The Problem
Web traffic is responsible for a large part of global carbon emissions. A report published by IEA in 2022 showed that data transmission network energy use was at ~340 TWh and data center energy use was at 320 TWh constituting over 3% of the global electricity usage. That number is rapidly increasing and is said to increase to as much as 13% by 2030. The driving forces behind those high numbers are the shifting habits of our internet use at work and daily life and the continual improvements in video streaming and online gaming that cause increasingly large amounts of data transfer and cloud computing time.
Many excellent carbon-aware solutions have been used to reduce the environmental impact of software, however, the global penetration is still very low as the instrumentation of individual apps or cloud infrastructure with carbon-aware solutions often introduces friction in software teams because they have to build, maintain and monitor the carbon-aware features. As a result the large majority of internet traffic still comes from applications where carbon-aware solutions have already reached diminishing returns or apps written before the carbon-aware concept was widely adopted. A solution would be to move all global operations to green data centers, but that is really hard due to pre-existing contractual requirements and lack of location flexibility for private data center setups. Also notably there are very few truly 24/7 “green” data centers yet there are very numerous data centers that are very “green” for a part of the day when the wind is blowing and the sun is shining or people are sleeping.
Solution
We propose a carbon-aware solution that monitors electrical grid data and intelligently directs web traffic to a region where carbon intensity is the lowest at the given time yet not too far away so as to not compromise the latency when it matters. This solution abstracts carbon awareness to the computing edge of the internet which is almost entirely controlled by a single digit of vendors and open source solutions. As such incorporating our solution in just a few key places can affect the internet globally.
CarbonAware SDK
The global DNS weighting is updated continuously using the CarbonAware API from sources like US Energy Information Administration's hour grid monitor.
Impact
The main selling point of our proposal is that by merely incorporating it into a single digit small number of products we can achieve almost a complete penetration of the global market including cloud, on-prem and transmission centers. Notably our solution will not reduce the carbon impact of the transmission center local to the end user and for latency-critical applications like online gaming, and companies using custom DNS solutions, but we estimate that these combined contribute no more than 20% of the total global internet electricity usage. The estimated the drop in carbon emissions in the remaining part of the used electricity by at least 60% and even close to 100% over time as more over time as companies increasingly move to multi data center setups and use auto-scaling.
We confirmed that within each global region like EMEA, Americas or Asia Pacific there is typically at least one data center per cloud provider where the wind is blowing or the sun is shining bringing the carbon intensity to nearly zero.
With our solution we still do not give up on the data centers with typically more carbon emissions. They will still serve the critical part of the latency-critical traffic and they will still fulfill the crucial reliability function. But notably the latency critical traffic rarely involves a meaningful amount of compute and data and when servers and transmission centers are on standby in the more polluting regions. More advanced users already use auto-scaling to completely shut down the unused machines, but even if still on, servers and switches use only a tiny fraction of the power when they run at low capacity.
We also impact the allegedly already green data centers that use energy storage or private green energy sources as when the traffic is redirected away from them, they can then sell back their green energy back to the not fully green local grid.
Future Plans
We hope to use more development time to bring the PoC to the most popular open-source implementations of DNS and LoadBalancers. Once incorporated there we will need more man hours to promote the idea to the biggest DNS and CDN vendors such as Route53 or CloudFlare so that it can be incorporated in their offering and set as the default option. We can also attempt to nudge the main vendors by lobbying the regulators. Specifically, regulators in the EU are typically really supportive of similar ideas and can help ensure the vendors deliver.
This solution can make a significant difference because we propose a low-friction, low-support, and low-integration solution. This is specifically powerful for cloud providers or enterprises. This is because in those settings, generally speaking, DNS and CDN is set up one time in company history and is managed almost fully by the global vendor, and it doesn’t require much human intervention. Once integrated the companies may be able to tune a few settings such as some routes in their service being more latency-critical than others but the large majority would always leave the default settings on.
Sample
Please take a look at a simple example of our solution powered by Constellix DNS and data downloaded from US Energy Information Administration together with code to deploy a mock infra in two AWS regions where the use can be observed:
https://github.com/drastawi/greenhack