Chameleon Changelog for April 2022

Dear Chameleon users,

It’s hard to believe that summer is soon beginning, especially given the cold here in Chicago – below are some new things on Chameleon that will hopefully make your summer projects better.

CHI@Edge release! If you’ve been following along with the previous few changelogs, you’ve already seen some exciting new features on our Edge testbed in what we call version 2.0. Notably, we announced an SDK which supports open device enrollment, allowing anyone to add a device to the testbed. If you are interested in enrolling your own devices, we’ve just created a screencast that walks you through the process – or you can read our tips & tricks blog from last month or follow our documentation). Our resource pool now consists of 30 devices, including Raspberry Pis (models 3 and 4), NVIDIA Jetson Nano – and a new addition: four NVIDIA Jetson Xavier NX. The Xavier NX has more than 10x improvement in GPU performance over the Jetson Nano, moving from the Maxwell to the Volta generation, as well as adding dedicated tensor cores, deep learning, and vision accelerators. It also moves from 4 to 16gb of ram, and 4 to 6 cpu cores. All this is done with only a 2x increase in peak power draw, from 10w to 20w.    Other improvements include a significantly smaller implementation footprint which leaves more space on the device for whatever it is you are doing, more flexible networking, and a better GUI experience that helps with debugging containers. This month, all of our devices were migrated to 2.0, and we did more testing and improved the stability of the system. As announced in our last changelog version 1 of CHI@Edge will no longer be supported. As always, if you have any edge questions, please check out the chameleon-edge-users group.

Easy Edge to Cloud experiment setup. Last month, we released a blog post showcasing a few experiments that were built on CHI@Edge. All of these experiments share a similar underlying structure, with experiments deployed over edge and cloud nodes. We’ve recognized a common structure in some edge to cloud experiments — an edge device collects data that is sent to a machine learning model in the cloud for processing. To help support this pattern, we’ve released a new Trovi artifact that sets up a simple experiment fitting this mold. Our hope is that this can be easily copied, modified, and extended to form a starting point for many experiments. See the artifact for more details.

Integration with new Trovi API and Git import/export. In December, we announced a new public API that turns Trovi – our experiment sharing portal – into an open platform, making it into an experiment repository that can be integrated with any testbed or cloud. We are happy to announce that this month, we’ve updated our portal to use this new, open Trovi as a backend. You will see no difference in our Jupyter/Trovi/Chameleon integration – but now other testbeds or clouds can develop Chameleon-style Trovi integrations so that in the future you will hopefully be able to use it as a base to discover and launch experiments on a variety of testbeds.We’ve also added a feature which allows you to import a Trovi artifact from any git repository. This means that if your experiment files are stored on GitHub, you’ll be able to share them with Trovi at the click of a button. For more information, check out our documentation on importing an artifact. In the spirit of creating an open environment, we’ve also added a section on exporting your Jupyter notebook to a GitHub repository. For now, this requires use of the git CLI, but we are considering improving this feature based on feedback.

CHI-in-a-Box release. You may have noticed under the “Experiment” tab on our website, Chameleon is growing quickly with associate sites CHI@NCAR and CHI@EVL, all made possible through CHI-in-a-box. If you are considering deploying CHI in a Box yourself you will be interested to know that now, CHI-in-a-box includes Chameleon image tools, allowing site operators to deploy the latest Chameleon-supported images and to clean up deprecated images at the site. Usage reporting has improved since its announcement last month based on the feedback from our associate sites. The usage reporting allows associate site operators to monitor bare metal node usage via our internal Grafana dashboard and the weekly KPI report emails.


Add a comment

No comments