[ECSS 2024] Presentations of the "Green ICT and ICT for Green" Workshop

Workshop Chair: •Marco Aiello, University of Stuttgart (Germany) Supported by IE's Green ICT Working Group 20th European Informatics Leaders Summit (ECSS 2024) “Green ICT and ICT for Green” Workshop 29 October 2024 On Sustainability metrics and measures Rafael Capilla Rey Juan Carlos University of Madrid (Spain) Informatics Europe Meeting, Malta, October 2024 Rafael’s research •First publication in 2013 about sustainable design decisions •Moving after to sustainability metrics in software architecture (i.e. architecture decay) 2015-todaySystems endurance •Sustainability in education, skills demanded: 2021-today •Understand sustainability as a high-level cross-cutting quality attribute affecting the development of systems •Without metrics and KPIs we cannot estimate how sustainable our systems are •11 papers + 1 special issue published since 2013 I started around 2012 working on sustainability topics… 2 Informatics Europe Meeting, Malta, October 2024 Sustainability dimensions •Environmental sustainability: Good set of metrics to estimate Green ICT/IT or sustainable ICT in various domains (e.g. climate, transportation, energy, etc) •Technical/Software sustainability: A significant number of metrics to measure sustainable aspects of software systems (develop sustainable systems or use software to evaluate sustainability aspects). Lack of a global sustainability indicator •Social/Individual: Less explored and hard to evaluate the impact of sustainability initiatives in society and individuals •Economic: Cost factors evaluating sustainability actions 3 Informatics Europe Meeting, Malta, October 2024 Software sustainability metrics 4 Informatics Europe Meeting, Malta, October 2024 •More than 40 code metrics to evaluate different quality aspects of code bases •Lack of a clear sustainability indicator (e.g. the case of maintainability) and how to combine existing metrics (e.g. complexity, coupling, instability, etc) •Many software metrics related to estimate technical debt can be used •Time/Effort to pay the debt •Detection of quality loss •TD ratings assess on quality defects Software sustainability metrics 5 Informatics Europe Meeting, Malta, October 2024 •Architecture metrics can help to evaluate sustainabilit y from a different angle Environment sustainability indicators 6 Informatics Europe Meeting, Malta, October 2024 •There are plenty of metrics to estimate the green aspects of systems •Several domains can benefit of estimating environmental sustainability and green ICT (e.g. waste and resource mgmt., transportation, energy, manufacturing, and many more) •We need to standardize these metrics and its correct use provided by annual sustainability reports •It is key to define good KPIs used by the metrics as indicators to be measure (e.g. context data, resource usage) •Some rankings use environment indicators to rank most sustainable organizations Environment sustainability indicators 7 Informatics Europe Meeting, Malta, October 2024 •Annual sustainability reports from companies •We studied 14 international companies using their sustainability reports and web sites as baseline •Uncover metrics used, ESGs, SDGs, and sustainability dimensions covered Environment sustainability indicators: CGI example 8 Informatics Europe Meeting, Malta, October 2024 •Reporting sustainability impact and coarse numbers •Carbon-emission, e-waste, resources usage •Impact of social dimension •Develop futureready skills Introduction The Growing Energy Demands of HPC and Quantum Computing •High-Performance Computing (HPC) drives advancements in scientific research, weather forecasting, financial modelling, and more. •Quantum computing promises to revolutionize fields like medicine, materials science, and artificial intelligence. •The Problem: Both HPC and quantum computing are energy-intensive, contributing to rising carbon emissions and environmental concerns. •Q: Are Sustainability and Competitive