Green Access is an experimental HPC-FaaS platform for "fungible" allocations, or allocations for running jobs across multiple allocations. Green Access implements Carbon Based Accounting (CBA) where the cost a running a job depends on the carbon footprint of that job. The Green-Access prototype consists of three python packages.
To install the green access service, clone the GreenAccess repository and install the
run pip install . inside the service/ folder.
To start the servive, authenticate with diaspora-event-sdk and create topics for producing events. In a Python terminal, run: from diaspora_event_sdk import Client
c = Client()
c.register_topic("ga-resources-<user>")
c.register_topic("ga-predictions-<user>")
Running those commands will require authentication with Globus Auth. Once authenticated, we create Kafka topics for passing resource information from the endpoints to the service. Update these topic names inside the green-access-service/common.py configuration file.
GreenAccess also relies on Redis and MySQL to store data. Those connection parameters can also be updated in green-access-service/common.py.
To start the web app run, uvicorn router:app. To start the faust consumer, run faust -A monitor -l info worker.
The Green Access client provides a Python client to interact with the Green Access service. To install the client, run pip install . inside of the client/ folder.
The Green Access Experiments package was used to run experiments on the Green Access platform. It includes several benchmark functions, as well as the ability to simulate users who follow certain policies. To install the package, run pip install . inside the experiments folder.