Newsletter‎ > ‎

Newsletter - Volume 4 Issue 5 - July 2016


Monthly bulletin of the IEEE Computer Society Special Technical Community on Sustainable Computing

Providing quick access to timely information on sustainable computing.

Sponsored by:





Share It!


Message from the editor

Cristina Rottondi, Dalle Molle Institute for Artificial Intelligence


I am delighted to introduce the July issue of the IEEE STC Newsletter, which includes two Workshop reports, two technical contributions and an interview. 

First, the chairs of the ACM GreenMetrics 2016 workshop summarize some of this year's highlights, and Dr. Athanasia Panousopoulou (Institute of Computer Science of the Foundation for Research and Technology - Hellas) reports on the CysWater 2016 Workshop.

Dr. Francesco Musumeci (Dept. of Electronics, Information and Bioengineering, Politecnico di Milano, Italy) discusses green approaches for the management of inter-datacenter networking.
Moreover, Dr. Giorgio Corani and Mauro Scannagatta (Dalle Molle Institute for Artificial Intelligence, Lugano, Switzerland) describe a classification approach for statistical air pollution prediction.


Finally, we feature the interview to prof. Kameshwar Poolla (Electrical Engineering & Computer Sciences, Mechanical Engineering, University of California, Berkeley), a prominent researcher in the area of energy business models.


As usual, the newsletter closes with the list of upcoming conferences and workshops in the field of sustainable computing.


Report from the ACM GreenMetrics workshop

By Niklas Carlsson (Linköping University, Sweden), Zhenhua Liu (Stony Brook University, US), Thu Nguyen, (Rutgers University, US), Catherine Rosenberg (University of Waterloo, Canada), and Adam Wierman (Caltech, US)

The seventh annual GreenMetrics Workshop was held on June 14, 2016 in Antibes Juan-les-Pins, France, in conjunction with the ACM SIGMETRICS/IFIP Performance 2016 conference. Topics of interest fall broadly into three main areas: designing sustainable ICT, ICT for sustainability, and building a smarter, more sustainable electricity grid. The workshop brought together researchers from the traditional SIGMETRICS and Performance communities with researchers and practitioners in the three areas above, to exchange technical ideas and experiences on issues related to sustainability and ICT.

The workshop program included three 45-min keynote talks, and nine 20-min presentations of technical papers. In the first keynote ``The New Sharing Economy for the Grid2050'', Kameshwar Poolla from UC Berkeley discussed three sharing economy opportunities in the electricity sector - sharing storage, sharing PV generation, and sharing recruited demand flexibility. He also discussed regulatory and technical challenges to these opportunities. In addition, he presented a micro-economic analysis of decisions by firms, and quantify the benefits of sharing to various participants.

Xue (Steve) Liu from McGill University presented the second keynote talk, titled ``When Bits Meet Joules: A View from Data Center Operations' Perspective''. He used data centers as an example to illustrate the importance of the co-design of information technologies and new energy technologies. Specifically, he focused on how to design cost-saving power management strategies for Internet data center operations.


Our third keynote talk was by Florian Dörfler from ETH Zurich, titled ``Virtual Inertia Emulation and Placement in Power Grids''. He presented a comprehensive analysis to address the optimal inertia placement problem, in particular, by providing a set of closed-form global optimality results for particular problem instances as well as a computational approach resulting in locally optimal solutions. He illustrated the results with a three-region power grid case study.

GreenMetrics 2016 Best Student Paper Award Ceremony. 
In photo: Catherine Rosenberg (award chair) and Navid Azizan Ruhi (winner of the award). 

The best student paper award was given to ``Opportunities for Price Manipulation by Aggregators in Electricity Markets'' by Ruhi et al. (See picture above from the award ceremony.) The award was determined by a committee of the invited speakers, chaired by Catherine Rosenberg, after considering both the papers and the presentations of the candidates. The authors quantified the profit an aggregator can obtain through strategic curtailment of generation in an electricity market. Efficient algorithms were shown to exist when the topology of the network is radial (acyclic). Further, significant increases in profit can be obtained through strategic curtailment in practical settings.

The full program and more details can be found at http://www.sigmetrics.org/greenmetrics/ 

Overall, the papers presented at the workshop reflected a current concern of energy consumption associated with proliferating data centers, and other fundamental issues in green computing. The workshop incited interesting discussions and exchange among participants from North America, Europe, and Asia. Next year’s workshop will take place at University of Illinois Urbana-Champaign (UIUC). Information about that edition will be posted on the workshop website.


