Green_Tech_Beyond_the_renewable_energy_integration.pptx.pdf

Transcript

1. Beyond the renewable energy integration Green Tech Sancheeta Kaushal
2. Disclaimer • Backend Engineer with a background in cloud • Run ClimateTechCollective community • Overview and introductory session. More breadth to help you understand the landscape • Just my opinions. Should be applicable to most contexts. Not a 1 size fits all • Alot of it will be obvious. Get obvious done
3. I am a cloud developer…. Why should I care about sustainability ? 2C’s - Cost and Compliance Let’s look at some carbon emission stats from IT Industry!!!
4. IT sector is responsible for ~2% of global carbon emissions. This is same as aviation sector contribution.
5. Simple, all we gotta do is plug in renewable energy. Problem Solved. Right ? • No, a research by Stanford University indicates servers in data centers operate at low levels of utilization, often as low as 12%. • The Uptime Institute found that 30% servers worldwide are unused. This results in a loss of $30 billion in wasted electricity per year worldwide. • A typical server consumes 30-40% of maximum power even when doing no work at all.
6. So, what’s the solution ? • Energy Monitoring & Reporting • Power Down Strategies • Energy Aware Hardware • Green Data Centers • Energy Efficient Algorithms
7. Energy Monitoring & Reporting • Scope 1 ◦ Direct greenhouse gas emissions that originate at the organisation. • Scope 2 ◦ Indirect greenhouse emissions associated with the generation of purchased electricity, heat, or cooling. • Scope 3 ◦ Indirect greenhouse emissions that result from sources not owned or controlled by the organization but are associated with its value chain.
8. Energy Monitoring & Reporting Give Scope 3 nature of emissions, it’s extremely important for cloud engineers to understand the behind the scenes of a cloud so that they can make an informed choice about their service providers.
9. Serverless computing Power Down Strategies Scheduled Shutdowns & Termination Containerization Auto Scaling Server Consolidation Idle Resource Detection Load Balancers Reserved Instance Resource Tagging
10. Energy Aware Hardware • Low power architectures for CPUs/GPUs/TPUs that minimize leakage current and idle power consumption. • Incorporate power gating mechanisms that enable the isolation and shut-off of unused hardware blocks or components when they are not in active use. • Energy Efficient Storage - Solid-state drives (SSDs), minimize power consumption during data access and idle states. • Energy-aware RAM models to anticipate memory access patterns and prefetch data into RAM as needed. This can reduce the frequency of high-energy memory access operations.
11. Energy Aware Hardware • For RAM, IOT devices, Mobile Phones, Wearables where data is frequently erased, energy costs become high. • This is due to Landauer's Principle - A fundamental concept in physics, connecting the thermodynamics and information theory and relates to the minimum amount of energy required to erase one bit of information in a computational process • Reversible Computing minimize energy dissipation by ensuring that every computation is theoretically reversible or can be "undone" without any energy loss. Eg. Qbits and magnetic poles vs voltage levels for 0 and 1 bit today.
12. Energy Aware Hardware • Dynamic Voltage and Frequency Scaling (DVFS) ◦ Power management technique used in computer systems, to optimize energy efficiency and reduce power consumption while maintaining acceptable performance levels ◦ Energy efficiency is achieved by reducing voltage and frequency based on workload and performance requirements ◦ Improves the lifetime of hardware and battery
13. What matters most to computer designers at Google is not speed, but power, low power, because data centers can consume as much electricity as a city. - Eric Schmidt
14. Green Data Centers • Power Usage Effectiveness (PUE) as a critical metric for data center ◦ Facility’s total power delivered divided by its IT equipment power usage, and the lower this figure is, the better. ◦ A PUE rating of 1.0 would be equivalent to a 100 percent efficient facility. ◦ Data centers average about 1.67, which means that for every 1.67 watts of electricity drawn by the facility, only 1 watt is being delivered to IT equipment.
15. Green Data Centers • Air, Hot Water & Liquid Cooling (Oil, Glycol, Flourinert etc. ) • Free Cooling using outside air • Variable Speed Fans • Heat Exchangers Or HVAC (Heat, Ventilation And Cooling) • Energy Efficient Lighting • Hot and Cold Aisle Containment
16. While the amount of computing done in data centers increased by about 550 percent between 2010 and 2018, the amount of energy consumed by data centers only grew by six percent during the same time period. - Google
17. Energy Efficient Algorithms • Time and Space complexity as the measure of algo efficiency ◦ Energy as the new metric • ML training takes huge amount of data and time and so the energy and cost is extremely high ◦ Transfer learning leverages pre-trained models as a starting point for training new models on specific tasks. ◦ Model Compression for running it efficiently ◦ Online learning algorithms updates models incrementally as new data arrives, rather than retraining the entire model from scratch.
18. Energy Efficient Algorithms • Energy efficiency important in IoT for prolonging battery life ◦ Low-Power Communication Protocols like MQTT (Message Queue Telemetry Transport) & CoAP (Constrained Application Protocol) ◦ Sleep Modes and Duty Cycling ◦ Adaptive sampling algorithms adjust the frequency at which sensors take measurements based on the variability and importance of the data.
19. Let’s take a minute and think about what do you want to take back to your work from this presentation?
20. THANK YOU