamic, reconfigurable, and resilient. To enable and support such features, employing AI has been suggested. Integrating AI in networks will likely require distributed AI deployments with resilient connectivity, e.g., for communication between RL agents and environment. Such approaches need to be validated in realistic network environments. In this demo, we use ContainerNet to emulate AI-capable and autonomic networks that employ the routing protocol KIRA to provide resilient connectivity and service discovery. As an example AI application, we train and infer deep RL agents learning medium access control (MAC) policies for a wireless network environment in the emulated network.

The Intelligent Transportation System (ITS) environment is known to be dynamic and distributed, where participants (vehicle users, operators, etc.) have multiple, changing and possibly conflicting objectives. Although Reinforcement Learning (RL) algorithms are commonly applied to optimize ITS applications such as resource management and offloading, most RL algorithms focus on single objectives. In many situations, converting a multi-objective problem into a single-objective one is impossible, intractable or insufficient, making such RL algorithms inapplicable. We propose a multi-objective, multi-agent reinforcement learning (MARL) algorithm with high learning efficiency and low computational requirements, which automatically triggers adaptive few-shot learning in a dynamic, distributed and noisy environment with sparse and delayed reward. We test our algorithm in an ITS environment with edge cloud computing. Empirical results show that the algorithm is quick to adapt to new environments and performs better in all individual and system metrics compared to the state-of-the-art benchmark. Our algorithm also addresses various practical concerns with its modularized and asynchronous online training method. In addition to the cloud simulation, we test our algorithm on a single-board computer and show that it can make inference in 6 milliseconds.

Macrodiversity is a key technique to increase the capacity of mobile networks. It can be realized using coordinated multipoint (CoMP), simultaneously connecting users to multiple overlapping cells. Selecting which users to serve by how many and which cells is NP-hard but needs to happen continuously in real time as users move and channel state changes. Existing approaches often require strict assumptions about or perfect knowledge of the underlying radio system, its resource allocation scheme, or user movements, none of which is readily available in practice. Instead, we propose three novel self-learning and self-adapting approaches using model-free deep reinforcement learning (DRL): DeepCoMP, DD-CoMP, and D3-CoMP. DeepCoMP leverages central control and observations of all users to select cells almost optimally. DD-CoMP and D3-CoMP use multi-agent DRL, which allows distributed, robust, and highly scalable coordination. All three approaches learn from experience and self-adapt to varying scenarios, reaching 2x higher Quality of Experience than other approaches. They have very few built-in assumptions and do not need prior system knowledge, making them more robust to change and better applicable in practice than existing approaches.

Serverless edge computing allows for highly efficient resource utilization, reducing the energy footprint of edge data centers. Indeed, the containers can be dynamically created and destroyed, allowing to adapt the workload to the available resources. Creating containers upon arrivals of service requests entails, however, a high start-up latency, which may be unsuitable for time-critical services. As alternative solution, pre-started containers (“warm containers”) are used to decrease start-up latency, but incurring in higher resource costs. In this work, we minimize the energy consumption of the active servers in the data center by optimally managing the various container states while meeting the target delay of the requested services. Further, in light of the problem complexity, we investigate how a simple threshold-based algorithm performs and show that it can closely match the optimum.

In resource-constrained cloud systems, e.g., at the network edge or in private clouds, it is essential to deploy microservices (MSs) efficiently. Unlike most of the existing approaches, we tackle this issue by accounting for two important facts: (i) the interference that arises when MSs compete for the same resources and degrades their performance, and (ii) the MSs’ deployment time. In particular, we first present some experiments highlighting the impact of interference on the throughput of MSs co-located in the same server, as well as the benefits of MSs’ parallel deployment. Then, we formulate an optimization problem that minimizes the number of used servers while meeting the MSs’ performance requirements. In light of the problem complexity, we design a low-complexity heuristic, called iPlace, that clusters together MSs competing for resources as diverse as possible and, hence, interfering as little as possible. Importantly, clustering MSs also allows us to exploit the benefit of parallel deployment, which greatly reduces the deployment time as compared to the sequential approach applied in prior art and by default in state-of-the-art orchestrators. Our numerical results show that iPlace closely matches the optimum and uses 21-92% fewer servers compared to alternative schemes while proving to be highly scalable. Further, by deploying MSs in parallel using Kubernetes, iPlace reduces the deployment time by 69% compared to state-of-the-art solutions.

Efficiently deploying microservices (MSs) is critical, especially in data centers at the edge of the network infrastructure where computing resources are precious. Unlike most of the existing approaches, we tackle this issue by accounting for the interference that arises when MSs compete for the same resources and degrades their performance. In particular, we first present some experiments highlighting the impact of interference on the throughput of co-located MSs. Then, we formulate an optimization problem that minimizes the number of used servers while meeting the MSs’ performance requirements. In light of the problem complexity, we design a low-complexity heuristic, called iPlace, that clusters together MSs competing for resources as diverse as possible and, hence, interfering as little as possible. Importantly, the choice of clustering MSs allows us to exploit the benefit of parallel MSs deployment, which, as shown by experimental evidence, greatly reduces the deployment time as compared to the sequential approach applied in prior art. Our numerical results show that iPlace closely matches the optimum and uses 10-63% fewer servers compared to alternative schemes, while proving to be highly scalable.

We formulate computation offloading as a decentralized decision-making problem with autonomous agents. We design an interaction mechanism that incentivizes agents to align private and system goals by balancing between competition and cooperation. The mechanism provably has Nash equilibria with optimal resource allocation in the static case. For a dynamic environment, we propose a novel multi-agent online learning algorithm that learns with partial, delayed and noisy state information, and a reward signal that reduces information need to a great extent. Empirical results confirm that through learning, agents significantly improve both system and individual performance, e.g., 40% offloading failure rate reduction, 32% communication overhead reduction, up to 38% computation resource savings in low contention, 18% utilization increase with reduced load variation in high contention, and improvement in fairness. Results also confirm the algorithm’s good convergence and generalization property in significantly different environments.

In many wireless multimedia applications, like speech enhancement or speaker localization, data sources need to be tightly synchronized. Synchronization is also a well-known problem for medium access control protocols of wireless sensor networks. In this paper, we unconventionally use the synchronization from a multimedia application to synchronize the wireless transmissions. From our analytical results, we derive a threshold for the synchronization accuracy below which the application synchronization could be used to drive a fixed-schedule time-division multiple access protocol. Above that threshold, Carrier Sense Multiple Access with Collision Avoidance is preferable and results in higher throughput. For evaluation, we simulate a synchronization algorithm for acoustic sensor networks and derive stability as well as high-throughput regions.

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.

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.

