A configurable experimental environment for large-scale edge to cloud research

Recent news

  • Chameleon at SC25

    We're heading to SC25 in St. Louis. Read what we'll be up to!

    SC2025 is taking place in St. Louis, Missouri from November 16 to 21, 2025. Kate Keahey, Chameleon PI, will be attending the conference and hopes to see some of you there. In this blog, we list some of the events where Chameleon is getting let out of the box!

  • Chameleon Newsletter & Changelog - October 2025

    Testbed updates, new features, webinars, and other exciting news from our user community
    by Marc Richardson

    October was Performance Month for Chameleon Cloud. We're excited to share a variety of performance upgrades for various testbed services, a new Trovi feature, new webinars, user resources, and awesome Trovi artifacts developed by our users!

  • Introducing MINCER’s Performance Measurement and Reproducibility Appliance

    Building a Standardized Framework for Reproducible Performance Measurements Across Diverse HPC Architectures
    by Gonzalez, Jhonny

    Reproducibility is a cornerstone of scientific computing, yet achieving consistent results across different hardware environments remains a significant challenge in HPC research. The MINCER project tackles this problem head-on by providing researchers with an automated performance measurement appliance on Chameleon Cloud. Using Docker containers and the PAPI framework, MINCER enables standardized collection of performance metrics across CPUs, NVIDIA GPUs, and AMD GPUs, making it easier to compare results and understand how system-level factors influence computational performance. This post explores how MINCER is helping make HPC experiments more reproducible and accessible to the research community.

  • Bare Metal or KVM? Which Should You Choose and When

    A detailed comparison of hardware access, reservation systems, and storage options for users
    by Mark Powers

    Not sure whether to use bare metal or KVM for your next Chameleon experiment? You're not alone. With KVM's recent addition of H100 GPUs and advance reservations, the choice isn't always obvious. This comprehensive comparison walks you through the practical trade-offs: when you need the isolation of bare metal for accurate benchmarking, when KVM's 6-month leases and resource sharing make more sense, and how to navigate the different storage and networking options. Make informed infrastructure decisions that accelerate your research.