Report on the 2nd International Workshop on Cyber-Physical Systems for Smart Water Networks

Dr Athanasia Panousopoulou, Dr Grigorios Tsagkatakis, Institute of Computer Science (ICS), Foundation for Research and Technology - Hellas (FORTH), Greece, Prof Panagiotis Tsakalides, Department of Computer Science, University of Crete, and Institute of Computer Science (ICS), Foundation for Research and Technology - Hellas (FORTH), Greece, Prof Baltasar Beferull Lozano, Department of Information and Communication Technology, University of Adger, Norway.


The first two decades of the 21st century have witnessed systematic efforts on transiting from sporadic ICT solutions for water management towards a framework of sustainable solutions on water resources utilization. This increasing interest in world-wide water management comes at no surprise; in 2015 the water crisis was ranked as the #1 global risk based on its societal impact as a measure of devastation.

Smart Water Networks (SWN) have in turn emerged as the engineering field that addresses the blend of networked data technologies with water infrastructures. By definition, SWN have an inherited dependence on Cyber-Physical Systems (CPS), since the latter provides the technological suite to deliver responsible, scalable, and secure architectures in dynamic environments. Nevertheless, numerous technical and research challenges must be jointly addressed. New design exemplars on smart sensors should be provided, tailored to the needs of the aging, harsh water infrastructure. Front-end raw sensing should be elevated to in-network edge-processing, which is capable of adapting to the spatial and temporal dynamics of underground and underwater environments. Distributed intelligence and control should be combined with cloud computing architectures to efficiently address both the optimized storage of the massive volumes of information generated, as well as the necessity of timely failure detection and isolation. Finally, novel paradigms on security and safety against intentional and unintentional contamination attacks should be embedded to the network backbone of a SWN architecture.

The discussion thus far highlights that the success of SWN in becoming the sustainable technology for overcoming water crisis depends on bridging the gap between researchers and engineers from the CPS community and practitioners from the Water Industry, in order to both share their experiences, as well as formulate novel CPS paradigms for fulfilling the vision of SWN. This has been the motivation for pin-coining in 2015 the International Workshop on Cyber-Physical Systems for Smart Water Networks (CySWater) as part of the Cyber-physical Week in Seattle, US. The 2016 edition of CySWater was held in conjunction with the CPS Week 2016 in Vienna, Austria, as the application-driven forum, where CPS modeling, analysis, deployment, and evaluation are tailored to the specific needs of different aspects of the water life cycle.

The theme of the workshop was oriented at control, signal processing, security, and experimentation of designing and evaluating the efficiency of CPS in the Water Management arena. The technical program of CySWater 2016 (available at http://www.ics.forth.gr/spl/cyswater2016/program.html) consisted of a keynote, 8 full research papers, which have received full indexing by the IEEE Digital Library (available at: http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=7468874), and a panel discussion for drawing the workshop’s concluding remarks.

The highlight of the event was the keynote given by Dr. Christopher Harman, research manager at the Norwegian Institute for Water Research (NIVA), Norway, on environment diagnostics. Dr. Harman explained the impact of the climate change on the sustainability of water ecosystems. The keynote highlighted the perspective of the water practitioner in terms of how new techniques for sampling, recording, and analyzing water resources could help in elevating dull symptoms monitoring to automated diagnosis of dangerous contaminants in water and the respective course of treatment. The link between wastewater treatment and community health was also addressed, highlighting the trend of using collective sampling of human biomarkers in sewage to detect the response of community to external factors, such as the impact of air pollution to negative health reactions.

CysWater2016 was characterized as a highly interactive event, since both the keynote as well as the technical presentations stimulated fruitful discussions between the delegates. The efficacy of game theory, compressive sensing, and machine learning in securing the critical information, facilitating edge processing, and optimizing the performance of water distribution over aging infrastructures, have been identified as promising approaches for enabling the vision of Smart Water Networks. Nevertheless, the need for standardized CPS practices in the water management arena that would dramatically improve the design of scalable solutions, the adoption of security-by-design approaches in the architecture of new water infrastructures, and the establishment of systematic efforts for collaborations between computer scientists, control engineers, and water engineers have been recognized as the key future SWN challenges.

