The integration of graphics processing units through virtualization offers significant potential to optimize resource utilization in distributed systems. Particularly in edge computing scenarios, the architecture of edge devices can be simplified by consolidating GPU usage at the edge gateway, which offers the improvement of the GPU utilization by sharing the GPU by multiple edge client devices. However, the effectiveness of this approach is heavily contingent upon network communication, which in edge environments often lacks the high-speed and low latency links as well as specialized protocols available in traditional data center settings for high performance computing, especially if wireless networks are deployed. Therefore, our study aims to comprehensively analyze and evaluate the performance implications of using the conventional transmission control protocol (TCP) versus the emerging QUIC protocol in such environments. We specifically focus on assessing QUIC's unique features, including concurrent streams, stream cancellation, and stream prioritization, which are not inherently present in TCP. Furthermore, we investigate the potential performance gains achievable through the adoption of asynchronous API calls, aiming to provide insights into optimizing GPU utilization in edge computing. We show the advantages of most of QUIC's features and asynchronous API calls by an experimental evaluation.

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.

Combining multiple networks at a host to access the Internet (multi-homing) allows for continuous data transmission even in scenarios with one intermittent connection. To this end, a multi-path capable transport layer protocol, such as multi-path TCP (MPTCP), is required to simultaneously use two or more networks for data transmission. A typical use case is to combine WLAN and cellular networks, in which the user stays connected to the cellular network, if the WLAN connection is lost due to user mobility. In that scenario, multi-path protocols have to detect the connection loss and have to retransmit corresponding lost packets immediately, so that the receiving network stack can forward the packets in sequence to the application. This paper analyzes MPTCPs connection loss detection mechanism and the corresponding process of the reinjection of lost packets. We show that the mechanisms for connection loss detection originate from single-path TCP (SPTCP) and have been simply adopted to MPTCP. They work slowly and cause large application level delays. Therefore, we propose a new reinjection method to speed up the reinjection after a connection loss. We evaluate our approach with emulation and real-world experiments and show that we reduce application level delays by up to 80%.

In multi-path scenarios, the default multi-path TCP (MPTCP) scheduler prefers the subflow with the minimum round-trip time (RTT) to send packets. Therefore, the scheduler requires reliable and timely information on the RTTs for the individual subflows to make proper scheduling decisions. We show for various situations that a reliable and timely RTT estimate is not always given. This shortcoming is induced from the original TCP behavior that the RTT is only updated if data is transmitted on a TCP flow. This means for MPTCP, that a subflow, which suffers from a short-term RTT increase, may not be selected again by the scheduler and so never gets an RTT update. Moreover, we demonstrate for various scenarios, that this leads to non-optimal scheduling decisions. We propose an RTT probing scheme to achieve reliable and timely RTT updates. Our probing scheme is designed for a fast reaction as well as little probing traffic. The advantages are demonstrated in emulated as well as real-world vehicular experiments.

Selecting the best path in multi-path heterogeneous networks is challenging. Multi-path TCP uses by default a scheduler that selects the path with the minimum round trip time (minRTT). A well-known problem is head-of-line blocking at the receiver when packets arrive out of order on different paths. We shed light on another issue that occurs if scheduling have to deal with deep queues in the network. First, we highlight the relevance by a real-world experiment in cellular networks that often deploy deep queues. Second, we elaborate on the issues with minRTT scheduling and deep queues in a simplified network to illustrate the root causes; namely the interaction of the minRTT scheduler and loss-based congestion control that causes extensive bufferbloat at network elements and distorts RTT measurement. This results in extraordinary large buffer sizes for full utilization. Finally, we discuss mitigation techniques and show how alternative congestion control algorithms mitigate the effect.

Intelligent Transportation Systems (ITS) and vehicle passengers request for high-speed and reliable data access from vehicle to infrastructure networks. For example, passengers like to consume multimedia services and new use cases arise from ITS and connected vehicular services that require high-speed and reliable connectivity solutions. To achieve these requirements multi-connectivity is a viable solution to increase throughput and reliability. Such options exist on various layers. On the network access layer multiple antennas are deployed for transmitting and receiving data in parallel data streams; on the transport layer different network access opportunities are utilized by multi-path protocols such as multi-path TCP. At the present study, we evaluate performance metrics of these options in a real-world vehicular scenario. We quantify throughput, delay, and reliability improvements of using multiple antennas and multi-path TCP and also indicate issues of multi-path TCP. Further, we demonstrate performance improvements by the change of the congestion control protocol.