Two of the most important metrics when developing Wireless Sensor Networks (WSNs) applications are the Quality of Information (QoI) and Quality of Service (QoS). The former is used to specify the quality of the collected data by the sensors (e.g., measurements error or signal's intensity), while the latter defines the network's performance and availability (e.g., packet losses and latency). In this paper, we consider an example of wireless acoustic sensor networks, where we select a subset of microphones for two different objectives. First, we maximize the recording quality under QoS constraints. Second, we apply a trade-off between QoI and QoS. We formulate the problem as a constrained Markov Decision Problem (MDP) and solve it using reinforcement learning (RL). We compare the RL solution to a baseline model and show that in case of QoS-guarantee objective, the RL solution has an optimality gap up to 1%. Meanwhile, the RL solution is better than the baseline with improvements up to 23%, when using the trade-off objective.

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 1.05× more flows than the greedy flow scheduling heuristics under varying network load.

To address the vital problem of selecting the most useful microphones in wireless acoustic sensor networks, this paper proposes a novel, general-purpose approach that accounts for both acoustic and network aspects and remains application-agnostic for broad applicability. The inter-channel correlation of single-channel signal features, together with tools from spectral graph theory, is used to assess the usefulness from an acoustic perspective. By only transmitting the features that characterize signal frames as opposed to the full signal waveform, the unique constraints of wireless sensor networks are accommodated. The source-to-sink transmission delay, resulting from embedding a distributed signal processing application into a wireless network, captures the usefulness from a network perspective. The experiments demonstrate the efficacy of the proposed method for an exemplary multichannel signal processing application.

In this work we use autonomous vehicles to improve the performance of Wireless Sensor Networks (WSNs). In contrast to other autonomous vehicle applications, WSNs have two metrics for performance evaluation. First, quality of information (QoI) which is used to measure the quality of sensed data (e.g., measurement uncertainties or signal strength). Second, quality of service (QoS) which is used to measure the network’s performance for data forwarding (e.g., delay and packet losses). As a use case, we consider wireless acoustic sensor networks, where a group of speakers move inside a room and there are autonomous vehicles installed with microphones for streaming the audio data. We formulate the problem as a Markov decision problem (MDP) and solve it using Deep-Q-Networks (DQN). Additionally, we compare the performance of DQN solution to two different real-world implementations: speakers holding/passing microphones and microphones being preinstalled in fixed positions.We show using simulations that the performance of autonomous vehicles in terms of QoI and QoS is better than the real-world implementation in some scenarios. Moreover, we study the impact of the vehicles speed on the learning process of the DQN solution and show how low speeds degrade the performance. Finally, we compare the DQN solution to a heuristic one and provide theoretical analysis of the performance with respect to dynamic WSNs.

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).

Modern services consist of interconnected components, e.g., microservices in a service mesh or machine learning functions in a pipeline. These services can scale and run across multiple network nodes on demand. To process incoming traffic, service components have to be instantiated and traffic assigned to these instances, taking capacities, changing demands, and Quality of Service (QoS) requirements 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 DeepCoord, 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 on 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, DeepCoord significantly improves flow throughput (up to 76%) and overall network utility (more than 2x) on real-world network topologies and traffic traces. It also supports optimizing multiple, possibly competing objectives, learns to respect QoS requirements, generalizes to scenarios with unseen, stochastic traffic, and scales to large real-world networks. For reproducibility and reuse, our code is publicly available.

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 1.1× more coflows than the Varys scheme while meeting their deadlines.

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 algorithms can be executed individually at each node in parallel and require 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 100×).

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.

Manufacturing companies can accomplish increased productivity by using data-driven learning methods based on production and machine data. To realise that big-data platforms are needed to simplify data provisioning and manipulation, for example to train machine learning algorithms. We introduce a method for the generation of big-data platform architectures. The method covers software selection which minimises implementation efforts and maximises the maintainability of used software solutions to implement architectures. We apply requirements from four large scale manufacturers to our method in a case study.

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. Our focus is on demonstrating the automatic network reconfiguration that is triggered by the IDS.

Network function virtualization (NFV) proposes to replace physical middleboxes with more flexible virtual network functions (VNFs). To dynamically adjust to ever-changing 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 under-allocation 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.

The Industry 4.0 sector is evolving in a tremendous pace by introducing a set of industrial automation mechanisms tightly coupled with the exploitation of Internet of Things (IoT), 5G and Artificial Intelligence (AI) technologies. By combining such emerging technologies, interconnected sensors, instruments, and other industrial devices are networked together with industrial applications, formulating the Industrial IoT (IIoT) and aiming to improve the efficiency and reliability of the deployed applications and provide Quality of Service (QoS) guarantees. However, in a 5G era, efficient, reliable and highly performant applications' provision has to be combined with exploitation of capabilities offered by 5G networks. Optimal usage of the available resources has to be realised, while guaranteeing strict QoS requirements such as high data rates, ultra-low latency and jitter. The first step towards this direction is based on the accurate profiling of vertical industries' applications in terms of resources usage, capacity limits and reliability characteristics. To achieve so, in this paper we provide an integrated methodology and approach for benchmarking and profiling 5G vertical industries' applications. This approach covers the realisation of benchmarking experiments and the extraction of insights based on the analysis of the collected data. Such insights are considered the cornerstones for the development of AI models that can lead to optimal infrastructure usage along with assurance of high QoS provision. The detailed approach is applied in a real IIoT use case, leading to profiling of a set of 5G network functions.

Upcoming sensing applications (acoustic or video) will have high processing requirements not satisfiable by a single node or need input from multiple sources (e.g., speaker localization). Offloading these applications to cloud or mobile edge is an option, but when running in a wireless senor network (WSN), it might entail needlessly high data rate and latency. An alternative is to spread processing inside the WSN, which is particularly attractive if the application comprises individual components. This scenario is typical for applications like acoustic signal processing. Mapping components to nodes can be formulated as wireless version of the NP-hard Virtual Network Embedding (VNE) problem, for which various heuristics exist. We propose a Reinforcement Learning (RL) framework, which relies on Q-Learning and uses either Greedy Epsilon or Epsilon Decay for exploration. We compare both exploration methods to the result of an optimization approach and show empirically that the RL framework achieves good results in terms of network delay within few number of steps.

As wireless sensor networks evolve towards comprising more powerful devices, they offer opportunities for distributing applications into the network and processing data in-network. Examples for such applications often come from the signal processing or data analysis domain. We describe such applications by overlay graphs, consisting of functional blocks with predefined interconnections that may even have feedback loops. We treat these functional blocks as virtual functions that can be easily moved (reprogrammed) among the network nodes. This property allows us to use the concept of virtual network embedding (VNE) for placement and routing. VNE has only partially been considered for wireless environments; specifically, the interaction between embedding and the wireless multicast advantage has not been fully explored. We cast our problem as mixed integer linear programming (MILP) formulation for uniform transmit power and mixed integer quadratic constraint programming (MIQCP) for power allocation and compare between both methods.