We are grateful to the CPS Week 2016 organizers, IEEE, as well as the workshop’s delegates for making CySWater2016 a successful event. Motivated by the acclaimed necessity of bringing together the CPS research community and the Water industry, the 2017 edition of CySWater is planned to be hosted at Pittsburg, US, while expanding the call towards water management standardization and policy making.


1^ http://www.ics.forth.gr/spl/cyswater2016/program.html


How green is the Cloud? Problems and solutions

Dr. Francesco Musumeci, Researcher Dept. Of Electronics, Information and Bioengineering, Politecnico di Milano (Italy)


Recent advances in cloud-based services and data-center (DC) infrastructures have led to high growth in the traffic managed by DCs and transported through the network. Correspondingly, all the involved network equipment consumes relevant amount of energy, which has become one main concern for DC and network operators, especially considering the consequent high emission of Green-House Gases (GHG) and CO2 volumes.

According to Greenpeace estimations [1], if cloud computing were a country, it would be the 5th largest electricity consumer in the world. This explains why efficient solutions must be adopted to make the cloud energy-friendly and environmentally sustainable i.e., in one word, “green”. A first step towards greening the cloud consists of exploiting Renewable Energy Sources (RES), such as solar or wind, for DCs power supply. In the last years, many players in the cloud industry have driven into this direction, e.g., Google has announced that more than 35% of its DCs operations are powered by RES [2], and this number is around 25% for the case of Amazon Web Service [3].

However, one drawback of RES is that their availability is not stable in time, as can be observed in Figure 1, where we show the distribution of solar and wind power along a day, based on data from [4] and [5], respectively. Due to this variability, distributed cloud systems exploiting RES are also known as “follow-the-wind/follow-the-sun” architectures.





Figure 1: Typical daily variation of solar [4] and wind [5] power.


Note that, in general, improving the ICT sustainability, i.e., reducing its carbon footprint, is not equivalent to reducing its energy consumption. In fact, exploiting RES may help in reducing CO2 emissions, but usually requires higher overall energy consumption within the network, since RES are typically available in remote locations, far from where the electricity is really needed. Thus, given that transporting energy is way less efficient than transporting information, “green DCs” are usually deployed far from their users, therefore additional energy is required to reach these DCs.

The trade-off between the “DC energy”, i.e., the energy consumed to perform data processing, storage and management within a DC, and the “transport energy”, i.e., the energy spent by the network infrastructure to carry information from the DC to the end users and vice-versa, has been investigated by several researchers, who have provided general solutions to reduce the overall network energy consumption. As a sample study, here I would like to mention the work in [6], which not only deals with the trade-off between DC- and transport-energy, but specifically focuses on CO2 emission reduction, by distinguishing between “green” and “brown” energy, the latter being energy consumed by producing CO2 emissions. The authors of [6] have proposed algorithms for the sustainable routing of online users requests for DC services, and compared them with two benchmark solutions: 1) shortest-path routing, and 2) best green DC, i.e., an approach where each user request is satisfied by the DC with current highest RES availability. The authors have shown that their proposed algorithms, properly taking into account the current network occupancy, RES availability over time and the physical distance between users and DCs, are able to substantially reduce CO2 emissions, with very low impact on the service availability (i.e., service blocking probability).

The second aspect here discussed, providing additional advantages in emissions reduction, which can be combined with the potential of using RES, is represented by virtualization and in particular by Virtual Machine (VM) migration. This technique is adopted for several purposes, such as cloud bursting, service redundancy, load balancing as well as to improve users quality of experience. In our scenario, VM migration can be exploited to greening the cloud, e.g., by transferring VMs (and so, the services they support) to physical servers running in RES-powered DCs. An example is shown in Figure 2 for the migration of VMs from brown DCs to green DCs across a USA-like network.





Figure 2: Example of VMs migration from brown to green, i.e., RES-powered DCs, adapted from [7].


Also in this context, a trade off arises between the emission reduction provided by utilizing the green DC instead of the brown DC, and the additional (brown) energy required to perform VM migration. This trade-off has been investigated, e.g., in [7], evaluating the effectiveness of different VM migration techniques with respect to standard network operation without VM migration. It has been shown that relevant emission reduction (up to 50% in the scenarios analysed by the authors) is obtained by properly performing VM migration, i.e., taking into account network/users information such as traffic prediction, RES availability and transport network consumption.

