Download files
Abstract
Improving energy efficiency of Internet equipment is becoming an increasingly important research topic, motivated by the need to reduce energy costs (and Carbon footprint) for Internet Service Providers, as well as increase of power density to achieve more switching capacity per-rack. To clearly understand how power consumption of Internet equipment is determined by the network elements such as sleep or active states of the device components, number of active ports or links, traffic patterns and network architectures, researchers have profiled the energy demand of commercial routing platforms and proposed ways to reduce network power consumption. However, these early works usually profile only coarse-grained power models (i.e., at the granularity of per line-card or per port) and the proposed power saving solutions involve significant architectural and/or protocol changes in the network. The cost and risk associated with such drastic changes increase the barrier to adoption by network operators, thus stretching the time-horizon at which they become practical for wide-scale deployment.
In this thesis we profile fine-grained power models for Internet equipment and propose a power saving scheme that requires minimal changes to existing router design, carries little risk of impacting network performance, is almost entirely transparent to network operators and is ready for incremental deployment.
We first profile power consumption of NetFPGA, an increasingly popular routing platform for networking research due to its versatility and low-cost, by conducting several experiments that allow us to decompose the power consumption of the NetFPGA routing card into fine-grained per-packet and per-byte components with reasonable accuracy. This work opens the doors for estimating network-wide energy footprints at the granularity of traffic sessions and applications (e.g., due to TCP file transfers), and provides a benchmark against which energy improvements arising from new architectures and protocols can be evaluated.
Second, we analyse the power consumption of Energy Efficient Ethernet (EEE) switches in several experiments and based on the results propose a power model to profile them. Energy Efficient Ethernet is an IEEE standard (802.3az) for improving energy efficiency in Ethernet devices, which was newly issued in Sep. 2010. Our work is the first evaluation of power consumption of EEE switches. The proposed model can be used to predict the energy savings when deploying the new switches and also for research on further power saving techniques such as energy efficient routing or dynamic link shutdown.
Third, we propose a simple and practical algorithm for activating buffers in backbone router line-cards incrementally as needed and putting them to sleep when not in use. We evaluate our algorithm on traffic traces from carrier and enterprise networks, via simulations in ns2, and by implementing it on a programmable-router test-bed. Our study shows that much of the energy associated with off-chip packet buffers can be eliminated with negligible impact on traffic performance. Dynamic adjustment of active router buffer size provides a low-complexity low-risk mechanism of saving energy that is amenable for incremental deployment in networks today.