In the context of continually growing data volumes, which have to be processed and transmitted in modern and future road traffic, the importance of the LTE technology will increase. While the extensively evaluated 802.11p standard shows its strength in regards to road safety applications, data-intensive comfort and infotainment services, which cause high data volume mainly in downlink direction, can best be provided by LTE. Therefore, the investigation of the download performance and availability is essential. In this paper, we present measurement results that evaluate how LTE's download speed is affected by relevant parameters. We perform two measurement campaigns in a major commercial LTE network. In a mobile scenario, we analyze position dependent parameters, such as frequency bands, the corresponding bandwidth, and the channel's signal quality, i.e., SINR and RSSI. Furthermore, a statistical performance evaluation provides insight into the parameters' impact on the download speed and their consistency in urban and rural areas. In addition, we compare the influence of main transport protocols, i.e., TCP and UDP. Finally, we identify diurnal and weekday dependent patterns. Concerning the download speeds in LTE, our results show on the one hand a higher spatial variability in cities and on the other hand that the signal quality and channel bandwidth are the most important influencing factors.

Closed-loop flow control protocols, such as the prominent implementation transmission control protocol (TCP), are prevalent in the Internet, today. TCP has continuously been improved for greedy traffic sources to achieve high throughput over networks with large bandwidth delay products. Recently, the increasing use for streaming and interactive applications, such as voice and video, has shifted the focus toward its delay performance. Given the need for real-time communication of non-greedy sources via TCP, we present an estimation method for performance evaluation of closed-loop flow control protocols. We characterize an end-to-end connection by a service curve that provides statistical guarantees for arbitrary traffic. The estimation is based on end-to-end measurements at the application level that include all effects induced by the network and by the protocol stacks of the end systems. From our measurements, we identify different causes for delays. We show that significant delays are due to queueing in protocol stacks. Notably, this occurs even if the utilization is moderate. Using our estimation method, we compare the impact of fundamental mechanisms of TCP on delays at the application level: in detail, we analyze parameters relevant for network dimensioning, including buffer provisioning and active queue management, and parameters for server configuration, such as the congestion control algorithm. By applying our method as a benchmark, we find that a good selection can largely improve the delay performance of TCP.

TCP is widely used for delay-sensitive applications including video streaming, voice, and instant messaging. Therefore, a performance parameter of interest is the end-to-end delay at the application level. In the past, much research was performed in the optimization of TCP's throughput. Besides throughput, delay is an important issue in todays networks. However, it was often disregarded. Amongst others, extensive queueing inside the network induces large delays that account for a significant share of the end-to-end delay. This phenomenon is termed bufferbloat. We show that another reason for extensive delays is queueing in the TCP stack that originates primarily from TCP's congestion control. We analyze this using the analogy of a queueing model for end-to-end delays from simulating TCP's congestion control mechanism. The causes of these delays could be identified and the connection between basic TCP mechanisms and network dimensioning and configuration is presented.

Closed-loop flow control protocols, such as the prominent implementation TCP, are prevalent in the Internet, today. TCP has continuously been improved for greedy traffic sources to achieve high throughput over networks with large bandwidth delay products. Recently, the increasing use for streaming and interactive applications, such as voice and video, has shifted the focus towards its delay performance. Given the need for real-time communication of non-greedy sources via TCP, we present an estimation method for performance evaluation of closed-loop flow control protocols. We characterize an end-to-end connection by a transfer function that provides statistical service guarantees for arbitrary traffic. The estimation is based on end-to-end measurements at the application level, that include all effects induced by the network and by the protocol stacks of the end systems. From our measurements, we identify different causes for delays. We show that significant delays are due to queueing in protocol stacks. Notably, this occurs even if the utilization is moderate. Using our estimation method, we compare the impact of fundamental mechanisms of TCP. In detail, we analyze buffer provisioning and its impact on delays at the application level. We find that a good selection can largely improve the delay performance of TCP.

