Classical particle tracking using frame-based Active Pixel Sensor (APS) struggles with poor lighting conditions, scenes with high dynamic range, dynamic changes in illuminance and restricted frame rate. We extend Particle Tracking Velocimetry (PTV) to such difficult environmental conditions by utilizing Event-Based Vision (EBV) sensors, which asynchronously generate timestamped events in response to pixel intensity changes.

Point cloud compression is essential to experiencing remote volumetric multimedia as it drastically reduces the required streaming data rates. Point attributes, specifically colors, extend the challenge of lossy compression beyond geometric representation to the joint reconstruction of texture and geometry. Current state-of-the-art methods separate geometry and attributes to compress them individually, requiring distinct compression models for each modality and resulting in a complex, multi-step compression pipeline that increases the overall coding latency. In contrast, using a single model to jointly compress geometry and attributes allows to highly simplify the encoding and decoding process, but requires to select the trade-off between the rate and the quality of both modalities at training time. Consequently, during inference, each model operates at the geometry-attribute-rate trade-off defined during training, making it necessary to train and deploy an ensemble of models for scenarios that demand quality-rate adaptivity. We propose a method to overcome this shortcoming by conditioning a model on the space of possible geometry and attribute trade-offs during training. This results in a single model that can cover a wide range of trade-offs between geometry quality, attribute quality and rate during inference. We find competitive rate-distortion performance on densely sampled point clouds compared to state-of-the-art standards and to existing learned compression approaches while being less effective on sparse geometry point clouds. By providing a single model that can adapt to a wide range of geometry-attribute-rate trade-offs at inference, our approach reduces coding latency and system complexity, advancing learned point cloud compression towards deployment in immersive multimedia systems.

Point clouds are a promising video representation for virtual and augmented reality. Their high-bitrate, however, has so far limited the practicality of live streaming systems. In this work, we leverage the delivery timeout feature within the Media Over QUIC protocol to perform implicit server-side adaptation based on an application's latency target. Through experimentation with several publisher and network configurations, we demonstrate that our system unlocks a unique trade-off on a per-client basis: applications with lower latency requirements will receive lower-quality video, while applications with more relaxed latency requirements will receive higher-quality video.

We optimize hierarchies of Time-to-Live (TTL) caches under network delays. A TTL cache assigns individual eviction timers to cached objects that are usually refreshed upon a hit where upon a miss the object requires a random time to be fetched from a parent cache. Due to their object decoupling property, TTL caches are of particular interest since the optimization of a per-object utility enables service differentiation. However, state-of-the-art exact TTL cache utility-based optimization does not extend beyond single TTL caches, especially under network delays. In this paper, we leverage the object decoupling effect to formulate the nonlinear utility maximization problem for TTL cache hierarchies in terms of the exact object hit probability under random network delays. We iteratively solve the utility maximization problem to find the optimal per-object TTLs. In addition, we propose a variant TTL policy, which we denote as exTTL to counteract the effect on the optimal utility of the storage mismatch between the actual realization of a TTL cache and its ideal infinite storage assumption. Further, we show that the exact model suffers from tractability issues for large hierarchies and propose a machine learning approach to estimate the optimal TTL values for large systems. Finally, we provide numerical and data center trace-based evaluations for both methods, showing the significant offloading improvement due to TTL optimization considering the network delays.

Fifth Generation (5G) networks and beyond are envisioned to provide user-focused communications, supporting diverse services with enhanced Quality of Service (QoS). Pivotal to this evolution is user equipment, which is increasingly performing advanced computational tasks beyond the edge of the network, known as the Extreme-Edge. Seamless integration of Extreme-Edge devices (EEDs) into the 5 G framework is however hindered, due to challenges in terms of device management, resource restrictions and interoperability issues. To address these barriers, we realize the Extreme-Edge Orchestrator (EEO), a management and orchestration framework enabling the extension of the 5 G cloud-to-edge continuum towards the Extreme-Edge. The EEO enables real-time resource monitoring and lifecycle management of network applications, including Artificial Intelligence/Machine Learning (AI/ML) tasks, deployed on EEDs. Unlike existing theoretical studies, our solution is deployed on an operational research-center-wide 5 G testbed and evaluated using an AI/ML-based network QoS prediction application, in the automotive domain. Our results show that the EEO supports efficient resource utilization, dynamic EED selection under device mobility scenarios and maintains robust service performance under computational stress-demonstrating its capability to support next-generation network services.

Given a quantum communication system that offers a service to a distributed quantum application, we consider the problem of optimizing purification strategies for link-level entanglements with respect to the offered teleportation fidelity. We study the trade-off between the entanglement generation and purification process, which may probabilistically fail, and the fidelity decay of application qubits due to waiting in memory for ready entanglements. Given application requests for entanglements that arrive at random time points, we devise a purification strategy that takes into account the current system state and a probabilistic notion of time-dependent risk to gauge the impact on the average teleportation fidelity. Numerical results show that the optimized purification strategy performs better when compared to always pumping across a number of different regimes and that always pumping is only optimal for certain intermediate decoherence time values.

Learned point cloud compression methods have achieved rate-distortion performance, which is comparable to or higher than conventional approaches. However, this often comes at the cost of high hardware requirements and thus low throughput during encoding and decoding. In this paper, we present an adaptive bitrate point cloud streaming system utilizing learned compression. While other learned compression techniques require to split geometry and attributes, resulting in high encoding latency, we deploy a unified model to handle both modalities together, which drastically reduces the coding complexity. We explore the capabilities of the learned encoder to derive multiple quality representations with only re-running a fraction of the encoding steps, making it a suitable fit for adaptive bitrate streaming. Furthermore, we ablate the encoding latency of each component in the encoder and decoder stack, identifying bottlenecks in the process.

Dynamic Point Clouds are a representation for three-dimensional (3D) immersive media that allows users to freely navigate a scene while consuming the content. However, this comes at the cost of substantial data size, requiring efficient compression techniques to make point cloud videos accessible. Addressing this, Video-based Point Cloud Compression (V-PCC) projects points into 2D patches to compress video frames, leveraging the high compression efficiency of legacy video codecs and exploiting temporal correlations in the two-dimensional (2D) images. However, clustering and projecting points into meaningful 2D patches is computationally intensive, leading to high encoding latency in V-PCC. Applying adaptive streaming techniques, originating from traditional video streaming, multiplies the computational effort as multiple encodings of the same content are required. In this light, transcoding a compressed representation into lower qualities for dynamic adaptation to user requirements is gaining popularity. To address the high latency when employing the full decoder-encoder stack of V-PCC during transcoding, we propose RABBIT, a novel technique that only re-encodes the underlying video sub-streams. This is in contrast to slow V-PCC transcoding that reconstructs and re-encodes the raw point cloud at a new quality setting. By eliminating expensive overhead resulting from calculations based on the 3D space representation, the latency of RABBIT is bounded by the latency of transcoding the underlying video streams, allowing optimized video codec implementations to be used to meet the real-time requirements of adaptive streaming systems. Our evaluations of RABBIT, using various optimized video codec implementations, shows on-par quality with the baseline V-PCC transcoding given a high-quality representation. Given unicast or multicast distribution of a point cloud stream and in-network or edge transcoders, our evaluations show the tradeoff between rate-distortion performance and the required network bandwidth.

Video streaming has evolved from push-based, broad-/multicasting approaches with dedicated hard-/software infrastructures to pull-based unicast schemes utilizing existing Web-based infrastructure to allow for better scalability. In this article, we provide an overview of the foundational principles of HTTP Adaptive Streaming (HAS), from video encoding to end user consumption, while focusing on the key advancements in adaptive bitrate algorithms, Quality of Experience (QoE), and energy efficiency. Furthermore, the article highlights the ongoing challenges of optimizing network infrastructure, minimizing latency, and managing the environmental impact of video streaming. Finally, future directions for HAS, including immersive media streaming and neural network-based video codecs, are discussed, positioning HAS at the forefront of next-generation video delivery technologies.

