Carbon footprint reduction of a cloud computing service using a - - PowerPoint PPT Presentation

Carbon footprint reduction of a cloud computing service using a predictive dynamic LCA model

Elsa Maurice, Thomas Dandres, Reza Farrahi Moghaddam, Kim Nguyen,Yves Lemieux, Mohamed Cherriet and Réjean Samson SAM8 Seminar, May 20-21, 2014 elsa.maurice@gmail.com

INTRODUCTION OBJECTIVES METHOD RESULTS DISCUSSION & CONCLUSION

PLAN

Electricity demand around the world INTRODUCTION

5

3

2

1

4

6th largest consumer Source: Greenpeace, April 2014

Canadian project to reduce ICT carbon footprint INTRODUCTION

Green Sustainable Telco Cloud

How to measure in real time the environmental impacts of such cloud computing system?

 Server nodes in Alberta, Ontario and Quebec  Time dependent electricity consumption  Dynamic server-load migrations depending on the emissions attributed to each server node in real time

INTRODUCTION

(1) Energy consumption depends on server workload: Dynamic demand

e(t) Instantaneous power demand related to

CPU/Memory use and α base power demand of servers (when CPU/Memory is not used)

Source: Zuker, R. C., et al. (1984). Blue Gold.

Electricity= E(t) depends on demand (t) and supply (t)

(2) Annual load curve for an electric utility: Dynamic generation

 Servers optimization and emission assessment need a precise monitoring of electricity grid mix variations at each hour of the day

Source: Gebert, S. (2012)

Data center power demand variability Electricity generation variability

Regionalization of the electricity flows (2012-2013) INTRODUCTION

Québec Ontario

Nuclear Hydro Natural gas Wind Coal

Alberta

 Electricity carbon footprint has to consider every energy sources  Approach by electricity grid mixes

Main Objective

• Improve the modeling of electricity flow in life cycle

assessment by taking into account temporal variations

• Compute dynamic electric grid mixes from historic data

Secondary Objectives

• Assess real time GHG emissions of an end-user of a

cloud computing service

• Develop a future oriented dynamic A-LCA method to

improve data center management OBJECTIVES

METHOD – LCA system boundaries

Conventional Attributional LCA approach METHOD – Electricity modeling in attributional LCA

Ontario Québec

A-LCA (static)

Demand Electricity production Alberta

Data collection:  Regionalized grid mix  Annual electricity generation (average)

Limits: temporal aggregation in LCA prevents the modeling of accurate emissions of processes that consumes electricity in an irregularly manner

Dynamic historical Attributional LCA approach METHOD – Electricity modeling in dynamic LCA

• Data collection:

 Regionalized grid mix  Historic data of electricity generation (hour by hour over years)  Real time data of electricity generation

• Limits: Difficult to collect data, hard to implement in a LCA model

Québec = Q(t) Alberta = A(t) Ontario = O(t)

Dynamic LCA = time dependent function

Demand = f(t) Electricity production = g(t)

? ? ?

Sources of data METHOD – Data collection

Real-time and historic data from Ontario Source of data: IESO Power to Ontario Demand (http://reports.ieso.ca/public/GenOutputCapability/) Real-time and historic data from Alberta Source of data: AESO Alberta Electric System Operator

(http://ets.aeso.ca/ets_web/ip/Market/Reports/CSDReportServlet/)

METHOD – Data processing (Ontario and Alberta)

Historic

data

Hours Energy output

Day 1

Hours Energy output

Day 2

Hours Emission kg C02

Per day

(2) Seasonal average

Each daily profile is combined

• ver each season

Season = function (hours) (1) Monthly average:

Each daily profile is combined

• ver one month

Month = function(hours) Grid-mix Carbon footprint

RESULTS – Ontario grid mix and GHG emission factor Electric grid mix variations

50 100 150 200 250 300 350 Nov-2012 Dec-2012 Jan-2013 Feb-2013 Mar-2013 Apr-2013

Daily greenhouse gas emission in Ontario (g CO2-eq./kWh)

Greenhouse gas emission factor gCO2/kWh Max: 298 g CO2-eq./kWh Min: 44 g CO2-eq./kWh 0% 20% 40% 60% 80% 100% Nov-2012 Dec-2012 Jan-2013 Feb-2013 Mar-2013 Apr-2013

percentage electricty per source (%)

