Fabian Kaup

PowerPi: Measuring and modeling the power consumption of the Raspberry Pi

An increasing number of households is connected to the Internet via DSL or cable, for which home gateways are required. The optimization of these - caused by their large number - is a promising area for energy efficiency improvements. Since no power models for home gateways are currently available, the optimization of their power state is not possible. This paper presents PowerPi, a power consumption model for the Raspberry Pi which is used as a substitute to conventional home gateways to derive the impact of typical hardware components on the energy consumption. The different power states of the platform are measured and a power model is derived, allowing to estimate the power consumption based on CPU and network utilization only. The proposed power model estimates the power consumption resulting in a RMSE of less than 3.3%, which is slightly larger than the maximum error of the measurements of 2.5%.

Upgrading Wireless Home Routers for Enabling Large-Scale Deployment of Cloudlets

Smartphones become more and more popular over recent years due to their small form factors. However, such mobile systems are resource-constrained in view of computational power, storage and battery life. Offloading resource-intensive tasks (aka mobile cloud computing) to distant (e.g., cloud computing) or closely located data centers (e.g., cloudlet) overcomes these issues. Especially, cloudlets provide computational power with low latency for responsive applications due to their proximity to mobile users. However, a large-scale deployment of range-restricted cloudlets is still an open challenge. In this paper, we propose a novel concept for a large-scale deployment of cloudlets by upgrading wireless home routers. Beside router’s native purpose of routing data packets through the network, it can now offer computing resources with low latency and high bandwidth without additional hardware. Proving our concept, we conducted comprehensive benchmark tests against existing concepts. As result, the feasibility of this concept is shown and provide a promising way to large-scale deploy cloudlets in existing infrastructures.

Measuring and modeling the power consumption of OpenFlow switches

The power consumption of network devices contributes to a considerable fraction of the energy expenses of data center and service providers. Recently, Software Defined Networking (SDN) has emerged as a novel networking paradigm that allows optimizing the traffic in a variety of ways, ranging from the Ethernet layer to the network layer and above. This makes SDN a versatile approach for a large number of applications. However, little is known about the power consumption of SDN-enabled networking devices, especially OpenFlow switches. This paper presents measurements and derived power consumption models for two SDN devices, namely an OpenFlow-based hardware switch (NEC PF 5240) and the Open vSwitch running on server grade hardware. The effect of configuration, management, and the managed traffic on the resulting power consumption is evaluated. Based on these measurements, a power model is derived, allowing for an estimation of the power consumption based on the configuration and network traffic only. Being available on the OpenFlow controller, the information about the power model enables an estimation of the power consumption of the full network without additional measurement equipment. The resulting model shows an error of less than 8% for the software switch and less than 1% for the hardware switch.

EnerSim: An energy consumption model for large-scale overlay simulators

Determining the energy consumption in large-scale overlay simulations is still an open issue as most existing simulation frameworks are agnostic to that aspect. Especially simulations including mobile devices, such as smartphones or tablet PCs, can benefit from having a energy consumption model in simulations such that newly developed large-scale overlay mechanisms can be evaluated with respect to their energy consumption on mobile devices. Therefore, this paper proposes a component-based energy consumption model, which is based on measurements of existing smartphones. The model causes little computational overhead, thus, being suitable for large-scale simulations. A brief evaluation shows that using our model, the energy consumption can be estimated with a mean error of ±4.7%. Furthermore, the measurements conducted to derive the model show that WiFi and Bluetooth communication are one, respectively two, orders of magnitude more energy efficient than cellular communication.

Can Multipath TCP save energy? A measuring and modeling study of MPTCP energy consumption

Mobile data consumption has seen a considerable rise and is predicted to further increase. This is mainly caused by mobile video consumption. A promising solution is offloading cellular traffic to WiFi. Here, Multipath TCP (MPTCP) enables a seamless user experience for handover and load balancing. However, its influence on the energy consumption of smartphones is not yet well analyzed. To this end, this paper determines the energy cost of using MPTCP for mobile constant bit rate streaming on two Nexus devices and compares it to the power consumption using single interface TCP streaming. From this, recommendations for an optimal assignment of sub-streams to interfaces are derived. The measurements indicate that using MPTCP on certain smartphones causes a 20% lower energy expense compared to the cost of both interfaces individually. However, MPTCP using multiple interfaces should only be used if the requested data rate cannot be supported by a single interface.

