What are the use cases for real time power and carbon intensity information?

[adrianco]

Most fundamentally, there is a gap between what can be done in a datacenter context today, and what can be done in the cloud, because some information is only known by the cloud provider. So one use case is to enable innovation that spans datacenter and cloud environments.

The CNCF Kepler project is performing attribution and allocation for container based workloads. Making Kepler's model less wrong and more useful in cloud environments is one use case.

Cost optimization vendors such as CloudZero want to add carbon optimization to their products. From the PRFAQ:
CloudZero CEO Erik Petersen said [APPROVED REAL QUOTE]“Sustainability has long been a concern for cloud engineering teams. But for as long as it’s been on engineers’ minds, the missing link in making sustainability a non-functional requirement has been the data. Every engineering decision is a buying decision — and consequently, an emissions decision — but without real-time data on cloud infrastructure’s cost and carbon consequences, engineers haven’t been able to prioritize efficiency as they build. GSF Real Time Cloud is a crucial step in establishing a universal definition of cloud sustainability; now it’s up to organizations to quantify and optimize their cloud efficiency in the name of sustainability — an existentially urgent concern for all of us.” — Erik Peterson CTO and Founder, CloudZero.

[PindyBhullar]

I think we need to break down the frequency of real time power and carbon intensity information. Split out the frequencies by the value it brings to the cloud providers and then the customers. The key question we have to ask is, once we have this data what can we do with that data.

[adrianco]

It's difficult to generate data to calculate SCI in the cloud, so a lot of approximations and short cuts are taken. This SCI case study shows some of the difficulties. Green-Software-Foundation/sci-guide#53

[adrianco]

I think we need to break down the frequency of real time power and carbon intensity information. Split out the frequencies by the value it brings to the cloud providers and then the customers. The key question we have to ask is, once we have this data what can we do with that data.

The proposal is that carbon intensity data is hourly, aligning with some of the 24/7 work and other data sources that provide forward predictions of carbon intensity for the next day, hour by hour.

Workload activity metrics are often at one minute resolution, and as utilization varies that can be turned into a power estimate for a defined instance type, or the power can be measured directly.

Both carbon intensity and power can be instrumented as confidence intervals with a most likely value, for use by monte-carlo models.