jimbou
commented
11 months ago Telma's notes from meeting on Friday 8/12 : Meeting with Green Software Foundation – 08.12 Why does It need to be a model? (Dimitris explains his idea of how to solve this) • Theoretically it would work. However, it is not how they would design it for the IEF. They want it to be a model so that it can interact and work with other models. Because of the way the framework was designed, this idea doesn’t match. Therefore, they would not be able to support this version.
Their encouragement • For parameter optimisation: get specific with examples. There are not many places where we can change the parameters (some ideas are times and regions. Carbon optimization is really the only thing we can play with). Not having the project be a model would limit the product. • Only use case they can think of right now: carbon optimisation with e.g. location • IEF is a low-level framework; we can technically do whatever we want. But they want models. They want to encourage people to go the way the IEF is designed to, aka. models. • Add more interesting use cases with the models
More info about models • We can do anything we want; the key thing with the standardisation is the interface. The preexisting ones are built with typescript but they can now call other languages (or are working on it?). Command line tool. We should be good, as long as it communicates inputs and outputs with the IEF. • Models are very abstract in the IEF • Functional interface (the actual functions exposed) and keys (names of keys) are also standardised. Units of measures are also standardized, e.g. energy not en or eng.
What parameters? • What we are optimising are parameters to a model. • Example: water model for water scarcity. That model might have some parameters to tune, but it doesn’t exist yet. We must create a very generic project as a lot of the models don’t exist yet and are being built. • Not much else that we can really change that exists right now.
Other comments • They wrote the input file first. They recommend we write a pseudo input file to understand better. • Parameters to change: location, time, instance types (would have to rerun everything for that one, but they seem interested in this) ◦ E.g. Check if there is a better instance type to have better utilisation • Carbon Aware SDK • A model which you can insert in a pipeline takes whatever data it gets as input (configurable, find the best carbon, etc.). Config advisor, optimisations. Some sort of configuration. What it does is basically based of the input data, analyze input data based on where it is running, find out if other location or time is better. Print on command line or elsewhere • Key thing: not changing or recalculating anything. Can be dropped into the pipeline
Users of IEF • Alpha, so not many people are using it. • Built for anyone interested in calculating sci.
When Asim replys back with the information on the final project idea + using Telma's notes we should transfer all the info for the final project to the report