Stein Kristiansen

Challenges and techniques for video streaming over mobile ad hoc networks

Developments in mobile devices and wireless networking provide the technical platform for video streaming over mobile ad hoc networks (MANETs). However, efforts to realize video streaming over MANETs have met many challenges, which are addressed by several different techniques. Examples include cross-layer optimization, caching and replication, and packet prioritization. Cross-layer optimization typically leverages multiple description video coding and multipath routing to provide the receiver(s) sufficient video quality. Caching and replication add tolerance to disruptions and partitioning. In this paper, we identify the challenges of realizing video streaming over MANETs, and analyze and classify the proposed techniques. Since 65 % of the identified involve cross-layering design, we study the distribution of joint optimization and parameter exchanges. Due to the importance and complexity of evaluating the techniques, we analyze the common methods, indicating that the research domain suffers from a problem of comparability.

On the forwarding capability of mobile handhelds for video streaming over MANETs

Despite the importance of real-world experiments, nearly all ongoing research activities addressing video streaming over MANETs are based on simulation studies. Earlier research shows that the limited resources of mobile handhelds, which are not modeled in most network simulators, can be a severe bottleneck. We study the capability of a modern handheld to perform one core task, which is the forwarding of video streams. We present end-to-end video quality and network measurements, along with an analysis of resource consumption. Our studies of the recent handheld Nokia N900 show that it can forward up to 3.70 Mbps. However, subjective video quality is compromised already at 3.35 Mbps, due to excessive delay. Our analysis unveils that direct memory access (DMA) relieves the CPU of forwarding overhead and that, due to the digital signal processor (DSP) support, additional coding overhead does not decrease the forwarding capacity. Finally, we find that power management impacts results considerably. It is possible to increase the forwarding capacity up to 27.4% by increasing the frequency of internal buses. Hence, our results demonstrate that the forwarding capacity is highly dependent on the internal state and activity of the device.

ns-2 distributed clients emulation: accuracy and scalability

ns-2 is a well known network simulator, recently extended with improvements to its emulation facility. Real-time constraints and the boundary between real-world and simulated entities impose scalability and accuracy limitations, and distort the simulated network as perceived by the involved real-world applications. This paper presents results from a performance evaluation of the ns-2 emulation facility. Conducting emulation experiments of differing magnitudes, and under varying emulation environment set-ups, we unveil central types of scalability limitations and obtainable accuracy. We find throughput limits using high and low end computers, and a significant throughput decrease when increasing the number of involved real-world applications. We furthermore show how end-to-end delay increases both with traffic load and an increasing number of involved real-world applications. Moreover, during these conditions, we find that the system treats these applications increasingly unfair by distributing total throughput unevenly between them, and by imposing different amounts of end-to-end delay.

Accuracy and scalability of ns-2's distributed emulation extension

ns-2 is a well known network simulator, recently extended with improvements to its emulation facility. Real-time constraints and the boundary between real-world and simulated entities impose scalability and accuracy limitations, and distort the simulated network as perceived by the involved real-world applications. This paper presents results from a performance evaluation of the ns-2 emulation facility. By carrying out emulation experiments at different scales and in different scenarios, we identify scalability limitations and quantify the emulation accuracy. We find throughput limits using high- and low-end computers, and a significant throughput decrease when increasing the number of involved real-world applications. Furthermore, we show how the end-to-end delay increases with both the traffic load and the number of real-world applications. In experiments with many applications, we also uncover significant variations in throughput and end-to-end delay among flows. Finally, our jitter measurements indicate how kernel scheduling causing bursty traffic transmission is responsible for most jitter. This effect is most often relatively modest, yet somewhat variable for experiments including several applications. Based on analysis of our results, we propose a set of modifications of the system that improves the performance. Our implementation of one of these modifications are demonstrated to increase throughput significantly while reducing the variation among flows. Implementation and evaluation of the remaining modifications is reserved for future work.

Modeling communication software execution for accurate simulation of distributed systems

Network simulation is commonly used to evaluate the performance of distributed systems, but these approaches do not account for the performance impact that protocol execution on nodes has on performance, which may be significant. We propose a methodology to capture execution models from communication software running on real devices where the execution models can be integrated with discrete event network simulators to improve their accuracy. We provide a set of rules to instrument the software to obtain the events of importance, and present techniques to create executable models based on the obtained traces. To make the models scalable, processing stages are reduced to statistical distributions. When the resulting models are executed in a device model with a scheduler simulator, we are able to model the dynamics of multithreading and parallel execution. Our initial results from a proof-of-concept extension to Ns-3 show that our models are able to accurately model protocol execution on the Google Nexus One with low simulation overhead.