The first step for entanglement distribution among quantum communication nodes is to generate link-level Ein-stein-Podolsky-Rosen (EPR) pairs between adjacent communication nodes. EPR pairs may be continuously generated and stored in a few quantum memories to be ready for utilization by quantum applications. A major challenge is that qubits suffer from unavoidable noise due to their interaction with the environment, which is called decoherence. This decoherence results in the known exponential decay model of the fidelity of the qubits with time, thus, limiting the lifetime of a qubit in a quantum memory and the performance of quantum applications. In this paper, we evaluate the fidelity of the stored EPR pairs under two opposite dynamical and probabilistic phenomena, first, the aforementioned decoherence and second purification, i.e. an operation to improve the fidelity of an EPR pair at the expense of sacrificing another EPR pair. Instead of applying the purification as soon as two EPR pairs are generated, we introduce a Purification scheme Beyond the Generation time (PBG) of two EPR pairs. We use discrete time Markov chain (DTMC) approach to analytically show the probability distribution of the fidelity of stored link-level EPR pairs in a system with two quantum memories at each node allowing a maximum of two stored EPR pairs. In addition, we apply a PBG scheme that purifies the two stored EPR pairs upon the generation of an additional one. We finally provide numerical evaluations of the analytical approach and show the fidelity-rate trade-off of the considered purification scheme.

We present a robust method to analyze a broad range of classical queueing models, e.g., the GI/G/1 queue with renewal arrivals, an AR/G/1 queue with alternating renewals (AR), as a special class of Semi-Markovian processes, and Markovian fluids queues. At the core of the method lies a standard change-of-measure argument to reverse the sign of the negative drift in the underlying random walks. Combined with a suitable representation of the overshoot, we obtain exact results in terms of series. Closed-form and computationally fast bounds follow by taking the series’ first terms, which are the dominant ones because of the positive drift under the new probability measure. The obtained bounds generalize the state-of-the-art class of martingale bounds and can be much sharper by orders of magnitude.

Adaptive streaming of 360° tiled video requires encoding tiles to multiple qualities to support client decisions in the light of fluctuating bandwidth and dynamic view-ports. Some static approaches allocate fixed encoding resources independently of the scene complexity introducing a significant resource overhead while other static approaches add a substantial time delay relative to the source. As tiles of 360 ° video have a skewed popularity with a statistical concentration of the user requests static approaches are resource inefficient. Specifically, in live streaming the tile popularity and encoding time statistics are not present beforehand.

Time-Sensitive Networking (TSN) enables deterministic and low-latency communication for real-time applications over Ethernet. That is accomplished by leveraging scheduling and shaping techniques configured for each egress port within the network switches. Although Time Aware Shaper (TAS) is a promising solution for TSN, its adoption often involves substantial complexity. In this work, we propose Residence Delay Aggregation (RDA), a novel asynchronous TSN mechanism that offers dynamic traffic scheduling adapted to the traffic load. Specifically, the proposed RDA mechanism provides upper bound delays similar to other asynchronous TSN mechanisms while improving the flexibility of traffic scheduling and reducing the deployment complexity.

QoS forecasting for cellular vehicular communications allows cooperative, connected and automated mobility applications to tailor their behavior to the expected communication conditions on the road. In a nutshell, vehicles may, for example, execute cooperative maneuvers if the communication quality of service is only above a certain quantitative level whereas if not they revert to the individual autonomous mode. In this paper, we propose and show empirical methods for estimating packet-based QoS metrics obtained from 5G network measurements with a direct application to vehicular applications. As many distributed vehicular applications possess strict QoS requirements, we focus here on bounding packet-based statistical QoS quantiles, specifically for latency and loss. Our approach is based on training regression neural networks in a federated learning fashion and show that it can obtain predictions on par with centralized training without the vehicles needing to transmit raw measurement data. In contrast to QoS prediction using physical layer information, we briefly discuss the embedding of such much simpler application-level service within the 5G architecture. We also validate our approach through recovering classical closed-form delay quantiles that are obtained from analytical models of simple queueing systems. We show that our approach goes beyond these simple models in that it provides quantile estimates for the complex scenario of cellular vehicle communications and under different application traffic patterns including empirical data traffic traces as well as 5G testbed measurements.

Major building blocks of communication networks such as flow control and congestion control rely on fresh estimates of the network state to control data traffic injection into the network. These measured metrics are usually implicitly considered as estimates of the future network state until updated. In this paper, we propose to directly and explicitly estimate packet-based predictive QoS metrics from network measurements. As many applications possess strict QoS requirements, we focus here on bounding packet delay quantiles. Our approach is based on training neural networks to predict the quantile of the delay distribution observed by future packets given some observations of packet delays. We validate our approach through recovering classical closed-form delay quantiles that are obtained from analytical models of simple queueing systems. We show that our approach goes beyond these simple models in that it provides quantile estimates for complex scenarios and under various traffic patterns including empirical data traffic traces.

A significant portion of Machine Learning (ML) studies are toward automation of the training process for finding a robust ML model, while requiring the least amount of human effort. Although the introduced Automated ML (AutoML) tools show promising results with ongoing efforts for faster search, the primary beneficiaries are data scientists. Meanwhile, one of the most valuable data scientist roles is their experience in converting unclear, and noisy empirical problems into ML problems. In other words, the tool's performance is still relative to data scientists' consultation in crafting (business) requirements into ML pipelines. Such an entanglement to experts can avoid full automation of ML use for domain specialists with no background in programming and algorithms. This makes domain users, such as business owners, still dependent on the support of ML practitioners in every new use case. In this paper, we conduct a study to evaluate the mediator feature in which multi-modality simulates interchanging information from business users to data scientists in actual exchange meetings. We consider an industrial scenario, where noisy quality inspection business objectives and data scientists' past actions are used as a foundation for choosing follow-up ML tasks for future applications under known business requirements.

Load balancers in distributed caching systems face a fundamental trade-off between networking and caching performance metrics. The first comprises how the network traffic of object requests and replies is balanced on the different links to/from the cache servers while the second denotes cache hit rates and response times. In a nutshell, the root of this trade-off lies in the combination of the skewed popularity and dynamic nature of incoming object requests as well as whether the load balancing function is agnostic to the caching application or not. In this paper, we present Lethe, a network data-plane load balancer for distributed Time-to-Live (TTL) caching. We con-sider TTL caching as it separates the object dynamics but still performs equivalently to many classical caching algorithms (e.g. LRU) under appropriate TTL parametrization. Lethe segregates the cache objects based on the pattern of the incoming requests and efficiently places objects in cache servers to balance the network traffic without sacrificing the caching system hit rate. We implement Lethe in P4 and experimentally show that it improves the average response time and cache hit rate as compared to application-agnostic load balancing even for a skewed and dynamically changing workload.

Transport layer congestion control relies on feedback signals that travel from the congested link to the receiver and back to the sender. This forward congestion control loop, first, requires at least one Round-Trip Time (RTT) to react to congestion and secondly, it depends on the downstream path after the bottleneck. The former property leads to a reaction time in the order of RTT+ bottleneck queue delay, while the second may amplify the unfairness due to heterogeneous RTT. In this paper, we present Reverse Path Congestion Marking (RPM) to accelerate the reaction to network congestion events without changing the end-host stack. RPM decouples the congestion signal from the downstream path after the bottleneck while maintaining the stability of the congestion control loop. We show that RPM improves throughput fairness for RTT-heterogeneous TCP flows as well as the flow completion time, especially for small Data Center TCP (DCTCP) flows. Finally, we show RPM evaluation results in a testbed built around P4 programmable ASIC switches.

