Power Profiling the Internet Core: A Sensitivity Analysis

Power Profiling the Internet Core: A Sensitivity Analysis

Aruna Prem Bianzino (Telecom ParisTech, France), Anand Raju (Telecom ParisTech, France) and Dario Rossi (Telecom ParisTech, France)
Copyright: © 2012 |Pages: 22
DOI: 10.4018/978-1-4666-1842-8.ch013
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

In this chapter, the authors perform a careful sensitivity analysis of a power model for the Internet core: their results show that, no matter how carefully the data upon which the power-consumption model relies is chosen and cross-verified, the uncertainty of the overall results remains disappointingly high. The authors believe that part of the solution lies in a community-wide effort, to which they offer some initial guidance that could, if not solve the issue, at least greatly improve the current situation.
Chapter Preview
Top

Introduction

Research on green ICT is evidently gaining momentum: a rich literature exists and continues to grow on energy-thrifty networks solutions and on network power profiling analysis. As a comprehensive overview of network energy-efficiency issues is out of the scope of this chapter, we refer the reader to the literature survey by Bianzino et al. (2011) or to earlier chapters of this book.

One of the most challenging aspects in green research is to gather a set of energy-related assumptions, such as device power profiles, which is both accurate for present-day devices and future-proof as well. Indeed, even considering present-day systems, it is often difficult to estimate various aspects of power measurements – as for example the power required by cooling in Katz (2009), or discrepancies between actual power drain and maximum drain reported by the equipment manufacturer in Juniper (2009a), and so on. Considering instead futuristic scenarios, it is clear that major advances originating from other areas are hardly foreseeable, but may deeply impact the overall results—as has been the case in recent years for voltage scaling (Weiser, et al., 1994) from a hardware perspective, for tickless kernels (LessWatts Project, 2009) from a software standpoint, and for adaptive link rate (Christensen, et al., 2010) from the communication networking front.

To further add complexity to the picture, there are many instances of inaccurate or grossly misinterpreted results published by energy analysts and media. Moreover, errors can easily propagate, as wrong numbers replicate themselves through direct references and citations, possibly under the camouflage due to manipulation. The above statement is especially true when the estimations involve rapidly changing technologies, requiring constant updates and time-to-time verifications. Developing methodologies and performing measurement is often challenging, e.g., because of precluded access to the real infrastructures, or proprietary data-centers, very large-scale networks spanning multiple domains, etc. Thus, unfortunately, gauging the exact nuance of green in the ICT context is inherently uncertain, hence prone to fallacies.

Evidence of existing fallacies in the literature are uncovered only sporadically, if ever. The best-known example is represented by a Forbes magazine article by Huber and Mills (1999), which claimed that PCs and networked devices were responsible for 8% of all electricity consumed across USA. Huber and Mills (1999) also projected a staggering growth up to 50% of all electricity usage in the following 10-20 years. These numbers were widely published as well as publicized by the media, but were later debunked by a study conducted at Lawrence Berkeley National Laboratory (LBNL) by Koomey et al. (2002), which resized the estimation to about 3%, for all office, telecommunications, and network equipment.

Along similar lines, more recently, an article published by Wissner-Gross (2009) on BBC News estimated that a couple of Google searches on a desktop computer produces about 14 g of CO2, which is roughly the equivalent of boiling an electric kettle. These figures were later countered by Google (2009), claiming instead that a typical search produces only 0.2 g of CO2. Relative overestimation in this case is in the order of 70×, which raises the question on how to disprove the erroneous figures, or at least whether the gap in such diverse estimations could be significantly narrowed down, e.g., by using more accurate input values for power models.

Complete Chapter List

Search this Book:
Reset