Extending network simulators with communication software execution models

Existing network simulators do not account for the overhead of communication software execution, which can be significant when devices in the network are resource constrained, like in MANETs and sensor networks. We propose an approach to extend network simulators to model the execution of communication software and to map the resulting behaviour onto existing protocol models. The approach can be used to extend arbitrary network simulators to model heterogeneous hardware and software in a scalable manner. We evaluate a proof of concept implementation in Ns-3 that follows the design principles and techniques provided. The results demonstrate that our models are both accurate and scalable.

A Methodology to Model the Execution of Communication Software for Accurate Network Simulation

Network simulation is commonly used to evaluate the performance of distributed systems, but these approaches do not account for the performance impact that protocol execution on nodes has on performance, which can be significant. We provide a methodology to extract from real devices models of communication software execution that can be used to extend network simulators to improve their accuracy. The models are obtained by instrumenting the target devices to obtain the events necessary to describe software execution. We specify which events must be captured, how to capture them, and how to transform the event traces into models that can be used to extend network simulators. The obtained models are based on high-level abstractions that can be used to describe the execution of a wide range of communication software, and the design principles to extend network simulators are not restricted to any specific network simulator. The same model of communication software execution can be used without modification in all discrete event-based network simulators that are extended according to our principles. The models are represented in a human-readable format that is suitable for modification and can therefore be used to predict how software modifications impact performance. We evaluate our models with two proof-of-concept extensions of Ns-3 that execute the models of two modern smartphones: the Google Nexus One (GN1) and the Nokia N900. We measure the accuracy of our models by comparing results from real experiments with those from simulations with our models and analyze the simulation overhead of our approach.

MAC layer support for delay tolerant video transport in disruptive MANETs

The overall goal of this work is to improve video delivery in emergency and rescue scenarios using sparse MANETs that might be prone to frequent link breaks and network partitions. The core idea of our approach is to reduce the number of MAC layer retransmissions that are likely to fail. We do not drop packets that could not be sent after the final retransmission. Instead we handle them in an overlay for storecarry-forwarding. The design of the overlay protocol takes the instability of the network into account, in such a way that each overlay entity works autonomously and keeps a minimum amount of state. Our experimental results show that we reduce packet loss seen on broken links, while at the same time significantly reducing overhead in terms of the total amount of packets transmitted at the physical layer.

DCEP-Sim: An Open Simulation Framework for Distributed CEP

Distributed Complex Event Processing (CEP) is gaining increasing interest for two reasons: (1) to scale system performance to handle higher workloads in real-time, and (2) to perform in-network processing, e.g., in mobile networks to reduce the amount of data that has to be transferred through the network. System scalability and the complexity of mobile systems are some of the major challenges when evaluating the performance of new Distributed CEP solutions. We propose an open framework for distributed CEP (DCEP-Sim) built on a well-established network simulator, i.e, ns-3. The design of DCEP-Sim is based on the engineering principles of separation of concerns and the separation of mechanisms and policies. By leveraging the ns-3 feature of object aggregation it is very easy to add new policies, e.g., placement or selection policies, and evaluate them without changing anything else in the DCEP-Sim. The fact that ns-3 includes many accurate network models implies that Distributed CEP simulation with DCEP-Sim will also be much more accurate than ad-hoc handcrafted simulations. We demonstrate in a use case how easy it is to configure performance evaluation experiments and we perform experiments to confirm that the integration of the Distributed CEP in ns-3 is good foundation for large-scale experiments. The evaluation results demonstrate that DCEP-Sim substantially reduces the effort and costs of Distributed CEP evaluation.

Mobile Distributed Complex Event Processing—Ubi Sumus? Quo Vadimus?

One important class of applications for the Internet of Things is related to the need to gain timely and continuous situational awareness, like smart cities, automated traffic control, or emergency and rescue operations. Events happening in the real-world need to be detected in real-time based on sensor data and other data sources. Complex Event Processing (CEP) is a technology to detect complex (or composite) events in data streams and has been successfully applied in high volume and high velocity applications like stock market analysis. However, these application domains faced only the challenge of high performance, while the Internet of Things and Mobile Big Data introduce a new set of challenges caused by mobility. This chapter aims to explain these challenges and give an overview on how they are solved respectively how far state-of-the-art research has advanced to be useful to solve Mobile Big Data problems. At the infrastructure level the main challenge is to trade performance against resource consumption; and operator placement is the most dominant mechanism to address these problems. At the application and consumer level, mobile queries pose a new set of challenges for CEP. These are related to continuously changing positions of consumers and data sources, and the need to adapt the query processing to these changes. Finally, proper methods and tools for systematical testing and reproducible performance evaluation for mobile distributed CEP are needed but not yet available.