Point clouds are a mature representation format for volumetric objects in 6 degrees-of-freedom multimedia streaming. To handle the massive size of point cloud data for visually satisfying immersive media, MPEG standardized Video-based Point Cloud Compression (V-PCC), leveraging existing video codecs to achieve high compression ratios. A major challenge of V-PCC is the high encoding latency, which results in fallback solutions that exchange the compression ratio for faster point cloud codecs. This encoding effort rises significantly in adaptive streaming systems, where heterogeneous user requirements translate into a set of quality representations of the media.

We investigate the application of a recently published metric for flexibility in the context of combined port queue schedules of network paths in Time-Sensitive Networks (TSN). TSN comprises a set of specifications for deterministic networking, including support for scheduled traffic with guaranteed deterministic end-to-end delays. Typically, scheduler resource allocation in TSN disregards flexibility of scheduler configurations. Essentially, the notion of flexibility of paths comprising multiple concatenated ports having each a TSN configuration is based on the number of possible embeddings, i.e., resource allocations, for a new flow of a given specification (size and delay deadline) along that path. This demonstration allows the user to define TSN schedules along network paths and, hence, illustrates the behavior and benefit of performing flexibility-aware TSN configuration.

Load balancing arises as a fundamental problem, underlying the dimensioning and operation of many computing and communication systems, such as job routing in data center clusters, multipath communication, Big Data and queueing systems. In essence, the decision-making agent maps each arriving job to one of the possibly heterogeneous servers while aiming at an optimization goal such as load balancing, low average delay or low loss rate. One main difficulty in finding optimal load balancing policies here is that the agent only partially observes the impact of its decisions, e.g., through the delayed acknowledgements of the served jobs. In this paper, we provide a partially observable (PO) model that captures the load balancing decisions in parallel buffered systems under limited information of delayed acknowledgements. We present a simulation model for this PO system to find a load balancing policy in real-time using a scalable Monte Carlo tree search algorithm. We numerically show that the resulting policy outperforms other limited information load balancing strategies such as variants of Join-the-Most-Observations and has comparable performance to full information strategies like: Join-the-Shortest-Queue, Join-the-Shortest-Queue(d) and Shortest-Expected-Delay. Finally, we show that our approach can optimise the real-time parallel processing by using network data provided by Kaggle.

A key metric to express the timeliness of status updates in latency-sensitive networked systems is the age of information (AoI), i.e., the time elapsed since the generation of the last received informative status message. This metric allows studying a number of applications including updates of sensory and control information in cyber-physical systems and vehicular networks as well as, job and resource allocation in cloud clusters. State-of-the-art approaches to analyzing the AoI rely on queueing models that are composed of one or many queuing systems endowed with service order, e.g., FIFO, LIFO, or last-generated-first-out order. A major difficulty arising in these analysis methods is capturing the AoI under message reordering when the delivery is non-preemptive and non-FIFO, i.e., when messages can overtake each other and the reception of informative messages may obsolete some messages that are underway. In this paper, we derive an exact formulation for the distribution of AoI in non-preemptive, non-FIFO systems where the main ingredients of our analysis are Palm calculus and time inversion. Owing to the rationality of the Laplace-Stieltjes transforms that are used in our approach, we obtain computable exact expressions for the distribution of AoI.

We consider Time-to-Live (TTL) caches that tag every object in cache with a specific (and possibly renewable) expiration time. State-of-the-art models for TTL caches assume zero object fetch delay, i.e., the time required to fetch a requested object that is not in cache from a different cache or the origin server. Particularly, in cache hierarchies this delay has a significant impact on performance metrics such as the object hit probability. Recent work suggests that the impact of the object fetch delay on the cache performance will continue to increase due to the scaling mismatch between shrinking inter-request times (due to higher data center link rates) in contrast to processing and memory access times. In this paper, we analyze tree-based cache hierarchies with random object fetch delays and provide an exact analysis of the corresponding object hit probability. Our analysis allows understanding the impact of random delays and TTLs on cache metrics for a wide class of request stream models characterized through Markov arrival processes. This is expressed through a metric that we denote delay impairment of the hit probability. In addition, we analyze and extend state-of-the-art approximations of the hit probability to take the delay into account. We provide numerical and trace-based simulation-based evaluation results showing that larger TTLs do not efficiently compensate the detrimental effect of object fetch delays. Our evaluations also show that unlike our exact model the state-of-the-art approximations do not capture the impact of the object fetch delay well especially for cache hierarchies. Surprisingly, we show for single caches that the impact of the delay on the hit probability can be non-monotonic and that the range of delays for which a positive effect exists arises as a root of a polynomial in the ratio of the expected TTL to the expected inter-request time.

Designing Computer Vision (CV) algorithms for production applications requires extensive knowledge of Machine Learning (ML), programming, and software architecture. Accordingly, keeping humans, especially domain users, in the loop of model decisions is a challenge. The algorithms created by the CV community usually omit to provide clear interpretable information on the finally selected model, except for numerical metrics. This leads to a transparency challenge, known as black-box ML, which may negatively impact the users’ trust in the sensitive use cases. Furthermore, the majority of interpretable methods for giving feedback to domain users are usually made after training, i.e., proposing a final model without the user participation.In this paper, we present a Visual Interpretation-based Control (VIC) technique, which is a simple but principled model evaluation criteria. VIC offers a decision-making strategy for enabling non-experts to dictate their intuition from the most important areas in an image and incorporate this within CV pipelines. Specifically, we supervise a generative adversarial network to penalize its generator for the specified region of interest. We further validate our method through an architecture selection strategy in one of the common AutoML benchmarks.

Fixed buffer sizing in computer networks, especially the Internet, is a compromise between latency and bandwidth. A decision in favor of high bandwidth, implying larger buffers, subordinates the latency as a consequence of constantly filled buffers. This phenomenon is called Bufferbloat. Active Queue Management (AQM) algorithms such as CoDel or PIE, designed for the use on software based hosts, offer a flow agnostic remedy to Bufferbloat by controlling the queue filling and hence the latency through subtle packet drops.In previous work, we have shown that the data plane programming language P4 is powerful enough to implement the CoDel algorithm. While legacy software algorithms can be easily compiled onto almost any processing architecture, this is not generally true for AQM on programmable data plane hardware, i.e., programmable packet processors. In this work, we highlight corresponding challenges, demonstrate how to tackle them, and provide techniques enabling the implementation of such AQM algorithms on different high speed P4-programmable data plane hardware targets. In addition, we provide measurement results created on different P4-programmable data plane targets. The resulting latency measurements reveal the feasibility and the constraints to be considered to perform Active Queue Management within these devices. Finally, we release the source code and instructions to reproduce the results in this paper as open source to the research community.

Sparse and imbalanced data is a common challenge that practitioners must overcome when implementing industrial ML applications. This challenge concerns deep learning-based quality inspection systems in particular, as they often are obligated to adhere to high constraints in terms of reliability and performance. As deep learning quality inspection systems are usually implemented in a supervised manner, they additionally require balanced datasets that may be difficult or costly to obtain in production environments. However, new approaches using Generative Adversarial Networks for synthetic image generation promise a remedy by increasing the data amount of sparse classes, such as faults or defects. This paper presents an experimental use case where we employ a state-of-the-art image generator model of StyleGAN2 to a quality inspection application in laser beam welding to increase the number of defect images for training an object detector. We evaluate the generated images and their influence on the object detector’s performance using several training configurations. Our results reveal that with the limited amount of data, we are able to generate synthetic images that look promising at first glance. However, in the evaluation based on the object detector, we find that introducing synthetic images had an adverse effect on detection performance and robustness of the system. Further research is required to generate defect images from sparse datasets that can improve the performance of object detection systems in quality inspection.