In conclusion, it is noteworthy that the use of renewable energy, especially if combined with virtualization techniques and VM migration, is a promising candidate to allow cloud networks sustainability, but care must be taken when designing and operating the network, since being able to exploit RES usually comes at the cost of spending higher energy (either renewable or non-renewable) volumes.

References
  1. G. Cook, “How Clean Is Your Cloud”, Greenpeace white paper, 2012.
  2. www.google.com/about/datacenters/renewable.
  3. aws.amazon.com/it/about-aws/sustainable-energy.
  4. X. Dong, T. El-Gorashi, and J. Elmirghani, “IP over WDM networks employing renewable energy sources,” IEEE/OSA J. of Lightwave Technology, vol. 29, n. 1, pp. 3-14, Jan. 2011.
  5. Y.-H. Wan, “Wind power plant behaviors: Analyses of long-term wind power data,”  National Renewable Energy Laboratory (NREL) Technical Report, Sep. 2004.
  6. M. Gattulli, M. Tornatore, R. Fiandra, A. Pattavina, “Low-Emissions Routing for Cloud Computing in IP-over-WDM Networks with Data Centers”, IEEE J. on Selected Areas in Communications, vol. 32, n. 1, pp. 28-38, Jan. 2014.
  7. U. Mandal, M. Habib, S. Zhang, B. Mukherjee, and M. Tornatore, “Greening the Cloud Using Renewable-Energy-Aware Service Migration,” IEEE Network, vol. 27, n. 6, pp. 36-43, Nov. 2013.

Air prediction via multi-label classification

Giorgio Corani and Mauro Scanagatta, IDSIA – Istituto dalle Molle di Studi sull’Intelligenza Artificiale Manno (Switzerland)


Statistical air pollution prediction is an important task in environmental modeling. The pollutants most commonly studied are ozone (Schlink et al., 2003) and particulate matter. Consider for instance the problem of ozone prediction. The decision maker typically needs to know the probability of ozone overcoming a threshold deemed relevant for health. The task is then to estimate the probability of ozone exceeding the threshold on the basis of different features, typically constituted by past values of meteorological variables and air pollutants. This is a classification problem. Bayesian networks (Koller and Friedman, 2009) are probabilistic models suitable for classification.

Typically the decision maker requires prediction regarding multiple variables such as ozone measured in multiple stations, assessed according to different indicators (1-hour maximum value and 8-hours moving average) and over different days (e.g., today and tomorrow). There are therefore multiple correlated variables to be predicted. We propose to setup a model able to jointly predict all such variables accounting for their dependencies. Multilabel classification is the machine learning area which studies how to jointly predict multiple dependent class variables (labels). We devise a multilabel classifier based on Bayesian networks to simultaneously predict multiple air pollution variables. This is the first application of multilabel classification in environmental modeling, as far as we know. We compare the multilabel classifier against the traditional approach of devising an independent model for each variable to be predicted.

We consider three case studies: (i) prediction of PM2.5 in eight stations in Shanghai for today and tomorrow (16 variables to be predicted); (ii) prediction of ozone in Berlin for today and tomorrow, considering the threshold for both 1-hour and 8-hours concentration (4 variables to be predicted); (iii) prediction of ozone in Burgas (Bulgaria) for today and tomorrow, considering the threshold for both 1-hour and 8- hours concentration (4 variables to be predicted).

In each case study the multilabel classifier consistently outperforms the independent approach, provides significantly higher accuracy of the predictions. In this way it provides better support for the decision maker. The forecast of the models have been thoroughly assessed from different viewpoints; for details see Corani and Scanagatta (2016). The application of multilabel classifiers in environmental modeling is not limited to air pollution. Instead, it is suitable to the many applications in which it is required to predicting multiple dependent discrete variables. For instance multilabel classification could become an important tool for ecological modeling, being able to simultaneously predict the presence/absence of different species accounting for pray-predators relations.



Figure 1: The Bayesian network model learned on the Berlin data set. The white nodes indicate variables exploited to compute the predictions (features). The grey nodes indicate the variables to be predicted. The arrows denote stochastic dependence among variables. The model computes the prediction on all the four grey variables accounting for their dependencies.


References

U. Schlink et al, A rigorous inter-comparison of ground-level ozone predictions, Atmospheric Environment 37 (23) (2003) 3237 – 3253.

