David Mytton

Sustainable dev environments in the cloud

Published in Cloud, Environment, IT Energy, Product Engineering. Tags: , , , .

Sustainable dev environments in the cloud

A year ago I considered the energy efficiency of application streaming vs running native, local applications. The technology being implemented by the likes of Mighty, Cloudflare, and more hybrid options like Whist, is still very early so I was mostly asking questions.

All those questions remain. Until these companies provide detailed carbon accounting for their products, it’s not possible to make comparisons, only speculate. I also focused primarily on consumer use-cases like web browsing with lots of tabs, or running heavy applications like Figma. With newer technologies like WebAssembly allowing the browser to do even more, this is only going to become more common.

A use-case I didn’t consider was that of software developers offloading workloads to the cloud.

I have recently been setting up a devcontainer for the codebase we’re working on for a new product at Console. This is a significantly better way of defining the runtime dependencies for developers because rather than installing languages and frameworks locally – which might conflict across projects – they can be self-contained and packaged in the same way you would distribute and deploy the application itself. You can then bootstrap the container against Docker running locally. Each branch can have a different instance, it’s easy to reset, and you can keep everything isolated.

With VS Code, the Remote Containers extension allows you to run that container locally on Docker, or on a remote server somewhere. GitHub Codespaces is a productised version of that. Prebuilds mean everything is ready to go when you spin up a new Codespace. GitHub themselves have moved all their development to this model. GitPod is an open source alternative.

Create a new GitHub Codespace. Note how there are only 4 regions and the options are not specific enough to know the precise location and estimate the carbon intensity.

Prior to configuring the devcontainer I was running everything on my M1 Macbook Air. The performance is amazing – tests run in a few seconds and I have the database, frontend and backend APIs, and the Firebase Auth emulator all running locally. I can develop anywhere!

However, one of our team has an older laptop that wasn’t coping well in the summer heat. It’s not even that old – 2018 – but still has all the fans running to deal with the dev environment. In the past, the simple solution would be to buy a new laptop. It was easy to justify buying the top spec because developer time is so valuable, and they need to run VMs and tests and builds with minimal lag. Today, developer time is even more valuable, but the decision is less clear. It may be better to offload that processing to the cloud.

Instead of buying a brand new machine – with all the carbon cost of that purchase, plus all the time wasted setting up a new system – we decided to use it as an opportunity to experiment with GitHub Codespaces. It took me an hour or two to configure all the dependencies correctly, but now we have the entire environment running in the cloud.

Not only have we saved the carbon involved with buying a new laptop, if we do decide to buy one then the spec doesn’t need to be so maxed out. Cheaper in dollars, but also cheaper in resources.

Some remaining questions

It’s clear that physical resources were saved, but calculating the actual saving is still not possible. I assume GitHub Codespaces runs on Azure, so inherits the work Microsoft is doing around sustainable computing. But to what extent? You can’t find out precisely which data center the environment is running in, so you can’t figure out the carbon intensity.

Is the processing more efficient in the cloud? We assume so, but to what degree? Could the savings from not replacing the laptop be offset by a more frequent hardware refresh cycle in the data center? What resources are used by the prebuilds? The Codespaces UI gives you a nice output showing the time saved using them, but what about extra services, like the GPUs that are involved if you use GitHub Copilot where you get AI code suggestions every time you type something?

It’s great that you can now get carbon footprint data from all three cloud providers, but what about the services built on top of them? Now we need that carbon data on a GitHub organization level. What is the carbon cost of my software development – builds, tests, deploys, code hosting, dev environments?