Congestion control mechanisms in computer networks rely mainly on a feedback loop having a reaction time equal to the flow RTT. Reducing this feedback time helps the sender to react faster to changing network conditions such as congestion. In this work, we propose reverse-path congestion notification on top of programmable networking switches. Our approach can significantly lower the reaction time, such that the congestion control implementation can adapt much faster to changing network conditions. The proposed approach aims to work with current TCP implementations with no required changes to the communication endpoints. Last, we show how the presented approach could be realized by utilizing off-the-shelf programmable switches.

End users can mutually share their computing resources in ad-hoc computing environments with code offloading. This augments the computational power of resource-constrained mobile devices and enables interactive user-facing applications that would otherwise exceed single device capabilities. However, ad-hoc computing comes along with new challenges such as heterogeneity and unreliability of devices. Resource consumers have to make task scheduling decisions without relying on a centralized scheduler to facilitate sub-second response times in environments with communication latencies that are in the order of the task execution times. In this paper, we present a decentralized low-latency task scheduling approach that minimizes job execution times in heterogeneous ad-hoc environments. We propose two decentralized task scheduling algorithms that select powerful computing resources for parallel task execution while avoiding delays that arise from congested devices. We provide an analytical model of the performance of these algorithms before conducting an extensive evaluation based on real-world applications and a realistic computing infrastructure. Our results show that decentralized scheduling can dynamically adapt to varying system load and outperform a central scheduler in both task and job execution times, which enables low-latency task offloading in ad-hoc environments.

Time-Sensitive Networking emerged as a convergent Ethernet-based real-time networking standard for industrial applications. To support real-time, jitter-free isochronous traffic the corresponding TSN mechanism denoted Time Aware Shaper requires special hardware support. In this work, we propose a path to building TSN networks on top of programmable switches. Specifically, we show here how to leverage a data structure amenable to programmable data planes known as Push-in First-out (PIFO) queue to support TSN traffic scheduling for isochronous real-time, as well as, best effort traffic.

In Time-Sensitive Networks (TSN) applications with the highest real-time flow requirements are deployed using the Time-Aware Shaper which requires careful planning and scheduling of flows before deployment. Such deployments lack support for dynamic industrial scenarios such as modular machine assembly and reconfiguration, which require a flexible transition between real-time tasks. In contrast, state-of-the-art techniques rely on flow rescheduling and deployment in conjunction with undesired network downtime. Existing works on adapting schedules to traffic admissions are limited in their ability to choose suitable flows to account for future tasks. In this paper, we aim to leverage the flexibility of scheduler configurations to enable TSN dynamic reconfigurability at runtime. We propose a notion of flexibility for TSN Time-Aware Shaper schedules which we utilize to decide the admissibility of consecutive real-time tasks.

Multipath TCP (MPTCP) promises improvements in Quality of Service through connection bundling. This leads to the belief that it will inevitably improve the Quality of Experience (QoE), especially, for mobile applications running on top. The networking and transport layer improvements stem from bundling multiple paths, e.g., WiFi and LTE, as well as increasing the connection reliability through redundancy. For example, a smartphone running an application over WiFi may switch to the cellular network without service interruption upon user movement that gets the device out of the WiFi range, thus avoiding outage events. However, the impact of MPTCP on QoE for different applications has not yet been fully understood.

Many software based traffic load generators suffer from packet rate variation which is known as rate jitter. In this Demo, we show how this varying rate burstiness can affect the device under test even if the generated average data rate seems constant. To this end, we compare a hardware rate shaping, which is implemented using a programmable P4-switch, and a conventional software load generator and show their impact on a software device under test. The results show, that microbursts within the test load significantly impact the experiment results. Our recommendation is to benchmark the traffic load generator before conducting measurement experiments especially when the device under test is sensitive to microbursts.

QoS requirements of current network control and management applications require the ability to conduct precise measurements of network elements, including switches, routers and Virtual Network Functions (VNFs). State-of-the-art network switches have a forwarding delay of 1µs and below and offer high bandwidths of hundreds Gigabits per second. This imposes high time accuracy and loss-detection requirements on measurement equipment that are not met by existing, software-based measurement tools. The use of specialized tools, meeting these requirements, is restricted by limited flexibility and high cost.In this work, we introduce P4STA, an open source frame-work that combines the flexibility of software-based traffic load generation with the accuracy of hardware packet timestamping. Our evaluation results, obtained using an off-the-shelf P4-programmable switch, show that a time resolution up to 1ns can be achieved on these programmable data plane platforms. Moreover we show how to combine the traffic load of multiple software-based load generators to achieve a measurement load of up to 100Gbit/s per port. Experiments on further programmable platforms, specifically on P4-SmartNICs and FPGAs, show similar results. With this work, we make P4STA available for the research community to advance high performance experiment measurements at nanosecond accuracy.

We analyse a data-processing system with n clients producing jobs which are processed in batches by m parallel servers; the system throughput critically depends on the batch size and a corresponding sub-additive speedup function. In practice, throughput optimization relies on numerical searches for the optimal batch size, a process that can take up to multiple days in existing commercial systems. In this paper, we model the system in terms of a closed queueing network; a standard Markovian analysis yields the optimal throughput in ωn4 time. Our main contribution is a mean-field model of the system for the regime where the system size is large. We show that the mean-field model has a unique, globally attractive stationary point which can be found in closed form and which characterizes the asymptotic throughput of the system as a function of the batch size. Using this expression we find the asymptotically optimal throughput in O(1) time. Numerical settings from a large commercial system reveal that this asymptotic optimum is accurate in practical finite regimes.

We study job assignment in large, heterogeneous resource-sharing clusters of servers with finite buffers. This load balancing problem arises naturally in today's communication and big data systems, such as Amazon Web Services, Network Service Function Chains, and Stream Processing. Arriving jobs are dispatched to a server, following a load balancing policy that optimizes a performance criterion such as job completion time. Our contribution is a randomized Cost-Based Scheduling (CBS) policy in which the job assignment is driven by general cost functions of the server queue lengths. Beyond existing schemes, such as the Join the Shortest Queue (JSQ), the power of d or the SQ(d) and the capacity-weighted JSQ, the notion of CBS yields new application-specific policies such as hybrid locally uniform JSQ. As today's data center clusters have thousands of servers, exact analysis of CBS policies is tedious. In this article, we derive a scaling limit when the number of servers grows large, facilitating a comparison of various CBS policies with respect to their transient as well as steady state behavior. A byproduct of our derivations is the relationship between the queue filling proportions and the server buffer sizes, which cannot be obtained from infinite buffer models. Finally, we provide extensive numerical evaluations and discuss several applications including multi-stage systems.

Recent advances in quality adaptation algorithms leave adaptive bitrate (ABR) streaming architectures at a cross-roads: When determining the sustainable video quality one may either rely on the information gathered at the client vantage point or on server and network assistance. The fundamental problem here is to determine how valuable either information is for the adaptation decision. This problem becomes particularly hard in future Internet settings such as Named Data Networking (NDN) where the notion of a network connection does not exist. In this paper, we provide a fresh view on ABR quality adaptation for QoE maximization, which we formalize as a decision problem under uncertainty, and for which we contribute a sparse Bayesian contextual bandit algorithm denoted CBA. This allows taking high-dimensional streaming context information, including client-measured variables and network assistance, to find online the most valuable information for the quality adaptation. Since sparse Bayesian estimation is computationally expensive, we develop a fast new inference scheme to support online video adaptation. We perform an extensive evaluation of our adaptation algorithm in the particularly challenging setting of NDN, where we use an emulation testbed to demonstrate the efficacy of CBA compared to state-of-the-art algorithms.

Recent approaches to tiled 360° adaptive bitrate video streaming present significant bandwidth savings at little risk of stalling when only parts of the video, e.g., the current and predicted viewport, are transferred in high quality while the rest of the 360° video tiles are transferred in a lower quality. While this is currently feasible for video on demand scenarios, it poses a difficult problem for 360° live streaming as naive methods produce a considerable overhead owing to the lack of tiling support in existing hardware encoders.