Advantage mutually exclusive? Environmental Impacts of Traditional HPC The Carbon Footprint of Classic Supercomputers Main Issues •Massive energy consumption for processing and cooling largescale HPC systems. •Significant e-waste generation due to the rapid obsolescence of hardware components. •Resource depletion and environmental impact associated with manufacturing specialized chips and hardware. Quantum Computing: A New Set of Challenges Unique Sustainability Concerns in the Quantum Realm •Extreme cooling requirements for superconducting qubits, often requiring temperatures near absolute zero (-273.15°C). •Energy-intensive fabrication processes for quantum processors, involving rare and energy-demanding materials. •Potential for increased e-waste as quantum technology rapidly evolves and hardware becomes obsolete. Counter •Q: Does quantum computing power consumption scale directly with the number of qubits? Real-World Case Study: Bad Practice Hypothetical Scenario: An Unsustainable HPC Center A fictional university research lab with an outdated HPC cluster: •Inefficient Cooling: Reliance on outdated, energy-intensive cooling systems. •E-waste Neglect: No program for responsible recycling or disposal of old hardware. •Lack of Optimization: No strategy for energy-efficient hardware upgrades or software optimization. Real-World Case Study: Good Practice Leading the Way: Sustainable HPC at ETH Zurich •ETH Zurich's approach to green HPC: •Renewable Energy: Utilizing renewable energy sources (hydropower, solar) to power their data centre. •Efficient Cooling: Implementing free cooling systems that use outside air and optimizing airflow within the data centre. •Hardware and Software Optimization: Prioritizing energy-efficient hardware and software to reduce energy consumption. EU Policies and Guidelines EU Initiatives for Green HPC and Quantum Computing Current Policies •European Green Deal: Overarching framework for climate neutrality by 2050, with implications for ICT and research infrastructure. •Energy Efficiency Directive (EED): Sets energy efficiency targets for data centres and promotes the use of energy-efficient technologies. •Ecodesign Directive: Establishes minimum environmental performance standards for ICT equipment, including servers and storage devices. •EuroHPC Joint Undertaking: Promotes the development of energy-efficient supercomputing infrastructure across Europe. Challenges for Education and Research Sustainability in Academia What are we up against? •Budget Constraints: Limited funding for investing in energyefficient infrastructure and sustainable practices. •Awareness and Training: Need for increased awareness and training on green ICT principles among researchers and students. •Balancing Needs: Balancing the increasing demand for computational resources with environmental responsibility. Conclusion and Call to Action Towards a Greener Future for HPC and Quantum Computing Go Green: •Addressing the environmental impact of advanced computing is crucial for a sustainable future. •Research and development of energy-efficient HPC technologies including quantum are essential. •Collaboration between academia, industry, and policymakers is needed to promote sustainable practices. •The change is a complex and nonlinear process that requires sustained effort, collaboration, and perseverance to overcome various barriers and challenges. SDU Software Engineering sdu.dk #sdudk 3 0 / Towards Sustainable Software Engineering Mikkel Baun Kjærgaard Professor, Vice-Head of Research, Head of Software Educations 3 0 / sdu.dk #sdudk SDU Software Engineering Sustainability vs. Software Engineering Software engineering is concerned with developing and maintaining software systems that behave reliably and efficiently, are affordable to develop and maintain, and satisfy all the requirements that customers have defined for them. [ACM] Provide functionality for the customers that provide value but also address quality attributes: • Sustainable Resource Use • Energy Efficiency • … Energy Efficiency Sustainable Resource Use sdu.dk #sdudk SDU Software Engineering Developer Tooling and Training (2019-) sdu.dk #sdudk SDU Software Engineering Heredia, J., Kirschner, R. J., Schlette, C., Abdolshah, S., Haddadin, S., & Kjærgaard, M. B. (2023). Labelling Lightweight Robot Energy Consumption: A Mechatronics-Based Benchmarking Metric Set. