Using Human-Centered Approaches to Make Cloud Computing Green

Abstract

Our daily life and work are migrating to cloud computing. One major consequence is the excessive amount of energy consumed by computation and storage infrastructures behind cloud computing (namely datacenters). In 2016, the US Department of Energy found that all datacenters consumed roughly 2% of all electricity consumed nationwide. This number is expected to grow exponentially as more and more industries are adopting the clouds.
One obvious solution is running datacenters on renewable-based power grids; however, a major challenge is the volatile energy supply from renewable resources, which results in intermittent unavailability of the clouds. For cloud service providers, unavailability, even for a one-second duration, poses a huge monetary loss, because users nowadays have almost zero tolerance of service delay and/or disruption.
Most prior works focused on designing more resilient systems and/or algorithms to minimize the unavailability intervals and frequencies. We adopt a drastically different approach, by exploring a human-in-the-loop design to tackle the challenge. We aim to design an interactive system that (i) takes user preferences and their tolerance (to unavailability) into account when tuning system configuration and design, and (ii) uses system metrics and availability predictions to encourage users to adopt a more energy-efficient behavior.

Presentations

• A Human-centered Approach to make Networked Entertainment Green: A Case Study of CDN, International Workshop on Networked Entertainment Systems (NES2022)
  

Students Trained 

• 1 Student