Virtual reality has been gaining popularity in recent years fueled by the proliferation of affordable consumer-grade devices such as Oculus Rift, HTC Vive, and Samsung VR. Amongst the various VR applications, 360° video streaming is currently one of the most popular ones. However, it poses a series of challenges to the serving content distribution systems. One challenge is the significantly increased bandwidth requirement for streaming such content in real time. Recent research has shown that only streaming the content that is in the user's (field-of-view) FoV in high quality can lead to strong bandwidth savings. This can be achieved by analyzing the viewers head orientation and movement based on sensor information. Alternatively, historic information from users that watched the content in the past can be considered to prefetch 360° video data in high quality assuming the viewer will direct the FoV to these areas. This paper presents a 360° video streaming system that transitions between sensor- and content-based predictive mechanisms. We evaluate the effects of our system on the Quality of Experience (QoE) of such a VR streaming system and show that the perceived quality can be increased between 50% and 80% compared to systems that only apply either one of the two approaches.

Countless novel approaches to communication protocols, overlay networks, and distributed middleware are published every year, yet the adoption of such novel findings in the global Internet landscape progresses at a slow pace. Many of such new communication mechanisms excel (only) under specific deployment conditions, while user mobility and application usage patterns lead to dynamic operation conditions. This mismatch is one reason that makes a wide deployment of new specialized mechanisms particularly hard as observed, for example, for multipath transport protocol extensions until the emergence of multipath transmission control protocol (TCP). This paper formalizes the concept of Transitions, i.e., a method to instrumentalize adaptivity at runtime in communication systems. It allows to exchange communication mechanisms in a running system to optimize the communication quality. In the following, we describe the building blocks required to: 1) capture the features and relations within a communication system and 2) express and optimize the decision making process in such a system. We show how this concept maps intuitively to the Internet model which makes a protocol-independent deployment of applications feasible in the future Internet.

This paper presents MACI, the first bespoke framework for the management, the scalable execution, and the interactive analysis of a large number of network experiments. Driven by the desire to avoid repetitive implementation of just a few scripts for the execution and analysis of experiments, MACI emerged as a generic framework for network experiments that significantly increases efficiency and ensures reproducibility. MACI incorporates and integrates established simulators and analysis tools to foster rapid but systematic network experiments.

QUIC is the emerging transport layer protocol, providing encrypted, stream-multiplexed, low-latency data transfer. In this paper, we propose multipath-enabled QUIC (MPQUIC) to leverage multiple network interfaces, such as WiFi and LTE on today's mobile devices. We show how our MPQUIC design conceptually evolves beyond existing multipathing protocols, such as MPTCP, as it provides fine-grained stream-to-path scheduling, reduced head-of-line blocking, and faster subflow establishment. We present an userland implementation of MPQUIC that is deployable without operating system changes. Our evaluation results show that MPQUIC increases throughput in comparison to traditional QUIC, TCP and even the currently de facto multipath transport protocol MPTCP. First real world measurements confirm that MPQUIC is deployable in the Internet to reduce download times. Moreover, we show that MPQUIC's conceptual advantages over MPTCP efficiently reduce head-of-line blocking in heterogeneous environments. With multipathing support, QUIC is ready to become the universal stream transport protocol in today's Internet.

Collaborative uploading describes a type of crowd-sourcing scenario in networked environments where a device utilizes multiple paths over neighboring devices to upload content to a centralized processing entity such as a cloud service. Intermediate devices may aggregate and preprocess this data stream. Such scenarios arise in the composition and aggregation of information, e.g., from smart phones or sensors. We use a queuing theoretic description of the collaborative uploading scenario, capturing the ability to split data into chunks that are then transmitted over multiple paths, and finally merged at the destination. We analyze replication and allocation strategies that control the mapping of data to paths and provide closed-form expressions that pinpoint the optimal strategy given a description of the paths' service distributions. Finally, we provide an online path-aware adaptation of the allocation strategy that uses statistical inference to sequentially minimize the expected waiting time for the uploaded data. Numerical results show the effectiveness of the adaptive approach compared to the proportional allocation and a variant of the join-the-shortest-queue allocation, especially for bursty path conditions.

Network state monitoring is a fundamental task for network management. However, determining the full network state in Software defined Networks requires disproportionately too many resources. This stems from the discrepancy between the established methods used for state monitoring compared to the varying contribution in terms of information obtained from every additionally monitored network node. This relationship may even become more complicated depending on the network state information of interest. One solution to overcome bottlenecks by reducing the overall monitoring footprint is the use of spatial sampling, which allows the estimation of the network state based a fraction of the overall state. In this work, we propose schemes to place a small number of measurement points in the SDN network to maximize the obtained network state information. Considering different conditions, we utilize routing information and graph theoretic centrality metrics, respectively, to estimate the amount of information a node provides. Based on this knowledge, we, furthermore, develop a mechanism to place multiple measurement points while avoiding redundant measurements. For demonstration purpose, we use the developed mechanisms to estimate the Flow Size Distribution in SDN environments. An emulative evaluation taking several known topologies shows the effectiveness of spatial sampling using the proposed scheme.

The Information-Centric Networking (ICN) paradigm is deemed to enable simpler and more efficient networking interaction by moving from a strict connection based relationship between client and server to an interest-based relationship between user and content. Hence, addressing shifts to content objects rather than any specific copy or location of the content. In order to achieve this new way of addressing the client uses the content name, which is propagated into the network. As a result, a higher efficiency is expected since ICN network nodes (e.g., routers) may reply to such requests using copies from their own caches, or different sources hosting a content object may be used in parallel. Thus, this concept promises better support for device mobility and implicit caching, and it provides inherent multicast support.However, the simplicity of this concept comes at a price. As established applications, such as adaptive bitrate video streaming, have been designed and optimized having a client-server networking environment in mind. In particular, the quality adaptation algorithms used by today's de-facto streaming standard Dynamic Adaptive Streaming over HTTP (DASH) uses bandwidth estimation techniques that are specifically designed for a client-server networking environment where all video segments are retrieved from the same host.This paper addresses issues related to adaptive video streaming in an ICN environment. It analyzes the video streaming behavior of state-of-the-art video quality adaptation algorithms such as PANDA and BOLA in emulated ICN environments. The analysis focuses on the impact ICN chunk-based throughput measurements and ICN caches have on quantitative measurements of Quality-of-Experience (QoE). The paper provides a detailed investigation of the chunk-based throughput estimation showing that it behaves fundamentally different due to ICN caching. Based on these results we provide extensions to existing adaptation algorithms (e.g., PANDA) that can significantly improve the QoE in ICN environments.

Music belongs to one of the most popular content categories overall, and it is nowadays mainly consumed using online streaming services. With YouTube being the largest source of traffic in most networks about half of all YouTube requests address music videos. To cope with the continuously growing demand for content and thus increasing network traffic, YouTube operates its own CDN, a globally distributed network of caches. This allows serving content from locations close to the users, which circumvents potential network bottlenecks and increases the user-perceived QoE due to reduced latency. Recently, proactive caching and prefetching has shown superior performance results compared with traditional reactive caching schemes such as LRU and LFU. Due to the substantial footprint of music videos on today's Internet, we propose a novel proactive caching strategy specifically for music videos. This strategy incorporates two key observations: i) Music genre and mood popularity varies over the course of the day and ii) A video's past views are predictive for its future popularity development. For the classification task, we use a Convolutional Neural Network while investigating several predictive models for the popularity estimation. The proposed caching system can increase the cache hit rate up to 4.5% which is substantial for caching systems.