Impact of WiFi offloading on video streaming QoE in urban environments

Video streaming is the most popular application in todays mobile Internet and its growing demands and popularity put more and more load on cellular networks. In a recent trend to mitigate the cellular load, followed by many providers, users are offered to offload mobile connections to WiFi hotspots, which are predominately deployed in urban environments. In this work, we conduct a simulative performance evaluation of the impact of WiFi offloading on the Quality of Experience (QoE) of video streaming. The evaluation is based on connectivity measurements from a German city and uses a simple QoE model for estimating the perceived quality of video streaming. Our findings show that, despite its benefits for operators, offloading to WiFi has a negative impact on video streaming QoE for some users when 3G/4G coverage is available. Only in the case of 2G coverage, WiFi offloading can significantly improve the perceived quality for users.

Behind the NAT — A measurement based evaluation of cellular service quality

Mobile applications such as VoIP, (live) gaming, or video streaming have diverse QoS requirements ranging from low delay to high throughput. The optimization of the network quality experienced by end-users requires detailed knowledge of the expected network performance. Also, the achieved service quality is affected by a number of factors, including network operator and available technologies. However, most studies focusing on measuring the cellular network do not consider the performance implications of network configuration and management. To this end, this paper reports about an extensive data set of cellular network measurements, focused on analyzing root causes of mobile network performance variability. Measurements conducted over four weeks in a 4G cellular network in Germany show that management and configuration decisions have a substantial impact on the performance. Specifically, it is observed that the association of mobile devices to a Point of Presence (PoP) within the operators network can influence the end-to-end RTT by a large extent. Given the collected data a model predicting the PoP assignment and its resulting RTT leveraging Markov Chain and machine learning approaches is developed. RTT increases of 58% to 73% compared to the optimum performance are observed in more than 57% of the measurements.

Lightweight mobile bandwidth availability measurement

Mobile data traffic is increasing rapidly and wireless spectrum is becoming a more and more scarce resource. This makes it highly important to operate the mobile network efficiently. In this paper we are proposing a novel lightweight measurement technique that can be used as a basis for advanced resource optimization algorithms to be run on mobile phones. Our main idea leverages an original packet dispersion based, technique to estimate both per user capacity and asymptotic dispersion rate. This allows passive measurements using only existing mobile traffic. Our technique is able to efficiently filter outliers introduced by mobile network schedulers. In order to verify the feasibility of our measurement technique, we run a week-long measurement campaign spanning two cities in two countries, different radio technologies, and covering all times of the day. The campaign demonstrates that our technique is effective even if it is provided only with a small fraction of the exchanged packets of a flow. The only requirement for the input data is that it should consist of a few consecutive packets that are gathered periodically. This makes the measurement algorithm a good candidate for inclusion in OS libraries to allow for advanced resource optimization and application-level traffic scheduling, based on current and predicted future user capacity.

Assessing the Implications of Cellular Network Performance on Mobile Content Access

Mobile applications such as VoIP, (live) gaming, or video streaming have diverse QoS requirements ranging from low delay to high throughput. The optimization of the network quality experienced by end-users requires detailed knowledge of the expected network performance. Also, the achieved service quality is affected by a number of factors, including network operator and available technologies. However, most studies measuring the cellular network do not consider the performance implications of network configuration and management. To this end, this paper reports about an extensive data set of cellular network measurements, focused on analyzing root causes of mobile network performance variability. Measurements conducted on a 4G cellular network in Germany show that management and configuration decisions have a substantial impact on the performance. Specifically, it is observed that the association of mobile devices to a point of presence (PoP) within the operators network can influence the end-to-end performance by a large extent. Given the collected data, a model predicting the PoP assignment and its resulting RTT leveraging Markov chain and machine learning approaches is developed. RTT increases of 58% to 73% compared to the optimum performance are observed in more than 57% of the measurements. Measurements of the response and page load times of popular websites lead to similar results, namely, a median increase of 40% between the worst and the best performing PoP.

Optimizing energy consumption and qoe on mobile devices

The increased availability and data rates of cellular 3G/4G networks combined with the growing use of mobile applications highly affect the Quality of Experience (QoE) perceived by the end-user. The QoE is affected in two ways: First, the data rates in the networks are low when multiple users simultaneously request content; second, the transmission of data over slow connections consumes a considerable amount of energy compared to faster connections. Both can be avoided by better management of the available resources. This paper proposes a new approach, taking the energy efficiency into account as a key QoE aspect. Based on user mobility models, the available connectivity can be predicted, from which estimates for the energy consumption and expected QoE can be derived. An architecture is sketched, which combines QoE prediction for current and future network connections with energy efficiency on mobile devices.