Services in Network Function Virtualization (NFV) can have a variety of requirements such as data rates, latencies, and cost that can change during the lifecycle of services. To meet these requirements, various hardware and software resources are suggested for implementing Virtualized Network Functions (VNFs). However, meeting all service requirements using one implementation option is not always possible. For example, to improve the performance of VNFs, using acceleration hardware is proposed. Although acceleration hardware can improve the performance of a network function, as they are expensive appliances, they increase the cost of services; this might not be desirable for a particular service user or load that can be handled by cheaper resources. Dynamically provisioning services can solve this problem in which different implementations of VNFs are switched on the fly as service requirements change. In this paper, we analyse this service provisioning approach in terms of performance, cost, and management overhead by experimenting an example VNF.

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 researchers and, as an example, eight initial data sets, focusing on the performance of virtualised network functions.

In recent years, the interest in interconnecting production machines has grown and new technologies such as augmented reality enter the shop floor. This evolution is driven by hyped topics such as Industry 4.0 and Internet of Things. These topics promise benefits through increasing transparency in factories, such as improvement in efficiency and reduced costs, e.g., due to data analysis. As a result, the demand on new concepts and technologies for factory networks is rising to fulfill the upcoming new requirements. Therefore, the development of 5G comes just in time. This paper is focused on one of the 5G core technologies network function virtualization (NFV). We propose two use cases that demonstrate how NFV enables flexible smart manufacturing. In NFV technology, virtual network functions (VNF) are composed to network services. We introduce the application of our vertical-specific network services that enable augmented reality on-demand and the flexible interconnection of production machines with services in company's cloud backend.

Network softwarization and the establishing solutions in the areas of virtualized resource management and orchestration increase the interoperability of heterogeneous, multi-domain, multi-technology infrastructures. Having different hosting platforms available, service providers can develop different deployment versions for services and virtual network functions, each optimized different resource types with different characteristics (e.g., CPUs vs. FPGAs). We formalize the problem of scaling, placement, and routing for heterogeneous services that consist of multi-version components and present a single-step optimization approach and a heuristic algorithm for solving it. We study the trade-offs between deployment costs and service performance and show that our solution approaches can adapt to different requirements, by instantiating different deployment versions of the service components in different locations.

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 manufacturing 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. During 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.

Given the recent development in embedded devices, wireless sensor 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.

When deploying resource-intensive signal processing applications in wireless sensor or mesh networks, distributing processing blocks over multiple nodes becomes promising. Such distributed applications need to solve the placement problem (which block to run on which node), the routing problem (which link between blocks to map on which path between nodes), and the scheduling problem (which transmission is active when). We investigate a variant where the application graph may contain feedback loops and we exploit wireless networks' inherent multicast advantage. Thus, we propose Multicast-Aware Routing for Virtual network Embedding with Loops in Overlays (MARVELO) to find efficient solutions for scheduling and routing under a detailed interference model. We cast this as a mixed integer quadratically constrained optimization problem and provide an efficient heuristic. Simulations show that our approach handles complex scenarios quickly.

The traffic demand in mobile access networks has grown substantially in recent years and is expected to continue to do so. The infrastructure of mobile access networks has to keep up with this trend and provide the data rates to satisfy the increasing demands. To achieve this, employing coordination mechanisms is essential to use available front-/backhaul resources efficiently. By exploiting recent network softwarization approaches such as SDN and NFV, these coordination mechanisms can be handled by virtualized control applications (CAs) that can be flexibly positioned in the network. In previous work, we have introduced the Flow processing-aware Control Application Placement Problem (FCAPP) to place these CAs appropriately in the backhaul network of a mobile access network. We have also presented a heuristic approach (FlexCAPF) that places and flexibly reassigns CAs fast and efficiently. But FlexCAPF works logically centralized, which might not be possible in every application scenario. In this work, we therefore provide DistCAPA - an alternative, distributed algorithm for tackling the same tasks as FlexCAPF.

One of the key issues in industrial-radio-based control applications is supporting their stringent requirements over a wireless link. In this paper, we propose a new model called Time-Limited Wireless Channel Capacity (TL-WChC) to describe the achievable data rate within a given duration of time and within a given violation probability. We establish a connection between fading characteristics of the wireless link with the channel capacity of our model and derive a closed-form solution; we apply it to Rayleigh fading channels as example. Our simulation results show the applicability and effectiveness of the proposed approach and model.

To adapt to continuously changing workloads in networks, components of the running network services may need to be replicated (scaling the network service) and allocated to physical resources (placement) dynamically, also necessitating dynamic re-routing of flows between service components. In this paper, we propose joint optimization of scaling, placement, and routing (JASPER), a fully automated approach to jointly optimizing scaling, placement, and routing for complex network services, consisting of multiple (virtualized) components. JASPER handles multiple network services that share the same substrate network; services can be dynamically added or removed and dynamic workload changes are handled. Our approach lets service designers specify their services on a high level of abstraction using service templates. JASPER automatically makes scaling, placement and routing decisions, 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 and effectiveness of the proposed approach.

The management of complex network services requires flexible and efficient service provisioning as well as optimized handling of continuous changes in the workload of the services. 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 fully automated 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. Empirical results show the applicability and effectiveness of the proposed approach.

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 realworld management and orchestration solutions to the emulated PoPs. During our interactive demonstration we focus on the integration between the emulated infrastructure and state-of-theart orchestration solutions like SONATA or OSM.

Software-Defined Networks (SDN) are constantly evolving and so is their software. One of the key advantages of SDN over traditional networks is the ability to rapidly develop and deploy new features. However, updating them often requires restarting the SDN controller and causes network downtime. In addition to such planned updates, unforeseen, accidental downtime is also a risk for SDN networks. While commercialized SDN controllers are adding mechanisms to deal with both planned and accidental downtime, they still are not competitive with conventional approaches, which typically use redundant hardware and special software to address these problems. In this paper we investigate how these two challenges to SDN can be addressed with dynamic reconfiguration and show how the state of the network can be managed by reconfiguration. Finally, we present a proof-of-concept implementation of our approach.

A current trend in networking and cloud computing is to provide compute resources at widely distributed sites; this is exemplified by developments such as network function virtualization. This paves the way for wide-area service deployments with improved service quality: e.g., user-perceived response times can be reduced by offering services at nearby sites. But always assigning users to the nearest site can be a bad decision if this site is already highly utilized. This paper formalizes two related decisions of allocating compute resources at different sites and assigning users to them with the goal of minimizing the response times while the total number of resources to be allocated is limited-a non-linear capacitated facility location problem with integrated queuing systems. To efficiently handle its non-linearity, we introduce five linear problem linearizations and adapt the currently best heuristic for a similar scenario to our scenario. All six approaches are compared in experiments for solution quality and solving time. Surprisingly, our best optimization formulation outperforms the heuristic in both time and quality. Additionally, we evaluate the influence of distributions of available compute resources in the network on the response time: the time was halved for some configurations. The presented formulation techniques for our problem linearizations are applicable to a broader optimization domain.

