Integration of carbon aware SDK with Impact Framework

[srini1978]

Today the carbon aware SDK is a mature open source project. The carbon aware SDK integrates with backend APIs- Watt Time and Electricity maps and helps identify green regions for running cloud workloads. The 2 techniques that carbon aware SDK enables are region shifting and time shifting. The resultant of these techniques helps in reducing the "I" value of the SCI (Software carbon intensity) formula.

The idea here is to integrate Impact framework with carbon aware SDK. We could embed carbon aware SDK (either as a SDK or API) in the computation pipeline of Impact framework such that the SCI-O pipeline runs off of carbon aware SDK.

Prize category

Help the participant understand which of the prize categories this idea would best fit

• Best Model
• Best Content
• Beyond Carbon
• Best contribution to the framework

User Story

Describe the benefits of this change, ideally in the following format:

As a developer, I would like to link embed carbon aware SDK in the manifest file of Impact framework so that the carbon intensity is calculated by the SDK.

Rationale

Provide the relevant background information to understand why this is a useful thing to work on

This is where you can provide more details that help the reader understand why you want to make this change

Impact

Finish this scentance: "If this idea is successful implemented the ....."

Implementation guidelines

• If there are specific implementation details that are REQUIRED or PREFERRED please describe them here
  e.g. this model MUST conform to the ModelPlugin interface
  e.g. this feature MUST be part of the supercomputer module
  e.g. this feature SHOULD be be a model plugin

Difficulty

Provide an assessment of the expected difficulty from 1-5 to help hackers find tasks at their preferred skill level
2/5

Scope Of Work

List some of the tasks that will be required to implement this idea

• A new model plugin
• A new feature in an existing model plugin
• A manifest file will have to be written
• Implement test cases
• A new feature to the framework itself (the cli tool)
• It will require changes to the documentation

Examples and resources

IF docs
Model building tutorial

Manifest File

Provide a dummy Mmnifest file to help the implementer understand how this functionality will be executed, provide them with the necessary data, observations and structure so they can move fast with development.

name: nesting-demo
description:
tags:
  kind: web
  complexity: moderate
  category: cloud
initialize:
  models:
    - name: teads-curve
      model: TeadsCurveModel
      path: "@grnsft/if-unofficial-models"
    - name: sci-e
      model: SciEModel
      path: "@grnsft/if-models"
    - name: sci-m
      path: "@grnsft/if-models"
      model: SciMModel
    - name: sci-o
      model: SciOModel
      path: "@grnsft/if-models"
    - name: sci
      model: SciModel
      path: "@grnsft/if-models"
graph:
  children:
    child: # an advanced grouping node
      pipeline:
        - teads-curve
        - sci-e
        - sci-m
        - sci-o
        - sci
      config:
        teads-curve:
          thermal-design-power: 65
        sci-m:
          total-embodied-emissions: 251000 # gCO2eq
          time-reserved: 3600 # 1 hour in s
          expected-lifespan: 126144000 # 4 years in seconds    
          resources-reserved: 1 
          total-resources: 1 
       sci:
          functional-unit-duration: 1 
          functional-duration-time: ''
          functional-unit: requests # factor to convert per time to per f.unit
      children:
        child-1:
          inputs:
            - timestamp: 2023-07-06T00:00
              duration: 10
              cpu-util: 50
              e-net: 0.000811 #kwh     
              requests: 380
        child-2:
          inputs: 
            - timestamp: 2023-07-06T00:00
              duration: 10
              cpu-util: 33
              e-net: 0.000811 #kwh     
              requests: 380

[srini1978]

carbonawareSDK

[jawache]

@srini1978 there are two ways a CA SDK plugin could be used in the IF, as far as I can see:

1. To be a source of grid-carbon-intensity data. If you are already using CA SDK in your organization and have configured various data sources inside it already, makes sense to use that instead of directly connecting to watttime or electricity-maps

      pipeline:
        - teads-curve
        - casdk-grid-carbon-intensity # returns grid-carbon-intensity for this region/location etc...

2. To simulate what would happen if this component was run in a carbon-aware way (time/region shifted), so something like so:

config:
  carbon-aware:
    time-shift: +-4hr
    region-shift: us-only
pipeline:
  - carbon-aware

What's your thoughts?

[jmcook1186]

or to find the low carbon option froma list of possible regions by wrapping a call to the CA-SDK bylocations/best endpoint

[danuw]

Some initial work has started here - https://github.com/danuw/if-casdk-plugin with @jmcook1186 's help. Will keep the suggestions above in mind.

Also for a while, I was wondering about creating a custom data source (for example having solar panels, I should be able to assume that my carbon intensity is 0 while I use my battery, or the sun powers the house or however we want to calculate this, my point being...: ) so that I can combine the carbon rating and my own data source forecast - and I assume the IF would be a good one for that right, ie combine data sources as the current SDK would need some customisation too at this stage so may be best to improve a/the model?

[srini1978]

Love both the ideas @jawache . We can also think of an extension capability where the IF could suggest better times/regions to run the component based on historical data around 7 day emissions or any other capability that CA-SDK has today available out of the box.