Workshops and Tutorials at PACT 2026

Chameleon: An Open Platform for Computer Science Experimentation

Scheduling: Monday, October 19 (afternoon)

Tutorial Presenters:

Kate Keahey

Argonne National Laboratory / University of Chicago

Mark Powers

University of Chicago

Abstract

This tutorial introduces Chameleon (chameleoncloud.org), a publicly available NSF-funded platform for Computer Science Research and Education. It provides access to diverse, state-of-the-art hardware which includes a range of accelerators (NVIDIA, AMD, and Intel), sophisticated HPC architectures and configurations (including e.g., Fugaku nodes), disaggregated hardware (including Liqid and GigaIO systems), configurable edge systems (flexibly deployed NVIDIA Jetson and Raspberry Pi), and high-bandwidth interconnects. The hardware is distributed across three sites at University of Chicago, Texas Advanced Computing Center (TACC), and the National Center for Atmospheric Research (NCAR) connected by fast networks, which provides opportunities for distributed experimentation. Chameleon is purpose-built to support bare metal reconfigurability, critical to support experimentation on topics such as power management or performance variability, allowing users to modify the full software stack from firmware, through custom operating system kernels and drivers, to user-level software. In addition, the system also supports virtual machines to support large scale experimentation as well as less configuration demanding research and education projects.

The system has been in existence for a decade and over that time supported a community of over 14,000, who have collectively reported over 1,300 publications from their work on the system. The tutorial will walk attendees through the full experimental lifecycle on Chameleon: from browsing the resource catalog and provisioning hardware, to constructing complex experimental topologies such as GPU clusters and multi-node HPC environments, to snapshotting, packaging, and sharing reproducible artifacts via Trovi, an artifact repository for reproducible research. Throughout, we will ground the discussion in use cases directly relevant to experiments in parallel computing, heterogeneous workloads, and AI/ML system benchmarking. Attendees will gain hands-on experience with Chameleon Cloud and leave with reusable experiment patterns they can immediately apply to their own research or education projects on the system. The tutorial will also include a selection of advanced topics and Q&A adapted to the needs of the audience.

Bios

Kate Keahey is a senior computer scientist at Argonne National Laboratory and senior fellow at the University of Chicago’s Consortium for Advanced Science and Engineering (CASE). She received her Ph.D. in computer science from Indiana University and joined Argonne in 2001. One of the pioneers of infrastructure cloud computing, she created and led the development of the Nimbus project—recognized as the first open-source Infrastructure-as-a-Service implementation—and has spent her career advancing the use of cloud computing concepts in scientific research. Her current research interests focus on cloud computing, resource management, and reproducibility. She is the PI and founder of the Chameleon testbed project, an NSF-funded reconfigurable experimental platform available since 2015. She is also co-founder and co-editor-in-chief of the SoftwareX journal, which won the PROSE Award for innovation in journal publishing in 2016. Keahey has presented Chameleon tutorials at SC, ACM REP, IC2E, and numerous workshops and user meetings, and has given keynotes and invited talks at major venues including IEEE CLUSTER.

Mark Powers is a research software engineer at the University of Chicago for Kate Keahey’s Nimbus team. He is the DevOps lead for the Chameleon testbed. His research interests focus on cloud and edge computing, system design, and reproducibility.

Back to Workshops