The Software-Defined Networking (SDN) ecosystem is still characterized by a multitude of different controller platforms, each with its own programming model, execution model, and capabilities. This creates a danger of a controller lock-in for both developers of SDN control applications and operators of SDN networks. Since no single controller platform appears to dominate the ecosystem for the foreseeable future, there is a need for portability of control applications between different platforms. We propose an architecture based on executing multiple instances of different controller platforms concurrently in a network to provide the SDN code the environment it was written for. It is built around a controller-independent network event routing element called Network Engine that provides composition and conflict resolution. Results obtained in realistic scenarios demonstrate the feasibility of the proposed approach, which increases both developer productivity and operational flexibility. A preliminary prototype of the architecture is available for testing as an open source project.

The fifth generation of wireless communication is expected to enable various new use cases that today's wireless communication systems are not able to support. One of these use cases is mission-critical machine-type communication (C-MTC) for automation processes in factory environments. The real-time requirements including firm deadlines for such communication impose very low latency and high reliability demands that can currently only be fulfilled by wired solutions. In this paper, we focus on the radio resource management aspect to enable C-MTC in the context of a factory automation scenario by proposing a joint real-time scheduling and interference coordination approach that supports low latency communication while provisioning radio resources for very high reliability.

To study the behavior between distributed applications, such as a cloud application, a model for the connection between the components of the application is needed. Various models have been introduced to generate Internet-like Autonomous System (AS) topologies. These models have been focused on replicating structural graph properties. One of the most promising models is the Positive Feedback Preference model (PFP). These models, however, lack the ability to predict routing paths and therefore realistic latency. We present a model to enrich the AS peering graph with peering points. Our model allows to calculating paths for the connections between the end system and to infer the latency from these paths. We introduce a new notion for the generation of our graph: the compactness of an AS. We introduce an algorithm based on the PFP algorithm to generate instances of the model. Verifying the generated model instances shows that the resulting latencies and the geographic properties match empirical data sets.

Research on accelerating big-data applications can be divided into job scheduling and flow scheduling. Job scheduling focuses on the timely and spacial placement of jobs on execution units. Flow scheduling, on the other hand, concentrates on routing of flows originating from actively running jobs. Although both job scheduling and flow scheduling work on accelerating big-data applications, their view on the problem and the available information is very different. We propose a new simulation tool to evaluate ideas that jointly solve the job and flow scheduling problem for big-data applications. Our tool combines the Yarn Scheduler Load Simulator with the distributed network emulator MaxiNet. With our work, the interdependency between the network and the jobs running on top of it can be included into the evaluation of new ideas, leveraging research on big-data applications with joint job and flow scheduling.

The growing number of network mechanisms results in a large amount of data flow processing to be performed in future wireless access networks. Determining suitable locations for data processing in a network is important for a good network performance but also a non-trivial task as data processing imposes many requirements on network resources. Further, data flows and network load generally change over time, possibly very quickly, so that a static, one-time placement is not an adequate solution. Hence, a framework that determines these locations should also flexibly reassign them over time. We address this need by a flexible flow processing-aware controller placement framework (FlexFCPF), which places and flexibly reassigns controller devices that are able to perform both network control (conventional SDN) and data flow processing for an efficient management of future wireless networks. We have developed and implemented a fast heuristic framework and we prove its efficiency by providing evaluation results of FlexFCPF's performance in a dynamic network simulation.

Future wireless access networks include a broad variety of applications. Examples for these range from network applications meant to optimize the usage of resources, e.g. Coordinated Multi-Point (CoMP) transmission, over coordinating network mechanisms for Inter-Cell Interference Coordination (ICIC) to virtualized applications to gain flexibility and save hardware costs such as Virtual Network Functions (VNFs). Still, there is one common thread for all these functions: the necessity for data flow processing. Data flow processing imposes many requirements on network resources and a good allocation of the locations for data processing in a network is a challenging task and crucial for network performance. Hence, data flow processing is a significant point to be considered when searching for efficient control architectures for future wireless networks. We propose a flow processing-aware controller placement framework, which introduces data flow processing to the research domain of future wireless network control and enhances the classic Software-Defined Networking (SDN) view of reducing network control to just packet forwarding. We have developed an optimization problem and a heuristic algorithm that verify the practical feasibility of flow processing-aware controller placement. Our evaluations reveal that the heuristic algorithm performs close to optimal, while reducing the runtime by about four orders of magnitude compared to the optimization problem and thus providing acceptable runtime for real-world application.

Preemptive Routing and Wavelength Assignment (RWA) algorithms preempt established lightpaths in case not enough resources are available to set up a new lightpath in a Wavelength Division Multiplexing (WDM) network. The selection of lightpaths to be preempted relies on internal decisions of the RWA algorithm. Thus, if dedicated properties of the network topology 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 WDM networks. 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 and thus 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 configurations including networks consisting of heterogeneous WDM hardware. To this end, we are using the layered graph approach together with a newly developed graph model that is used to determine conflicting lightpaths.

The traffic demand in wireless access networks has grown substantially in recent years and is expected to continue doing so, both in terms of total volume and data rate required by individual users. To overcome the current limitations of wireless access networks, operators already push for very dense and heterogeneous wireless networks, commonly known as DenseNets. Efficiently controlling DenseNets is a task for current research, as simply scaling existing networks by orders of magnitude brings along several problems, e.g. increased energy consumption and the explosion of signaling. Further, new innovations such as Network Functions Virtualization (NFV) which target the operation expenses for networks with a large variety of proprietary hardware appliances, reveal that data flow processing should not be neglected in this research domain. Software-Defined Networking (SDN), which separates the control plane and the data forwarding plane, has been identified as a promising approach for an efficient control of DenseNets. However, the controllers of classic SDN approaches are confined to the routing of data flows, while the support of sophisticated processing mechanisms, e.g. NFV or Coordinated Multi-Point transmission (CoMP), is not considered. Data flow processing imposes many requirements on data rate, latency and processing capacity. Hence, an adequate allocation of the locations for data processing in a network is a challenging task and crucial for network performance. In this paper we propose our flow processing-aware controller placement framework that takes into account data flow processing and control applications, both enhancing the SDN architecture specifically for DenseNets.

Saving energy in mobile networks can be achieved by intelligently deactivating basestations. One way to know when to activate or deactivate basestations is understanding user behavior. In this paper, we look at the segment download behavior of users of mobile video streaming in adaptive bit-rate streaming systems with segmented downloads. From a large trace of HTTP requests we extract mobile video sessions and their inter-download times. We evaluate if Hidden Markov Models are feasible to model and predict when a user will request segments in the future. We further analyse how choosing model parameters influences the prediction quality.

Today's video delivery solutions for mobile terminals often use the HTTP Live Streaming (HLS) protocol, which has two interesting features: firstly, the video is divided into playable segments of a certain length, which allows to download and buffer segments before they are required for playback. And secondly, those segments can be available in different quality levels, which can be selected according to the available transmission capacity. Combining these features with scheduling and channel capacity prediction of wireless transmissions into a cross-layer scheduling approach could improve the QoE for users by reducing playback interruptions and by providing the best possible video quality depending on the user's wireless channel capacity. In this paper we propose a mixed integer quadratically constrained program (MIQCP) to implement the aforementioned cross-layer scheduling approach. We evaluate its performance compared to greedy strategies to assess the potential for a future polynomial time implementation.

