What has Amadeus learned about green software by building it for real? Last year, I met Florent MOREL, Head of Green Software Engineering at IT company Amadeus, when he and colleague Stefania Dante joined my Building Green Software course.
Recently, we caught up for an in-depth chat about what Amadeus was up to, what they had learned from what they were doing, and what value they had got beyond helping to save the planet, because that’s just table stakes.
Rock me, Amadeus
--------------------------------
Florent and his team's mission at Amadeus is to reduce emissions from software systems and I was keen to hear how they’re going about it.
Their approach is to give engineers and application owners the data to understand the climate impact of their systems, identify minimizing actions, make them, and then judge the results. Perfect. A strategy based on that amazing tool for improvement: the feedback loop.
They decided to start by building an in-house carbon measurement tool. They soon discovered things were not that easy…
Florent and his team realized there were two big problems with delivering the data:
The first was that perfect carbon emissions data was impossible to get. Few public grids provided accurate real time data for the electricity they supplied, and most data centres are supplied with electricity from a public grid. If grids couldn’t do it yet, it was almost impossible for enterprises.
The second, more subtle but even more important, issue was they realized that perfect data would be tricky to act on in a way that led to incremental, measurable improvements. For example, if your system emits more carbon today than yesterday, is that because of a change you made or was it windier yesterday and the local electricity was just cleaner?
Counterintuitively, the team decided that direct, real-time carbon emissions weren't the best output for the tool they were building. Their engineers needed data that was more consistent and thus better for identifying and testing iterative improvements - those lovely feedback loops.
But what data was that exactly? For them, they judged it was better to use a proxy for carbon: power consumption. It became their north star metric.
Their tool then used a flat, averaged conversion factor to turn that power consumption into carbon emissions numbers that were consistent, comparable, and realistic enough.
1/n (continued in comments)