Day

Electric grid mix in Ontario at 3pm (november-may)

OTHER Total WIND Total HYDRO Total GAS Total COAL Total NUCLEAR Total

Dynamic A-LCA approach: data collection METHOD – Data collection

Real-time and historic data from Ontario Source of data: IESO Power to Ontario Demand (http://reports.ieso.ca/public/GenOutputCapability/) Real-time and historic data from Alberta Source of data: AESO Alberta Electric System Operator

(http://ets.aeso.ca/ets_web/ip/Market/Reports/CSDReportServlet/)

X No real-time or precise historic data from Quebec Monthly data of Quebec electricity generated from Statistic Canada Extrapolation of Quebec daily electricity generation from neighbours consumption and import-export at interconnections

Dynamic A-LCA future oriented model for Quebec

Quebec electricity mix (t) = Production (t) + Imports (t) - Exports (t)

Quebec imports/exports modeling: METHOD – Quebec electricity modeling

New-Brunswick

(50% Coal, 50% natural gas)

Ontario

(Coal)

New-York

(50% Coal, 50% natural gas)

New-England

(Natural gas)

Quebec

(Hydro)

 Historic data of import/export at interconnections  Marginal electricity sources identified by Ben Amor in its thesis (2006-2008 study) Marginal sources of electricity by region

Dynamic A-LCA future oriented model for Quebec METHOD – Quebec electricity modeling Dynamic A-LCA model

Study of historical electrical trends of Quebec neighbors Calculation of a multi parameters function of net imports: Net import (day, hour) = A×Demand + B×Temperature + C×Price + D×Time + E

• A,B,C, D and E = Constants
• i = Ontario, New-York, New England or New Brunswick
• Net import day j of region i, Net import = Import (province i from quebec) – exportation (province i to quebec)
• Demand day j-1 or predicted
• Temperature day j
• Price day j-1 or predicted

Dynamic A-LCA predictive model

• f regioni

RESULTS Dynamic A-LCA future oriented model for Quebec System of equations

Example : December 2013, interconnection: Quebec  New-England

Total Net Import QC (MWh) =- 0,0790 * Day Ahead_demand (MWh) + 5,6229 * Temperature (°C) + 4,6307 * Day Ahead_Local Marginal Price ($/MWh) - 0,3962 * Hours - 739,8551

 One equation per month  One equation per interconnection (Ontario, New-Brunswick, New-England and New-York)  Good deviation factor: 0,84 < r2 < 0,99 Total Net Import QC x Marginal technology = Environmental impact

RESULTS – Quebec grid mix Implementing imports in dynamic electricity grid mix

98,50% 98,75% 99,00% 99,25% 99,50% 99,75% 100,00% janv-13 Feb-2013 mars-13 Apr-2013 May-2013 June-2013 July-2013 August-2013 September-13 October-13 November 2013 December-2013 Months

Quebec local electricity generation (2013)

Wind Natural gas Nuclear Hydro 88% 90% 92% 94% 96% 98% 100% 1 3 5 7 9 11 13 15 17 19 21 23 Hours

Quebec local electricity generation with net imports (December 2013)

Import-export modeling have inserted a significant part of Coal in the Quebec electricity mix

GHG emissions of electricity consumed in Quebec

DISCUSSION Smart network management

• Optimization of punctual or daily activities such as maintenance,

instant messaging… to reduce environmental impacts

• Optimization of server-load migrations across the Green

Sustainable Telco Cloud … to reduce environmental impacts

Combine Dynamic LCA with consequential LCA

• to take into account marginal sources of electricity of domestic

electricity generation

Keep on collecting data to improve model precision

• Update marginal technologies according to 2014 data

Application

• Carbon tax
• Transparency
• Minimization of the emissions directly by the end-user

CONCLUSION A-LCA Dynamic approach

• The difference between the annual average electric grid mix and the real-

time electric grid mix may be significant

• A better understanding of electricity flows at different scales of time: day,

season, year

• Improve the integration in A-LCA of electricity generation temporal

variations

• Improve inventory in A-LCA. Environmental impacts in Dynamic LCA are

computed with a higher accuracy

• Future oriented models can help to make decisions and to anticipate the

electricity trades between Canada and United States

QUESTIONS

Thank you for your attention!!

AKNOWLEDGEMENTS