The process of planning a virtual topology for a Wavelength Devision Multiplexing (WDM) network is called Virtual Topology Design (VTD). The goal of VTD is to find a virtual topology that supports forwarding the expected traffic without congestion. In networks with fluctuating, high traffic demands, it can happen that no single topology fits all changing traffic demands occurring over a longer time. Thus, during operation, the virtual topology has to be reconfigured. Since modern networks tend to be large, VTD algorithms have to scale well with increasing network size, requiring distributed algorithms. Existing distributed VTD algorithms, however, react too slowly on congestion for the real-time reconfiguration of large networks. We propose Selfish Virtual Topology Reconfiguration (SVTR) as a new algorithm for distributed VTD. It combines reconfiguring the virtual topology and routing through a Software Defined Network (SDN). SVTR is used for online, on-the-fly network reconfiguration. Its integrated routing and WDM reconfiguration keeps connection disruption due to network reconfiguration to a minimum and is able to react very quickly to traffic pattern changes. SVTR works by iteratively adapting the virtual topology to the observed traffic patterns without global traffic information and without future traffic estimations. We evaluated SVTR by simulation and found that it significantly lowers congestion in realistic networks and high load scenarios.

In OpenFlow, multiple switches share the same control plane which is centralized at what is called the OpenFlow controller. A switch only consists of a forwarding plane. Rules for forwarding individual packets (called flow entries in OpenFlow) are pushed from the controller to the switches.

Managing interference by Coordinated Multi-Point (CoMP) transmission/reception is an effective mechanism to achieve high data rates in future cellular networks, like Long Term Evolution (LTE)-Advanced. For CoMP, sets of Base Stations (BSs) have to be selected to jointly serve User Equipments (UEs). These sets are typically selected based on wireless characteristics only. However, using CoMP also poses strict capacity and latency requirements on the backhaul network, which are difficult to fulfill even with future optical technologies. Hence, these requirements additionally need to be taken into account when deciding which BSs jointly serve a given UE. We have developed a BSs selection heuristic for CoMP that takes into account both aspects: the wireless channels and the backhaul network status. This heuristic can also identify, for a particular wireless channel situation, which bottlenecks in the backhaul network make a desired BSs selection infeasible. We exploit this to dynamically adapt the backhaul network to the wireless requirements. We call this network reconfiguration. Our simulations show that the heuristic's solution quality is close to the optimum while execution time and memory consumption are reduced by multiple orders of magnitude compared to solving the problem via mathematical optimization. This allows real-world deployment of the heuristic. In addition, we simulate the network reconfiguration in a future backhaul network scenario based on Passive Optical Networks (PONs). The results illustrate how our approach helps to better exploit available backhaul resources.

Wireless communication is prone to errors due to always changing channel conditions. Multiple transmissions over independent channels can be leveraged by combining erroneous packets. To be able to combine, reliability information for bits, so-called soft-bits, of the previous messages has to be saved. Especially for systems with limited storage capacity, such as sensor nodes, this poses the question of how detailed the stored reliability information has to be in order to achieve a desired performance. We compare different reliability information approximations, introduce two quantization options specifically tailored for combining, and compare their Bit Error Rate (BER) performance by simulation. We conclude that our quantization with logarithmic spacing using a recursive Log Likelihood Ratio (LLR) approximation shows the most promise, because it has the best BER results and the lowest implementation complexity.

In wireless communication, diversity can be exploited for improved robustness. For wireless sensor networks, cooperation diversity protects a transmission from a source to a destination node by having a relay node overhear and repeat the source's transmission. Additionally, the data readings of the source and the relay node can be similar. The relay's data can hence be interpreted as a distorted version of the source's data. We show how to use forward error correction techniques to employ these similarities to reduce errors in the source-relay transmission. Moreover, we propose a new protocol, Coded Cooperation with Similar Data (CCSD), to take advantage of similar data to improve the source-to-relay link of Coded Cooperation (CC). CCSD is a conservative extension of CC, i.e., if data of source and relay are not similar, CCSD behaves like CC. We evaluate the performance of CCSD by simulation under realistic, symbol-wise Rayleigh fading. A comparison with the established Selection Decode-and-Forward (SDF) cooperation protocol, CC, and direct transmission shows superiority of the new CCSD protocol, especially for situations with a bad source-to-relay channel, good relay-to-destination channel, and strong error correcting codes.

Using Coordinated Multi-Point (CoMP) transmission and reception techniques poses challenging latency and capacity requirements on the backhaul network infrastructure of cellular access systems. On a small scale, these requirements can be fulfilled by using upcoming optical technologies like Wavelength Division Multiplexing (WDM) Passive Optical Networks (PONs). On a metro scale, however, it is unclear to which extent CoMP is feasible. For a metro-wide cellular network deployment, we propose different backhaul network architectures and analyze their capability of fulfilling CoMP requirements. The analysis shows a trade-off between the ability of reusing existing metro network infrastructure and the area that can be covered using CoMP techniques. Based on this, we provide not only an understanding of which architecture approach fits best for a certain scenario, but also point out necessary hardware upgrades required for supporting CoMP in a desired target area.

Coordinated Multi-Point (CoMP) transmission/reception, and especially Joint Processing (JP), is a promising solution for managing interference in cellular mobile access networks. Its successful deployment, however, strongly depends on the capability of the backhaul infrastructure as strict capacity and latency requirements have to be fulfilled.

Exploiting base station cooperation in wireless mobile access networks leads to benefits in wireless transmission capacity, inter-cell interference management, and cell edge user experience. The clustering of cooperative Base Station (BS) sets, necessary for achieving the desired wireless performance, poses several challenges in the backhaul architecture design. This paper addresses the problem of placing and connecting controller/processing nodes within the backhaul infrastructure, which coordinate and/or process signals of cooperating base stations. We formulated a Mixed Integer Linear Programm (MILP) for this problem and a heuristic algorithm that approximates the optimal solution. While the heuristic's solution quality is close to the optimum, the runtime and memory requirements are multiple orders of magnitude lower compared to solving the MILP. This advantage allows to use the proposed heuristic either for backhaul/core network pre-planning or for on-the-fly network reconfiguration during ongoing mobile network operation.

Coordinated Multi-Point (CoMP) transmission/ reception is a promising solution for interference management in wireless cellular systems. Its successful deployment, however, strongly depends on the capability of the mobile backhaul network architecture to support the capacity, latency, and synchronization requirements of cooperation. In this paper, we deal with the “feasibility” aspects related to CoMP transmission/reception. We analyze how different backhaul topologies and technologies can support Base Station (BS) cooperation. We study, for different traffic scenarios and backhaul connectivity levels, which BS clusters are actually feasible compared to the ones desirable from the Radio Access Network (RAN) perspective. We found out that a significant mismatch exists between the desired wireless cluster, as defined by the RAN, and the feasible one, as allowed by the backhaul characteristics. Based on these findings, we explore different approaches to this problem, highlighting how the adoption of layer-2 switching techniques and multicast capabilities can already improve the cooperation feasibility. Finally, we propose an algorithm that includes the backhaul network feasibility information in the wireless cluster formation process. As a result, our system avoids unnecessary signaling and user data exchange among BSs which would have not been eligible for taking part in the desired cooperative cluster.