Content delivery networks (CDNs) carry more than half of the video content in today's Internet. By placing content in caches close to the users, CDNs help increasing the Quality of Experience, e.g., by decreasing the delay until a video playback starts. Existing works on CDN cache performance focus mostly on distinct caching metrics, such as hit rate, given an abstract workload model. Moreover, the nature of the geographical distribution and connection of caches is often oversimplified. In this work, we investigate the performance of cache hierarchies while taking into account the presence of a mixed content workload comprising multiple categories, e.g., news, comedy, and music. We consider the performance of existing caching strategies in terms of cache hit rate and deterioration costs in terms of write operations. Further, we contribute a design and an evaluation of a content category-aware caching strategy, which has the benefit of being sensitive to changing category-specific content popularity. We evaluate our caching strategy, denoted as ACDC (Adaptive Content-Aware Designed Cache), using multiple caching hierarchy models, different cache sizes, and a real world trace covering one week of YouTube requests observed in a large European mobile ISP network. We demonstrate that ACDC increases the cache hit rate for certain hierarchies up to 18.39% and decreases transmission latency up to 12%. Additionally, a decrease in disk write operations up to 55% is observed.

In the past, resource sharing has been extensively studied for OLAP workloads. Naturally, the question arises, why studies mainly focus on OLAP and not on OLTP workloads? At first sight, OLTP queries - due to their short runtime - may not have enough potential for the additional overhead. In addition, OLTP workloads do not only execute read operations but also updates. In this paper, we address query sharing for OLTP workloads. We first analyze the sharing potential in real-world OLTP workloads. Based on those findings, we then present an execution strategy, called OLTPShare that implements a novel batching scheme for OLTP workloads. We analyze the sharing benefits by integrating OLTPShare into a prototype version of the commercial database system SAP HANA. Our results show for different OLTP workloads that OLTPShare enables SAP HANA to provide a significant throughput increase in high-load scenarios compared to the conventional execution strategy without sharing.

A decisive advantage of software-defined networking (SDN) is its support for flexible network reconfigurations. Considering that, software-defined networks require accurate and timely data-plane state information. Network monitoring mechanisms usually require considerable resources on SDN controllers as well as on the data-plane elements. In this paper, we propose an optimization of the statistic transmission to reduce costs on both control- and data-plane regardless of the used monitoring application and statistic provisioning tool. To this end, we intercept the statistic message exchange and 1) aggregate multiple requests coming from different monitoring applications/controllers, 2) filter irrelevant statistic messages with respect to their information gain before delivering them to the control applications, and 3) deploy statistic caching. The proposed system, denoted STATISTIC REQUEST RELAY, forms a logically centralized statistic relay between controllers and the managed data-plane network. Our evaluation shows that the number of statistics processed on controllers as well as statistic requests on switches is reduced significantly while the performance penalty is negligible when using statistic aggregation and filtering as proposed here.

State-of-the-art software-defined wide area networks (SD-WANs) provide the foundation for flexible and highly resilient networking. In this work, we design, implement, and evaluate a novel architecture (denoted as SABR) that leverages the benefits of software-defined networking (SDN) to provide network-assisted adaptive bitrate streaming. With clients retaining full control of their streaming algorithms, we clearly show that by this network assistance, both the clients and the content providers benefit significantly in terms of quality of experience (QoE) and content origin offloading. SABR utilizes information on available bandwidths per link and network cache contents to guide video streaming clients with the goal of improving the viewer’s QoE. In addition, SABR uses SDN capabilities to dynamically program flows to optimize the utilization of content delivery network caches. Backed by our study of SDN-assisted streaming, we discuss the change in the requirements for network-to-player APIs that enables flexible video streaming. We illustrate the difficulty of the problem and the impact of SDN-assisted streaming on QoE metrics using various well-established player algorithms. We evaluate SABR together with state-of-the-art dynamic adaptive streaming over HTTP (DASH) quality adaptation algorithms through a series of experiments performed on a real-world, SDN-enabled testbed network with minimal modifications to an existing DASH client. In addition, we compare the performance of different caching strategies in combination with SABR. Our trace-based measurements show the substantial improvement in cache hit rates and QoE metrics in conjunction with SABR indicating a rich design space for jointly optimized SDN-assisted caching architectures for adaptive bitrate video streaming applications.

YouTube is the most popular platform for streaming of user-generated videos. Nowadays, professional YouTubers are organized in so-called multichannel networks (MCNs). These networks offer services such as brand deals, equipment, and strategic advice in exchange for a share of the YouTubers’ revenues. A dominant strategy to gain more subscribers and, hence, revenue is collaborating with other YouTubers. Yet, collaborations on YouTube have not been studied in a detailed quantitative manner. To close this gap, first, we collect a YouTube dataset covering video statistics over 3 months for 7,942 channels. Second, we design a framework for collaboration detection given a previously unknown number of persons featured in YouTube videos. We denote this framework, for the detection and analysis of collaborations in YouTube videos using a Deep Neural Network (DNN)-based approach, as CATANA. Third, we analyze about 2.4 years of video content and use CATANA to answer research questions guiding YouTubers and MCNs for efficient collaboration strategies. Thereby, we focus on (1) collaboration frequency and partner selectivity, (2) the influence of MCNs on channel collaborations, (3) collaborating channel types, and (4) the impact of collaborations on video and channel popularity. Our results show that collaborations are in many cases significantly beneficial regarding viewers and newly attracted subscribers for both collaborating channels, often showing more than 100% popularity growth compared with noncollaboration videos.

Video streaming is responsible for the largest portion of traffic in fixed and mobile networks. Yet, forecasts expect this amount to grow further. Especially for mobile devices connected to cellular networks, high QoE video streaming can be a challenge as the user data volume is metered and eventually limited. Also, the connection quality may vary severely. Prefetching videos is an approach to mitigate this issue. Here, videos that the user is likely to watch in advance are prefetched on the user's smartphone, e.g., while he is connected to WiFi. However, this approach can only be efficient if only the videos that are interesting for the respective user are prefetched. This constitutes a major estimation and prediction challenge. To this end, this paper presents three contributions: First, a user study over multiple months that draws valuable insights on the user video request behavior. Second, we propose a novel privacy-preserving prefetching framework denoted vFetch that prefetches videos based, e.g., on the user's affinity of YouTube channels. Third, a trace-based evaluation and parameter study that demonstrates vFetch's efficiency with a hit rate of ~50% for a 50 GB cache.

Fork-Join (FJ) queueing models capture the dynamics of system parallelization under synchronization constraints, for example, for applications such as MapReduce, multipath transmission and RAID systems. Arriving jobs are first split into tasks and mapped to servers for execution, such that a job can only leave the system when all of its tasks are executed. In this paper, we provide computable stochastic bounds for the waiting and response time distributions for heterogeneous FJ systems under general parallelization benefit. Our main contribution is a generalized mathematical framework for probabilistic server scheduling strategies that are essentially characterized by a probability distribution over the number of utilized servers, and the optimization thereof. We highlight the trade-off between the scaling benefit due to parallelization and the FJ inherent synchronization penalty. Further, we provide optimal scheduling strategies for arbitrary scaling regimes that map to different levels of parallelization benefit. One notable insight obtained from our results is that different applications with varying parallelization benefits result in different optimal strategies. Finally, we complement our analytical results by applying them to various applications showing the optimality of the proposed scheduling strategies.

Multipath TCP enables remarkable optimizations for throughput, load balancing, and mobility in today's networks. The design space of Multipath TCP scheduling, i.e., the application-aware mapping of packets to paths, is largely unexplored due to its inherent complexity. Evidence in this paper suggests that an application-aware scheduling decision, if leveraged right, pushes Multipath TCP beyond throughput optimization and thereby provides benefits for a wide range of applications.

The current body of research on Dynamic Adaptive Streaming over HTTP (DASH) contributes various adaptation algorithms aiming to optimize performance metrics such as the Quality of Experience. Intuitively, the heterogeneity of the streaming environment and the underlying technologies lead many of the developed approaches to possess clear performance affinities denoted here as sweet spots. We observe, however, that systematic comparisons of these algorithms are usually conducted within homogeneous player environments.