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (pp. 1789-1796). IEEE. sdu.dk #sdudk SDU Software Engineering Model of the Energy Consumption for Collaborative Robot A hybrid model based on a data-driven and process-driven methodology. Data-Driven Energy Estimation of Individual Instructions in User-Defined Robot Programs for Collaborative Robots, Heredia, J., Schlette, C. & Kjaergaard, M. B., Oct 2021, In: IEEE Robotics and Automation Letters. 6, 4, p. 6836-6843 !"#$"% &!"#"% !"#$!%&'()*+$,-./-++*/-. &!"#"% ;99)&:%$&.'#+./+3'214*+812G&:$&.'+7.G2)# !"#"$%&"''())("*(*+ !"#"$%,-".#/0)1""$(*+ ,-2(*(*+%2*3%4*0-+5% 67$('(82$("* !"#$"% &!"#"% !"#$!%&'()*+$,-./-++*/-. !""#$#%&'()*+,-'#&. 8!"#$%&&'()')*%'%+,%- 8!"#$%&&'#.')*%'$/#0+- 81&)(2/3&*'('$#450)3.,'$#.)3.00486%�"$%'(7(3/(23/3)89'%.%",89'"%:0%&)& !"#$"% &!"#"% !"#$!%&'()*+$,-./-++*/-. )/011%&.%2 52,%6 -#7%"6 +7+"%4%336 E,'('#!"#24#)1+1F ;"4&$F G%1)#51.1&$!&0F 8!9(:3+;%<7+''+;%(23( =#4>.24%+"666 <'(3+;$<11*,*('#$:)0?5$F3,8*#(,#?3($/0)?#*0'9\$F$G(")$:"9( 70+8"55"=M$7O$AOM$V9="'06*,M$FO$%$<'30#8M$7OM$QRQYM$H30,((5*'+9$01$#8($&<<<]F:7$ SX#8$&'#(3'"#*0'")$:0'1(3(',($0'$`#*)*#;$"'5$:)0?5$:0=>?#*'+O$F:7 !"#$"% &!"#"% !"#$!%&'()*+$,-./-++*/-. E<2+F.%G+$.+K12%$&'4+LI9%:$ !"#"$%&'( )*"+,$ <%228&!+4# H8!.)!&0 A'"'1($,# ='+,%)" -,.&$/01+# <8((!$8.82# )'"!0&)# ')8$1+%( ')8$1+!%& I'1$,!&0 +%%." 1&)# 21+'(!1." 2+,.#/%3( 245$&/ J%.!$!'" 1&)# $8"+%2'( )'21&) K&)8"+(!1.# @%.8+!%&" SDU Software Engineering sdu.dk #sdudk 3 0 / Thank you for your attention Read more here: https://www.sdu.dk/en/forskning/sdusoftwareengineering https://www.sdu.dk/en/forskning/cis Contact info: Mikkel Baun Kjærgaard, [email protected].dk Thanks to all the people who contributed to the work presented 3 0 / Monica VITALI monica.[email protected] SADP Sustainable Application Design for Greener Applications in the Cloud October 29, 2024 WHO AM I? Associate Professor at Politecnico di Milano in Italy I research new strategies to improve the efficiency of data centers and clouds by applying techniques derived from the Artificial Intelligence and Machine Learning fields. I am very interested in adaptation and self-adaptation to discover how a complex system can heal itself when some problems occur. 2 CLOUD NATIVE APPLICATIONS Alternative implementations of the same functionality can be provided 9 CLOUD NATIVE APPLICATIONS Microservices are black boxes for the infrastructure provider All these features are hidden and don’t adapt with the context of execution Time sensitive Security sensitive Not time sensitive 10 SADP - Sustainable Application Design Process SUSTAINABILITY AWARENESS Microservice annotation with computational requirements, QoS constraints, and power consumption metadata MICROSERVICE ENRICHMENT Designers provide different execution modalities for the microservices composing the application MICROSERVICE CLASSIFICATION Application components are annotated with their relevance for the overall process 03 01 02 1 2 3 11 SADP 1 - Sustainability Awareness 12 SADP 1 - Sustainability Awareness 1 1 13 SADP 2 - Microservice Classification 2 2 1 1 14 SADP 3 - Microservice Enrichment 3 2 2 1 1 15 SADP - Supporting green applications design https://github.com/valentinbootz/a pp-cloud-native-sustainability 16 DESIGN IS ONLY A PIECE OF THE PUZZLE INFRASTRUCTURE PROVIDER Makes energy mix composition information available to service providers Manages application components exploiting the application enriched design SERVICE PROVIDER Enriches application design with sustainable features Gets insights on application sustainability and sets sustainability targets