Present broadcast approaches for wireless multihop networks distribute packets quickly to all nodes (i.e., with low latency) by constructing small broadcast trees, thereby reducing the number of forwarding transmissions. While these trees are sufficient in non-fading environments, we show that they have a low delivery rate under fading. As a solution, we (1) incorporate the Rayleigh fading model directly into tree construction to re-obtain complete distribution with high probability. To still achieve low latency at the same time, we combine transmissions at individual nodes to exploit cooperation diversity. Since in broadcasts, a packet has to be retransmitted by nodes along the tree anyway, we do not have to pay the multiplexing loss which hampers cooperation diversity in the unicast case. Thus, we (2) additionally exploit cooperation diversity during tree construction to gain improved reliability while still keeping the size of the tree low. This enables us to significantly decrease the time for broadcasts while still distributing packets to all nodes under fading with high probability. To justify our heuristic approach, we (3) show that finding minimum latency cooperative broadcasts is NP-complete.

Structured Peer-to-Peer (P2P) systems are increasingly important for scalable data dissemination and search. Current distributed approaches for resolving complex search queries, like multi-attribute and range queries, typically require multiple query messages to resolve a single search request. To reduce the message overhead and the search latency, some approaches like the Multi-Attribute Addressable Network (MAAN) use static replication. However, this results in high main memory requirements and large data transfers each time a device joins the P2P network. Those drawbacks can be tolerated for P2P networks that mainly consist of fixed, powerful nodes like PCs but are intolerable for resource-constrained nodes with high churn, like mobile devices. As mobile devices will play a significant role in accessing and distributing data in the future, we propose and evaluate an improved search mechanism for such a scenario. Compared to MAAN, our approach significantly reduces the memory footprint and bandwidth requirements (up to a factor of five in our sample scenario). At the same time, the good latency properties of MAAN are remained on average. This is achieved via a dynamic replication scheme which introduces an adjustable trade-off between memory footprint and search latency. Thereby, our approach makes efficient, distributed resolution of complex queries in resource-constrained P2P networks feasible.

Rate adaptation based on signal-to-noise ratio (SNR) measurements is a common channel adaptation scheme to increase throughput in wireless communication systems. To use rate adaptation efficiently in cooperative wireless networks, an adaptation algorithm must consider multiple channels (source- destination, source-relays, and relays-destination) to select modulation and code rates that maximize throughput. In this paper we analyze the potential gains that combining cooperation with rate adaptation brings in three steps: (1) We derive the theoretical capacity bounds for ideal rate adaptation schemes for typical topologies. (2) We propose an offline heuristic for computing SNR thresholds aimed at reaching the derived bounds. (3) Using this heuristic, we compare rate adaptation for maximal ratio combining (MRC), where links are equally adapted, with soft-bit MRC (SBMRC), where links are individually adapted. We find that adapting the rate per link is superior in terms of throughput.

Interference limits throughput in wireless networks. To protect themselves against interference, many wireless protocols create areas around receivers in which no node is allowed to transmit. If such an exclusion area is small, more transmissions can proceed simultaneously but observe higher interference, creating a tradeoff between network capacity and link capacity.

Multiuser dynamic OFDMA for the downlink has been extensively studied, specifically, in terms of fast close-to- optimal subcarrier allocation heuristics and the efficient representation of signaling information. Although these functions provide the fundament of dynamic OFDMA, a complete multiuser OFDMA system requires more functionality. Focusing on WLAN systems, we discuss such additional functionality required for enabling the IEEE 802.11 link and Medium Access Control (MAC) sublayer to leverage OFDMA advantages. We identify necessary extensions, study the resulting overhead, and introduce a lightweight design for a complete dynamic OFDMA IEEE 802.11a system. Studying its performance shows that with our lightweight integration dynamic OFDMA can improve IEEE 820.11a UDP throughput by up to 154% and UDP latency by up to 63% even if the full overhead is taken into account.

Cooperative relaying has been shown to provide diversity gains which can significantly improve the packet error rate (PER) in wireless transmissions. In ad hoc wireless routing where packets may travel over a number of hops before reaching the destination, hop-wise cooperative relaying may severely reduce network capacity. This approach was mainly addressed in literature so far. In this paper, we efficiently apply cooperative relaying along a complete path and over multiple hops at the same time. We use information from the routing layer to improve the medium access control (MAC) layer's performance. Simulations and testbed implementation show appealing gains through diversity resulting in up to 66% better PER performance and up to 148% goodput increase compared to conventional approaches.

Opportunistic Relaying (OR) and Selection Decode-and-Forward (SDF) cooperation protocols can both substantially improve the performance of wireless networks but fundamentally differ in utilized redundancy, relays, and channel knowledge. To analyze when SDF or OR improves error and data rate, we (1) derive their general outage probability and capacity for arbitrary relay configurations, (2) systematically benchmark both approaches in two-hop configurations, (3) study how often beneficial configurations occur in large networks, and, finally, condition our capacity results by this occurrence probability. Our results clearly show that OR maximizes the outage capacity at high acceptable error rate while SDF succeeds if a low error rate is required. SDF performs best if even the relays can cooperate among themselves, supported frequently in networks with more than three neighbors. Consequently, cooperating relays, adapting between OR and SDF, and joining these two approaches should be the focus of future protocol design. To this end, our paper provides a theoretical basis, adaptation rules, and design guidelines.

We describe and analyze minimum path difference (MPD), a metric to improve threshold-based selection decode-and-forward (SDF) protocols in cooperative wireless networks with channel coding. By observing the decoding process, MPD provides a more frequent channel quality estimation than CRC or SNR and, thus, allows SDF to forward only the parts of a message that are likely to be correct. In this paper, we describe MPDpsilas efficient integration into Viterbi decoders and SDF protocols, analyze its accuracy and threshold selection, and study its end-to-end Bit Error Rate in practical cooperative WLANs where ideal block fading and ideal channel codes cannot be assumed. Our studies validate that, unlike all prior methods, in these practical scenarios, MPD significantly improves SDFpsilas performance and experiences only a marginal performance penalty if suboptimal, constant thresholds are selected.

We propose a practical medium access control (MAC) protocol and transceiver design for mobile cooperative WLANs. Our MAC protocol integrates selection decode-and-forward (SDF) cooperative relaying into IEEE 802.11. Unlike previous approaches, its cooperative signaling cycle allows communication if the direct link fails even for small signaling frames. Further SDF functions are efficiently supported by our ready- to-use transceiver design. We implement this design, including our MAC protocol, on a software defined radio (SDR) resulting in a cooperative IEEE 802.11g prototype. Using this prototype, we validate feasibility and performance of our approaches by extensive field measurements in indoor and railroad scenarios with an actual train moving the cooperating terminals.