State-of-the-art Software Defined Wide Area Networks (SD-WANs) provide the foundation for flexible and highly resilient networking. In this work we design, implement and evaluate a novel architecture (denoted SABR) that leverages the benefits of SDN to provide network assisted Adaptive Bitrate Streaming. With clients retaining full control of their streaming algorithms we clearly show that by this network assistance, both the clients and the content providers benefit significantly in terms of QoE and content origin offloading. SABR utilizes information on available bandwidths per link and network cache contents to guide video streaming clients with the goal of improving the viewer's QoE. In addition, SABR uses SDN capabilities to dynamically program flows to optimize the utilization of CDN caches.; [email protected] by our study of SDN assisted streaming we discuss the change in the requirements for network-to-player APIs that enables flexible video streaming. We illustrate the difficulty of the problem and the impact of SDN-assisted streaming on QoE metrics using various well established player algorithms. We evaluate SABR together with state-of-the-art DASH quality adaptation algorithms through a series of experiments performed on a real-world, SDN-enabled testbed network with minimal modifications to an existing DASH client. Our measurements show the substantial improvement in cache hitrates in conjunction with SABR indicating a rich design space for jointly optimized SDN-assisted caching architectures for video streaming applications.

Despite known QoE shortcomings, Dynamic Adaptive Streaming over HTTP (DASH) has been tied with TCP for many years now. The advent of HTTP/2 powered by transport protocols such as QUIC provides an excellent opportunity to revisit adaptive bitrate streaming with respect to QoE. QUIC promises improved congestion control, zero-RTT connection establishment and multiplexing logical streams. In this work, we adapt state-of-the-art DASH players with buffer-based and hybrid (rate/buffer-based) quality adaptation logic to use QUIC. Our main focus lies in contrasting the QoE performance of DASH algorithms running on top of QUIC versus TCP in various environments. Interestingly, we find through testbed and Internet measurements that QUIC does not provide a boost to current DASH algorithms but instead a degradation in the chosen quality bitrates.

Network function (NF) virtualization decouples NFs from the underlying middlebox hardware and promotes their deployment on virtualized network infrastructures. This essentially paves the way for the migration of NFs into clouds (i.e., NF-as-a-Service), achieving a drastic reduction of middlebox investment and operational costs for enterprises. In this context, service chains (expressing middlebox policies in the enterprise network) should be mapped onto datacenter networks, ensuring correctness, resource efficiency, as well as compliance with the provider's policy. The network service embedding (NSE) problem is further exacerbated by two challenging aspects: 1) traffic scaling caused by certain NFs (e.g., caches and WAN optimizers) and 2) NF location dependencies. Traffic scaling requires resource reservations different from the ones specified in the service chain, whereas NF location dependencies, in conjunction with the limited geographic footprint of NF providers (NFPs), raise the need for NSE across multiple NFPs. In this paper, we present a holistic solution to the multi-provider NSE problem. We decompose NSE into: 1) NF-graph partitioning performed by a centralized coordinator and 2) NF-subgraph mapping onto datacenter networks. We present linear programming formulations to derive near-optimal solutions for both problems. We address the challenging aspect of traffic scaling by introducing a new service model that supports demand transformations. We also define topology abstractions for NF-graph partitioning. Furthermore, we discuss the steps required to embed service chains across multiple NFPs, using our NSE orchestrator (Nestor). We perform an evaluation study of multi-provider NSE with emphasis on NF-graph partitioning optimizations tailored to the client and NFPs. Our evaluation results further uncover significant savings in terms of service cost and resource consumption due to the demand transformations.

The dynamics of the application-layer-based control loop of dynamic adaptive streaming over HTTP (DASH) make video bitrate selection for DASH a difficult problem. In this work, we provide a DASH quality adaptation algorithm, named SQUAD, that is specifically tailored to provide a high quality of experience (QoE). We review and provide new insights into the challenges for DASH rate estimation. We found that in addition to the ON-OFF behavior of DASH clients, there exists a discrepancy in the timescales that form the basis of the rate estimates across (i) different video segments and (ii) the rate control loops of DASH and Transmission Control Protocol (TCP). With these observations in mind, we design SQUAD aiming to maximize the average quality bitrate while minimizing the quality variations. We test our implementation of SQUAD together with a number of different quality adaptation algorithms under various conditions in the Global Environment for Networking Innovation testbed, as well as, in a series of measurements over the public Internet. Through a measurement study, we show that by sacrificing little to nothing in average quality bitrate, SQUAD can provide significantlygt ; better QoE in terms of quality switching and magnitude. In addition, we show that retransmission of higher-quality segments that were originally received in low-quality is feasible and improves the QoE.

In this work we outline a framework for measurement-based performance evaluation in SDN environments. The SDN paradigm, which is based on a strict separation of the network logic from the underlying physical substrate, necessitates a comprehensive global view of the network state. To augment the network representation, we propose mechanisms for extracting traffic characteristics from network observations which are used to derive performance metrics. Such metrics can be exploited by SDN applications to optimize the performance of SDN services. Given the bursty nature of network traffic and the well known adverse impact of this property on network performance, we propose an approach for extracting flow autocorrelations from switch counters. Our main contribution is a random sampling approach that reduces the monitoring overhead while enabling a fine grained characterization of the flow autocorrelation structure. We analytically evaluate the impact of random sampling and demonstrate how services may use the estimated traffic properties to compute useful performance metrics.

The application-layer based control loops of dynamic adaptive streaming over HTTP (DASH) make video bitrate selection a complex problem. In this work, we review and present new insights into the challenges of DASH rate adaptation. We identify several critical issues that contribute to the degradation of DASH performance with respect to the rate control loops of DASH and TCP. We then introduce a novel DASH quality adaptation algorithm SQUAD, which is specifically designed to ensure high quality of experience (QoE). We implement and test our algorithm together with a number of state-of-the-art quality adaptation algorithms. Through extensive experiments on both testbed and cross-Atlantic Internet scenarios, we show that by sacrificing little to none in average quality bitrate, SQUAD provides significantly better QoE in terms of number and magnitude of quality switches.

To prevent rejecting requests, cloud platforms typically provision for their peak demand. Thus, a platform's idle capacity can be significant, as demand varies widely over multiple time scales, e.g., daily and seasonally. To reduce waste, platforms have begun to offer this idle capacity in the form of transient servers, which they may unilaterally revoke, for much lower prices - ~50-90% less - than on-demand servers, which they cannot revoke. However, transient servers' revocation characteristics - their volatility and predictability - influence their performance, since they affect the overhead of fault-tolerance mechanisms applications use to handle revocations. Unfortunately, current cloud platforms offer no guarantees on revocation characteristics, which makes it difficult for users to optimally configure (and correctly value) transient servers. To address the problem, we propose the abstraction of a transient guarantee, which offers probabilistic assurances on revocation characteristics. Transient guarantees have numerous benefits: they increase the performance of transient servers, enable users to optimally use and correctly value them, and permit platforms to control their freedom to revoke them. We present policies for partitioning a variable amount of idle capacity into classes with different transient guarantees to maximize performance and value. We then implement and evaluate these policies on job traces from a production Google cluster. We show that our approach can increase the aggregate revenue from idle server capacity by up to ~6.5× compared to existing approaches.

