Quantum Computing -based Optimization for Sustainable Data Workflows in Cloud Infrastructures

Created by Valter Uotila (Qumpula Quantum)

Motivation

Data centers create huge carbon footprint

Google has developed a Carbon Footprint tool which enables an individual user to track their computing emissions in detail

This work is inspired by the possibility to access such data

Problem definition

Assume a user provides a workload consisting of multiple data science and data management tasks

Assuming that we can estimate the emissions, how should we divide the workload among the data centers so that the total carbon footprint of the workload is minimized?

Solution in theory

The formalized problem is simply solved by finding the shortest path in a weighted graph

Problem is formalized as a QUBO where the binary variables are work-data center pairs $(w_i, d_j)$. If the variable is 1, the work $w_i$ should be executed in the center $d_j$.

We introduce two simple constraints: every work must be executed on some data center and the emissions are minimized

Implementation

The QUBO is implemented with

the Ocean software than can be solved on Amazon Braket and
with Qiskit that can be run on IBM Quantum systems

More details in the jupyter notebook on Github

Results

Quality of quantum computing based solution is compared to the classical method

More details and results in the paper draft on Github

Results

Future work

Collecting the results and presenting it as an interesting demonstration of quantum computing and sustainability to the database community

Finding automatically correct scaling values for the constraints

Me and my colleagues are also researching other interesting optimization and machine learning problems on quantum computers!