For block-fading Rayleigh channels, this work considers a network coding approach to cooperative transmission. The network under consideration consists of two sources, S and P, that transmit to a common destination, D. Using half-duplex channels, S and P transmit their packets to D; they also overhear each other's transmission. Then both sources cooperate by transmitting a network-coded packet, which is used at D as incrementally redundant information. We first analyze outage probability and show that this scheme achieves full diversity when good quality inter-source channels exist; the result is also verified through simulation. Then, using the outage result we investigate the 'optimal' rate and energy allocations that minimize the outage probability. The results show that the outage-based performance is more sensitive to the energy allocation than to the rate allocation.

In wireless networks, cooperative transmission is used as a means to combat channel fading. In this system, a source and a relay transmit each others' messages to a common destination using either amplify-and-forward or decode-and-forward strategies; the protocols in cooperative transmission can also broadly be categorized as static and adaptive. The idea in a network-coding-based cooperative transmission protocol is to allow the source and relay to combine messages by a network coding operation; a modulo-2 sum operation implements this network coding. In this work, the performance of various decode-and-forward-based cooperative transmission protocols without and with additional network coding is investigated. Outage probability is used as a measure of performance, and results for symmetrical source-relay and source/relay-destination channels are presented; also a source-relay-channels-based comparison is made. Based on these outage results, network-coding-based protocols are found to be suitable when the source-relay channels are unreliable; when the source-relay channels are good, protocols without network coding perform better. Moreover, to improve the performance of static protocols, we have introduced a sequence of decoding at the destination, and the corresponding outage results show that these protocols can achieve full diversity.

In future wireless networks, cooperating users may create virtual antenna arrays to profit from user cooperation diversity. Even then, fading can severely reduce the quality of real-time media streams. To improve streaming quality in cooperative networks, we propose, firstly, the asymmetric allocation of diversity branches. Secondly, we employ this method in our traffic-aware asymmetric cooperation diversity (TACD) scheme to prioritize important parts of the media streams. We demonstrate by outage analysis and simulation that this approach significantly increases the quality of real-time media streams without introducing additional delays.

In cooperative diversity systems, single antenna nodes may share their antennas to achieve a performance comparable to multi-antenna systems. Cooperative diversity is efficiently provided by the coded cooperation algorithm (Hunter and Nosratinia, 2002) where users cooperate by mutually transmitting their FEC-coded data. This paper studies the effect of user mobility on the outage probability performance of coded cooperation with two users. For this study the authors assumed that users do not always move and that for moving users, fading channel characteristics depend on the motion speed. Considering these two mobility factors, scenarios were defined for which the outage probabilities were analytically derived. The effects of user speed and the level of cooperation are further investigated by simulation. Finally a simple method for the dynamic selection of partners required for successful cooperation by adapting the cooperation level was proposed and discussed the effect of mobility and transmission power on the performance of this method.

Abstract To solve the IP mobility problem, the use of multicast has been proposed in a number of different approaches, applying multicast in different characteristic ways. We provide a systematic discussion of fundamental options for multicast‐based mobility support and the definition and experimental performance evaluation of selected schemes. The discussion is based on an analysis of the architectural, performance‐related, and functional requirements. By using these requirements and selecting options regarding network architecture and multicast protocols, we identify promising combinations and derive four case studies for multicast‐based mobility in IP‐based cellular networks. These case studies include both the standard any‐source IP multicast model as well as non‐standard multicast models, which optimally utilize the underlying multicast. We describe network architecture and protocols as well as a flexible software environment that allows to easily implement these and other classes of mobility‐supporting multicast protocols. Multicast schemes enable a high degree of flexibility for mobility mechanisms in order to meet the service quality required by the applications with minimal protocol overhead. We evaluate this overhead using our software environment by implementing prototypes and quantifying handoff‐specific metrics, namely, handoff and paging latency, packet loss and duplication rates, as well as TCP goodput. The measurement results show that these multicast‐based schemes improve handoff performance for high mobility in comparison to the reference cases: basic and hierarchical Mobile IP. Comparing the multicast‐schemes among each other the performance for the evaluated metrics is very similar. As a result of the conceptual framework classification and our performance evaluations, we justify specific protocol mechanisms that utilize specific features of the multicast. Based on this justification, we advocate the usage of a source‐specific multicast service model for multicast‐based mobility support that adverts the weaknesses of the classical Internet any‐source multicast service model. Copyright © 2006 John Wiley & Sons, Ltd.

This paper presents an end-to-end reservation protocol for quality-of-service (QoS) support in the medium access control layer of wireless multihop mesh networks. It reserves periodically repeating time slots for QoS-demanding applications, while retaining the distributed coordination function (DCF) for best effort applications. The key features of the new protocol, called "distributed end-to-end allocation of time slots for real-time traffic (DARE), are distributed setup, interference protection, and scheduling of real-time data packets, as well as the repair of broken reservations and the release of unused reservations. A simulation-based performance study compares the delay and throughput of DARE with those of DCF and the priority-based enhanced distributed channel access (EDCA) used in IEEE 802.11e. In contrast to DCF and EDCA, DARE has a low, nonvarying delay and a constant throughput for each reserved flow

Within the last decade, the orthogonal frequency- division multiplexing (OFDM) transmission scheme has become part of several standards for wireless systems. Today, OFDM is even a candidate for fourth-generation wireless systems. It is well known that dynamic OFDMA systems potentially increase the spectral efficiency. They exploit diversity effects in time, space, and frequency by assigning system resources periodically to terminals. Informing the terminals about new assignments creates a signaling overhead. Up to now, this overhead has not been taken into account in studies on dynamic orthogonal frequency-division multiplexing access (OFDMA) systems. Yet this is crucial for a realistic notion of the performance achieved by dynamic approaches. In this paper, we close this gap. We introduce two forms of representing the signaling information and discuss how these affect system performance. The study of the signaling impact on the performance is conducted for an exemplary dynamic approach. We find that the throughput behavior of dynamic OFDMA systems is significantly influenced by the signaling overhead. In many situations, neglecting the overhead leads to wrong performance conclusions. Also, the performance difference between dynamic and static schemes is now much more sensible to the specific parameter set of the transmission scenario (e.g., frame length, subcarrier number, etc.). This leads to the proposal of access points which should adapt certain system parameters in order to provide optimal performance.

Widespread concerns about potential health hazards caused by electromagnetic exposure have been raised in many countries. Although there is no ultimate consensus on radiation power levels and their respective impacts on human health, methods are necessary to reduce exposure as much as possible because of the rapid deployment of wireless communications systems. In our work we scrutinize the exposure reduction potential of the multi-hopping approach for two standardized but different wireless network types: IEEE 802.11b as a representative example for networks with distributed channel access control and HIPERLAN/2 as a representative example for centrally controlled networks. Our results show that multi-hopping has the potential to reduce the received electro magnetic power and energy for both network types although it suffers from a decreased network capacity. When attempting to compensate for this decrease by adapting the data rate used over an individual link, the ensuing transmission power adaptation required to maintain acceptable error rates does not necessarily neutralize the benefit of multi-hopping. We show that the sophisticated use of multi-hopping in conjunction with data rate adaption positively affects the received electromagnetic power and energy.