D. Koller, N. Friedman, Probabilistic Graphical Models: Principles and Tech- niques, MIT Press, 2009.

G. Corani, M. Scanagatta, Air pollution prediction via multi-label classification, Environmental Modelling & Software, Volume 80, June 2016, Pages 259–264


Community Highlight: Interview with Prof Kameshwar Poola


In this feature we ask a prominent researcher in the field of sustainable computing to share their journey and lessons along the way with the broader community. In this issue we have the privilege to sit down with Kameshwar Poolla, known for his research on Economics and Engineering of Future Energy Systems.


      Name: Kameshwar Poolla
      Current position   Cadence Distinguished Professor, Electrical Engineering & Computer Sciences, 
Mechanical  Engineering,  University of California, Berkeley 
       Past affiliations: University of Illinois, Urbana-Champaign
       Alumni: Indian Institute of Technology, Bombay
        Currently working on: Sharing Economy business models for the Smart Grid





Favorite memory as a student/advisor/researcher:  As a student, attending inspiring seminars by Tom Kailath, Petar Kokotovic, Pravin Varaiya, and so many others. Once I attended an awesome talk on stellar evolution by Hans Bethe. As an advisor, discovering that my students are much smarter than me. As a researcher, when my mentor Pravin Varaiya said “good job” after I proved a nice result.

Could you share a research contribution from your research, and explain why this is something that you are particularly proud?

Broadly, I work at the intersection of electricity and markets. I am interested in how intermittent renewable energy can be sold in existing and new markets, and how to mitigate the integration costs that arise from intermittency of renewables. I started working on these topics 5 years ago. In 2012, my excellent student EIlyan Bitar (now a Professor at Cornell) and I worked selling random renewable energy. Most energy is sold in day-ahead wholesale markets. If you are selling wind energy, this is a problem because forecast of your production are uncertain by 30%. It is like selling apples, but you don’t know how many apples you have. Eilyan and I cast this as a classical Newsvendor problem and explored how wind producers might participate in existing markets, the marginal value of a small amount of storage, and how recourse markets 4 hours before delivery can greatly increase the quantity of wind sold in wholesale markets.

Explain one thing that makes your work exciting for you? 


I am really excited by learning new things. Our research is at the confluence of game theory, optimization, microeconomics, and technology. And we hope in a small way, to influence energy policy through our research. We hope to enable reductions in greenhouse gas emissions and support our collective sustainable energy future.

What do you think is (are) the important problem(s) to be solved in the next 10 years within sustainable computing?

David Culler articulates the need for energy proportional computing. This is very important at the level of a single processor, multiple processors, and distributing computing platforms. Being able to reduce the energy footprint of information technologies through job scheduling, geographic load balancing and integrated cooling system management is very important. Any solutions proposed must make sound economic sense, and this constraint is hugely challenging. The electricity used in the IT sector is only 2-3%, but growing fast. Everyone must help in reducing CO2 emissions, and sustainable computing is part of the solution.

What courses and skills are most important for students wanting to work in this area?

I think students working on sustainable computing, or sustainability in general, must be intimately familiar with optimization, random processes, and be able to implement control/scheduling algorithms to test their research ideas. So courses in these subjects combined with software skills, and a burning ambition to create new technologies make for certain success.


More interviews can be found here.

Upcoming Events

Short Name

Main Topic

Location

Dates

Paper Due

Notification

ICGCTI 2016


Green computing and technology

Kuala Lumpur, Malaysia


Sep 6, 2016 - Sep 8, 2016

Aug 6, 2016

Aug 16, 2016

SmartGreens 2017

Smart cities and green ICT systems

Porto, Portugal

23-25 April 2017

Nov 23, 2016

Jan 25, 2017

ICGCTI 2017

Green computing and technology

Osaka, Japan

30-31 March 2017

Aug 23, 2016

Sept 28, 2016

ICSCC 2017 Boston

Sustainable Computing and Communications

Boston, USA

24-25 April 2017

Aug 4, 2016

Nov 4, 2016

ICSCC 2017 San Francisco

Sustainable Computing and Communications

San Francisco, USA

7-8 June 2017

Oct 31, 2016

Nov 21, 2016


Journal and Special Issue Call For Papers


Journal Papers                                                                     Due
Sustainable Computing                                                     (Open)
IEEE Transactions on Sustainable Computing                 (Open) 

Share It!