This paper develops a notion of capacity-delay-error (CDE) boundaries as a performance model of networked sources and systems. The goal is to provision effective capacities that sustain certain statistical delay guarantees with a small probability of error. We use a stochastic non-equilibrium approach that models the variability of traffic and service to formalize the influence of delay constraints on the effective capacity. Permitting unbounded delays, known ergodic capacity results from information theory are recovered in the limit. We prove that the model has the property of additivity, which enables composing CDE boundaries obtained for sources and systems as if in isolation. A method for construction of CDE boundaries is devised based on moment-generating functions, which includes the large body of results from the theory of effective bandwidths. Solutions for essential sources, channels, and respective coders are derived, including Huffman coding, MPEG video, Rayleigh fading, and hybrid automatic repeat request. Results for tandem channels and for the composition of sources and channels are shown.

In this paper we present spectrum usage results from a measurement campaign conducted in Berlin (Germany) during the International Radio Exhibition in September 2013. The measurements were performed in order to understand the characteristics of spectrum usage in the UHF TV band within a large indoor area. The measurement campaign relies on a spectrum sensing grid which is installed across multiple halls on the fairground at the Messe Berlin. We present our measurement setup and report our main findings regarding duty cycles and spatial correlation. Additionally, we focus on the analysis of possible coexistence threats for professional wireless microphone systems.

Numerous methods for available bandwidth estimation have been developed for wireline networks, and their effectiveness is well-documented. However, most methods fail to predict bandwidth availability reliably in a wireless setting. It is accepted that the increased variability of wireless channel conditions makes bandwidth estimation more difficult. However, a (satisfactory) explanation why these methods are failing is missing. This paper seeks to provide insights into the problem of bandwidth estimation in wireless networks or, more broadly, in networks with random service. We express bandwidth availability in terms of bounding functions with a defined violation probability. Exploiting properties of a stochastic min-plus linear system theory, the task of bandwidth estimation is formulated as inferring an unknown bounding function from measurements of probing traffic. We present derivations showing that simply using the expected value of the available bandwidth in networks with random service leads to a systematic overestimation of the traffic departures. Furthermore, we show that in a multihop setting with random service at each node, available bandwidth estimates requires observations over (in principle infinitely) long time periods. We propose a new estimation method for random service that is based on iterative constant-rate probes that take advantage of statistical methods. We show how our estimation method can be realized to achieve both good accuracy and confidence levels. We evaluate our method for wired single-and multihop networks, as well as for wireless networks.

This paper establishes a link between information and queueing theory using an envelope-based approach. Unlike classical, equilibrium information theory, information envelopes focus on the dynamics of sources and coders, using functions of time that bound the number of bits generated. In the limit the information envelopes converge to the expected value and recover the entropy of a source, respectively, the average codeword length of a coder. In contrast, on short time scales and for sources with memory it is shown that large deviations from known equilibrium results occur with non-negligible probability. These can cause significant network delays. Using results from the stochastic network calculus, the envelopes yield a characterization of the operating points of source coders by the triplet of capacity, delay, and error. In the limit, assuming an optimal coder the required capacity approaches the entropy with arbitrarily small probability of error if infinitely large delays are permitted. We derive a corresponding characterization of channels and prove that the model has the desirable property of additivity, that allows analyzing sources and channels as if in isolation.

We develop a stochastic foundation for bandwidth estimation of networks with random service, where bandwidth availability is expressed in terms of bounding functions with a defined violation probability. Exploiting properties of a stochastic max-plus algebra and system theory, the task of bandwidth estimation is formulated as inferring an unknown bounding function from measurements of probing traffic. We derive an estimation methodology that is based on iterative constant rate probes. Our solution provides evidence for the utility of packet trains for bandwidth estimation in the presence of variable cross traffic. Taking advantage of statistical methods, we show how our estimation method can be realized in practice, with adaptive train lengths of probe packets, probing rates, and replicated measurements required to achieve both high accuracy and confidence levels. We evaluate our method in a controlled testbed network, where we show the impact of cross traffic variability on the time-scales of service availability, and provide a comparison with existing bandwidth estimation tools.

Recent deployment of IP based multimedia distribution, especially broadcast TV distribution has increased the importance of simple and fast restoration during IP network failures for service providers. In this paper, we propose and evaluate a simple but efficient method for fast rerouting of IP multicast traffic during link failures in managed IPTV networks. More specifically, we devise an algorithm for tuning IP link weights so that the multicast routing path and the unicast routing path between any two routers are failure disjoint, allowing us to use unicast IP encapsulation for undelivered multicast packets during link failures. We demonstrate that, our method can be realized with minor modification to the current multicast routing protocol (PIM-SM). We run our prototype implementation in Emulab which shows our method yields to good performance.