Emerging "urban" ad hoc networks resulting from a large number of individual WLAN users challenge the way users could explore and interact with their physical surroundings. Robust and efficient service discovery and routing protocols in such networks are a necessary ingredient. Although a lightweight service discovery proposal integrated with ad hoc routing exists, an implementation and performance evaluation with respect to overhead and correctness have so far been missing.Moreover, the different service providers in an urban scenario, which (more or less) frequently and actively change their status, demand more flexible handling of cached information on neighboring providers than what is currently proposed. We therefore contribute mechanisms that maintain cache consistency and show that explicitly removing cache entries on existing neighboring providers is well invested effort. We finally evaluate whether ad hoc network latency can implicitly help our protocol in retrieving the physically closest provider, using an 802.11 model. Additionally, we provide a general architectural framework for enabling lightweight service discovery on top of most reactive routing protocols.

When optimizing the operations of large wireless ad hoc networks, neither global nor local information-based approaches fits well: they require either information about the entire network structure, which is in most cases impossible to get, or are not capable of optimizing beyond a very narrow horizon. We propose a novel optimization scheme based on regional information to compute the network-wide optimizations, taking the peculiarities of large ad hoc networks into account, and obtain an "emergent algorithm" out of a global optimization algorithm. Our solution uses a clustering algorithm to define regions but needs neither cluster maintenance nor inter-cluster communication protocols, thus is expected to be very robust. The problem of distributed frequency assignment is used as a case study to demonstrate the performance of our method as compared to algorithms based on local- or global information.

In multihop ad hoc networks the IEEE 802.11 MAC protocol, in its distributed version, dramatically degrades the performance in terms of throughput and delay. This is due to a protocol property, which tries to provide equal probability of channel access to all nodes. But in multihop ad hoc network a more frequent channel access might be necessary at certain nodes that are intermediate hops and forward other nodes' data. This paper proposes a modification to the protocol that leads to a higher throughput and lower delay by combining the ACK for one packet with the channel access procedure for another, queued-to-be-sent packet. Hence, nodes receiving many packets, probably forwarder, have a higher chance of accessing the channel. Additionally, these changes provide a scheme of packet forwarding more robust to different layouts than original IEEE 802.11, where performance can heavily deteriorate in certain configurations.

The error patterns of a wireless digital communication channel can be described by looking at consecutively correct or erroneous bits (runs and bursts) and at the distribution function of these run and burst lengths. A number of stochastic models exist that can be used to describe these distributions for wireless channels, e.g., the Gilbert-Elliot model. When attempting to apply these models to actually measured error sequences, they fail: measured data gives raise to two essentially different types of error patterns which can not be described using simple error models like Gilbert-Elliot. These two types are distinguished by their run length distribution; one type in particular is characterized by a heavy-tailed run length distribution. This paper shows how the chaotic map model can be used to describe these error types and how to parameterize this model on the basis of measurement data. We show that the chaotic map model is a superior stochastic bit error model for such channels by comparing it with both simple and complex error models. Chaotic maps achieve a modeling accuracy that is far superior to that of simple models and competitive with that of much more complex models, despite needing only six parameters. Furthermore, these parameters have a clear intuitive meaning and are amenable to direct manipulation. In addition, we show how the second type of channels can be well described by a semiMarkov model using a quantized lognormal state holding time distribution.

Mobile communication is increasingly oriented towards the usage of all IP networks as fixed-network components. An open problem is how to provide QoS guarantees that are competitive with that of existing cellular networks. In particular, an appropriate handover support is missing: a handover should not be performed to a base station that is not able to support the desired QoS. The paper presents a performance evaluation of a mechanism that integrates QoS support and handover mechanisms that in IP networks such that a handover is conditionalized upon availability of sufficient QoS resources. We show that this scheme is able to efficiently provide such conditional handover support, that is competitive with standard hierarchical mobile IPv6 regarding the amount of carries traffic, and that it outperforms hierarchical mobile IPv6 in arranging traffic within the network, resulting in a superior efficiency especially for mixed QoS/non-QoS traffic.

In a wireless, multi-hop sensor network, choosing transmission power levels has an important impact on energy efficiency and network lifetime. Two algorithms for dynamically adjusting transmission power level on a per-node basis are proposed here. Network lifetime, convergence speed as well as resulting network connectivity are used as figures of merit for these two algorithms. They have been evaluated in an indoor sensor environment. The network lifetime metrics of these two local algorithms are also benchmarked against power control algorithms using global information. We show that these local algorithms outperform fixed power level assignment and that the lifetime achieved by them is usually within a factor of two of globally computed solution while being scalable.

Java, in combination with Web browsers' abilities to load and execute untrusted Java applets in a secure fashion, has made computing over the Web a possibility. Now the challenge is to fully utilize this potential, given the limitations imposed by browsers. This paper presents KnittingFactory, an infrastructure to facilitate Web-based computing, which addresses this challenge. It supports building distributed applications, specifically those consisting of Java applets executing in browsers. It is composed of: (i) a distributed name service to assist users in locating other participants of a distributed computation via standard browsers; (ii) an embedded class server to eliminate the need for external HTTP servers for serving applet code; and (iii) a technique for direct applet-to-applet communication. In this paper, we describe the design and implementation of KnittingFactory and demonstrate its benefits by applying it to three distinct areas of Web-based computing. First, we apply our distributed name service to a client/server architecture to enable RMI clients to locate servers on unknown hosts. Second, we use the embedded class server to extend the capability of Charlotte, a parallel computing environment. Finally, we build a collaborative application using our direct applet-to-applet communication technique which does not require a forwarding agent. © 1998 John Wiley & Sons, Ltd.

Using Java for high-performance distributed computing aggravates a well-known problem: the choice between efficient message-passing environments and more convenient distributed shared memory systems which often provide additional functionalities like adaptive parallelism or fault tolerance – with the latter being imperative for Web-based computing. This paper proposes an extension to the DSM-based Charlotte system that incorporates advantages from both approaches. Annotations are used to describe the data dependencies of parallel routines. With this information, the runtime system can improve the communication efficiency while still guaranteeing the correctness of the shared memory semantics. If the correctness of these annotations can be relied upon, additional optimizations are possible, ultimately sharing primitive data types such as int across a network, making the overhead associated with accessing and sharing objects unnecessary. In this case, the annotations can be regarded as a compact representation of message-passing semantics. Thus, a program's efficiency can be improved by step-by-step incorporation of semantic knowledge. The possibility to freely mix and to easily switch between unannotated code, annotated code and shared primitive data types entails a big flexibility for the programmer. A number of measurements show significant performance improvements for annotations and annotation-based shared primitive types. © 1998 John Wiley & Sons, Ltd.