Dear Chameleon users,
Happy almost Fourth of July! We don't have any dangerous or spectacular pyrotechnics to display for you, but we do have some "explosive" news regarding the edge capabilities we're developing for Chameleon users!
Updates from a summer… on the edge! Last month we announced the start of the "summer on the edge" preview period for CHI@Edge, an experimental IoT/edge testbed we've been cooking up. Since then, we've been busy building out the first parts of the system so it's useful enough to do some real work with. Today, we're happy to announce that there is support for reserving edge devices for your exclusive use, similar to how reservations work in other Chameleon sites. And, you have a full-fledged container workflow system that allows you to launch containers on the devices from your own Docker images, connect the containers/devices on isolated private networks, no matter how distantly located, and manage the entire experiment with a variety of interfaces: GUI, CLI, and Jupyter/Python.
We started with a handful of Raspberry Pi 4s and have been steadily enrolling more devices as we get them. Notably, several Jetson Nanos equipped with NVIDIA TX2 GPUs are ready to go! Your containers on the Jetsons can have full access to the onboard GPU, allowing you to perform some more interesting edge experiments leveraging ML/AI or other tasks well-suited to compute accelerators.
If you're interested in using CHI@Edge resources for your experiments now or in the future, or just want to follow along and learn about this work, join the chameleon-edge-users mailing list. Please let us know what thoughts you have about what we have so far, and what you still would like or need. In the coming months, we will be working on integrating more types of peripherals and devices, with the ultimate aim of open-sourcing device enrollment so you can add your own devices to the testbed!
Finally, invite users to projects via email! This has been a long-standing request that we're happy to have finally had time to address in June. Before, when adding users to a project, the PI would have to first ask all collaborators to first sign up for a Chameleon account, and then give the PI their account details so the PI could complete the process. Now, PIs can simply enter email addresses of their collaborators or students even if they don't yet have a Chameleon account. We'll send them a personal invitation email. When the recipient signs up for Chameleon, they are automatically added to the PI's project. This should make onboarding students in particular much easier!
Hardware catalogue gets a nip and tuck. Chameleon's browseable hardware catalogue gives both a high-level and fine-grained way to dive through the set of hardware available for experimentation. But, there were some important aspects that weren't obvious, such as "which nodes have SSDs?" or "which nodes have multiple networks attached"? The catalogue now is more streamlined, allowing you to filter nodes by their high-level classification, but also filter by total storage, number of cores, or RAM size. The "reserve" helper also has been updated to allow making reservations when multiple resource constraints are in place, allowing you to specify a reservation, e.g., for nodes having more than X cores but less than Y RAM. Give it a spin!