Adaptive resource allocation arises naturally as a technique to optimize resource utilization in communication networks with scarce resources under dynamic conditions. One prominent example is cellular communication where service providers seek to utilize the costly resources in the most effective way. In this work, we investigate an uplink resource allocation scheme that takes into account the buffer occupation at the transmitter to retain a given level of quality of service (QoS). First, we regard exact results for the class of Poisson traffic where we investigate the sensitivity of the resource adaptation and QoS level to the actuating variables. We show relevant resource savings in comparison with a static allocation. Further, we regard a queueing setting with general random arrival and service processes. In particular, we consider the service of wireless fading channels. We show two different resource adaptation mechanisms that depend on the strictness of different assumptions. Finally, we present simulation results that show substantial resource savings using the queue-aware scheduling scheme, where we provide insight on the implementation and operation of such an adaptive system.

In a Fork-Join (FJ) queueing system an upstream fork station splits incoming jobs into N tasks to be further processed by N parallel servers, each with its own queue; the response time of one job is determined, at a downstream join station, by the maximum of the corresponding tasks' response times. This queueing system is useful to the modelling of multi-service systems subject to synchronization constraints, such as MapReduce clusters or multipath routing. Despite their apparent simplicity, FJ systems are hard to analyze.

The ever-increasing need to diversify the Internet has recently revived the interest in network virtualization. Wide-area virtual network (VN) deployment raises the need for VN embedding (VNE) across multiple Infrastructure Providers (InPs), due to the InP's limited geographic footprint. Multi-provider VNE, in turn, requires a layer of indirection, interposed between the Service Providers and the InPs. Such brokers, usually known as VN Providers, are expected to have very limited knowledge of the physical infrastructure, since InPs will not be willing to disclose detailed information about their network topology and resource availability to third parties. Such information disclosure policies entail significant implications on resource discovery and allocation. In this paper, we study the challenging problem of multi-provider VNE with limited information disclosure (LID). In this context, we initially investigate the visibility of VN Providers on substrate network resources and question the suitability of topology-based requests for VNE. Subsequently, we present linear programming formulations for: (i) the partitioning of traffic matrix based VN requests into segments mappable to InPs, and (ii) the mapping of VN segments into substrate network topologies. VN request partitioning is carried out under LID, i.e., VN Providers access only information which is not deemed confidential by InPs. We further investigate the suboptimality of LID on VNE against a “best-case” scenario where the complete network topology and resource availability information is available to VN Providers.

Many of today's Internet applications such as mobile web browsing and live video streaming are delay and throughput sensitive. In face of the great success of cellular networking, especially with the advent of the high-speed LTE access tech-nology, it is noteworthy that there is little consensus on the performance experienced by applications running over cellular networks. In this paper we present application-level performance results measured in a major commercial LTE network. We replicate measurements in a wired access network to provide a reference for the wireless results. We investigate the performance of common web application scenarios over LTE. In addition, we deploy controlled measurement nodes to discover transparent middleboxes in the LTE network. The introduction of middle-boxes to LTE results in a faster connection establishment on the client side and notable performance gain for HTTP. However, this improvement comes at the price of ambiguity as some middlebox operations may introduce unnecessary timeouts. Further, we pinpoint LTE specific delays that arise from network signalling, energy saving algorithms and Hybrid Automatic Repeat reQuest (HARQ). Our analysis provides insights into the interaction between transport protocols and LTE.

Universal access to Internet is crucial, and as such, there have been several initiatives to enable wider access to the Internet. Public AccessWiFi Service (PAWS) is one such initiative that takes advantage of the the available unused capacity in home broadband connections and allows Less-than-Best Effort (LBE) access to these resources, as advocated by Lowest Cost Denominator Networking (LCDNet). PAWS has been recently deployed in a deprived community in Nottingham, and, as any crowd-shared network, it faces limited coverage, since there is a single point of Internet access per guest user, whose availability depends on user sharing policies. To mitigate this problem and extend the coverage, we consider a crowd-shared wireless mesh network (WMN) in which the home routers are interconnected as a mesh. Such a network provides multiple points of Internet access and can enable resource pooling across all available paths to the Internet backhaul. In this paper, we investigate the potential benefits of a crowd-shared WMN for public Internet access by performing a comparative study between such a network and PAWS. To this end, we present a software-defined WMN control plane for the coordination of traffic redirections through the WMN and an algorithm for Internet access point selection. Our simulation results show that a crowd-shared WMN can provide much higher utilization of the shared bandwidth and can accommodate a substantially larger volume of guest user traffic.

We present AutoEmbed, a fully-automated framework for VN embedding across multiple substrate networks. To automate VN embedding, AutoEmbed deploys functions over three layers: (i) Service Providers, (ii) VN Providers, and (iii) Infrastructure Providers (InPs). AutoEmbed enables VN Providers to partition VN requests among multiple substrate networks based on resource and network topology information that is not treated as confidential by InPs.

In packet switched networks, statistical multiplexing of independent variable bit rate flows achieves significant resource savings, i.e., N flows require considerably less than N times the resources needed for one flow. In this work, we explore statistical multiplexing using methods from the current stochastic network calculus, where we compare the accuracy of different analytical approaches. While these approaches are known to provide identical results for a single flow, we find significant differences if several independent flows are multiplexed. Recent results on the concatenation of nodes along a network path allow us to investigate both single- as well as multi-node networks with cross traffic. The analysis enables us to distinguish different independence assumptions between traffic flows at a single node as well as between cross traffic flows at consecutive nodes of a network path. We contribute insights into the scaling of end-to-end delay bounds in the number of nodes n of a network path under statistical independence. Our work is complemented by numerical applications, e.g., on access multiplexer dimensioning and traffic trunk management.

The MIMO wireless channel offers a rich ground for quality of service analysis. In this work, we present a stochastic network calculus analysis of a MIMO system, operating in spatial multiplexing mode, using moment generating functions (MGF). We quantify the spatial multiplexing gain, achieved through multiple antennas, for flow level quality of service (QoS) performance. Specifically we use Gilbert-Elliot model to describe individual spatial paths between the antenna pairs and model the whole channel by an $N$-State Markov Chain, where $N$ depends upon the degrees of freedom available in the MIMO system. We derive probabilistic delay bounds for the system and show the impact of increasing the number of antennas on the delay bounds under various conditions, such as channel burstiness, signal strength and fading speed. Further we present results for multi-hop scenarios under statistical independence.

Packet switched networks achieve significant re source savings due to statistical multiplexing. In this work we explore statistical multiplexing gains in single and multi-hop networks. To this end, we analyze performance metrics such as delay bounds for a through flow comparing different results from the stochastic network calculus. We distinguish different multiplexing gains that stem from independence assumptions between flows at a single hop as well as flows at consecutive hops of a network path. Further, we show corresponding numerical results. In addition to deriving the benefits of various statistical multiplexing models on performance bounds, we contribute insights into the scaling of end-to-end delay bounds in the number of hops not a network path under statistical independence.

Fractional Brownian motion (fBm) emerged as a useful model for self-similar and long-range dependent Internet traffic. Asymptotic, respectively, approximate performance measures are known from large deviations theory for single queuing systems with fBm traffic. In this paper we prove a rigorous sample path envelope for fBm that complements previous results. We find that both approaches agree in their outcome that overflow probabilities for fBm traffic have a Weibull tail. We show numerical results on the impact of the variability and the correlation of fBm traffic on the queuing performance.

Fractional Brownian motion (fBm) became known as a useful model for Internet traffic incorporating its self-similar and long-range dependent properties. In this paper we derive end-to-end performance bounds for a through flow in a network of tandem queues under fBm cross traffic. We build on a previously derived sample path envelope for fBm, which possesses a Weibullian decay of overflow probabilities. We employ the sample path envelope and the concept of leftover service curves to model the remaining service after scheduling fBm cross traffic at a system. Using composition results for tandem systems from the stochastic network calculus we derive end-to-end statistical performance bounds for individual flows in networks under fBm cross traffic. We discover that these bounds grow in O(n(log n)1/2-2H) for n systems in series where H is the Hurst parameter of the fBm cross traffic. We show numerical results on the impact of the variability and the correlation of fBm traffic on network performance.