SERVICE CONSUMER Empowered sustainability awareness through indicators and certifications SUSTAINABILITY QUALITY of SERVICE 17 CONTEXT-AWARE DEPLOYMENT - 20% energy INFRASTRUCTURE PROVIDER 18 ▪2.8–3.8% of total EU electricity use for data centers [1]. ▪Cloud-native applications consist of multiple layers of technologies, services, and platforms using these data centres. ▪Identifying choices that impact energy efficiency is challenging. How can we evaluate the energy efficiency of continuously developing cloud-native applications and their platforms? Energy Efficiency of Cloud-Native Applications [1] Kamiya, G. and Bertoldi, P., Energy Consumption in Data Centres and Broadband Communication Networks in the EU, Publications Off of the European Union, Luxembourg, 2024, doi:10.2760/706491, JRC135926. Physical Machine Physical Machine Physical Machine Physical Machine Physical Machine Execution Platform Infrastructure Services VM Container OS / Hardware Service Runtime Framework Physical Machine Execution Platform Infrastructure Services VM Container OS / Hardware Service Runtime Framework Application Service Layer Platform Layer Isolation Layer Application Layer … Architectural Functional Configurations Technological Platform Service Choices What do we do in cases of evolving requirements and a large solution space? ▪Prototyping ▪Experimentation ▪Software Quality Management 5 Evaluating Energy Efficiency of Applications Code Test Deploy 6 Clue In support of this, we build Clue, a tool to continuously evaluate the energy-efficiency of changes in software development ▪Git-oriented experiments ▪CI/CD pipeline compatible ▪Can be used to experiment/prototype define workload model build and commit prototype runs experiments generates reports evaluate prototype ▪System Qualities ▪Energy Consumption* ▪Resource Utilization ▪Carbon Intensity 7 What do we evaluate (so far)? * We use existing energy meters, e.g., Kepler, that use estimation and sampling. How does clue work? 8 ▪Need to define an experiment ▪Requires IaC-capable deployments to a staging/dev environment ▪Relies on Prometheus for collecting measurements 1. Tee Store[2] 2. On-Prem bare-metal Kubernetes cluster, with socket meters (for inner validation) 3. 4x Workload profiles [Fixed, Backoff, Stress, Shape] 9 Seeming Clue in action. Monolith Prototype Serverless Service Prototype Runtime Replacement Prototype Service Reduction Prototype Service Layer Platform Layer Application Layer [2] J. von Kistowski et. al. , "TeaStore: A Micro-Service Reference Application for Benchmarking, Modeling and Resource Management Research," 2018 MASCOTS, doi: 10.1109/MASCOTS.2018.00030. 10 System Quality Branch Latency p95 [s] Failure Rate [%] Costs Projected Total [$] Projected Consumed [$] Per Request [¢/1000] Baseline 0.17 - 16.37 3.5 - 11.51 0.58 - 0.84 0.27 - 0.41 24.01 - 0.26 Runtime Replacement 0.10 - 12.42 2.3 - 0.03 0.58 - 0.82 0.27 - 0.40 23.11 - 0.10 Monolith Architecture 0.04 - 42.78 0.89 - 41.80 0.16 - 0.26 0.08 - 0.11 10.10 - 0.77 Service Reduction 0.20 - 8.36 1.9 - 1.78 0.69 - 0.86 0.28 - 0.41 24.98 - 0.10 Serverless 1.76 - 15.38 5.1 - 9.31 4.08 - 4.60 0.67 - 0.94 63.49 - 0.53 Comparing Pulsing and Stress Workloads 11 System Quality Branch Latency p95 [s] Failure Rate [%] Costs Projected Total [$] Projected Consumed [$] Per Request [¢/1000] Baseline 0.17 - 16.37 3.5 - 11.51 0.58 - 0.84 0.27 - 0.41 24.01 - 0.26 Runtime Replacement 0.10 - 12.42 2.3 - 0.03 0.58 - 0.82 0.27 - 0.40 23.11 - 0.10 Monolith Architecture 0.04 - 42.78 0.89 - 41.80 0.16 - 0.26 0.08 - 0.11 10.10 - 0.77 Service Reduction 0.20 - 8.36 1.9 - 1.78 0.69 - 0.86 0.28 - 0.41 24.98 - 0.10 Serverless 1.76 - 15.38 5.1 - 9.31 4.08 - 4.60 0.67 - 0.94 63.49 - 0.53 Comparing Pulsing and Stress Workloads Serverless has a high (idle/startup) cost, that also impacts performance and reliability. The Monolith can’t scale enough for the stress workload but has very good idle cost and performance. Runtime Replacement outperforms in all categories.