Chameleon Changelog for February 2021

Dear Chameleon users,


Welcome to a new month, full of possibility and intrigue! In the Northern Hemisphere we anxiously await the vanquishment of our common enemy, "Wintry Mix." In the southern hemisphere, I'm not sure what you're up to because I've been too hesitant to look. One can only take so much contrast. In any event, here's what we've been up to whilst waiting out the end of our winter.



A new major release of CHI-in-a-Box. CHI-in-a-Box is the packaging of Chameleon--a set of software and provisioning tools that allows anybody to stand up their own Chameleon site. Over the last months we have been working on a new release that eliminates some major shortcomings. CHI-in-a-Box now includes project and account management via Chameleon's federated identity, and utilization accounting. We also vastly improved the documentation and tooling, wrote a new Installation Guide, fixed several bugs, and added support for CentOS 8 on the control plane. With this release, there is no longer a need for an associate site to onboard, manage, and offboard user accounts for a Chameleon associate site--you can now set up a CHI-in-a-Box site and any user already tied to a Chameleon allocation can simply use your site like one of our core sites, e.g. CHI@TACC or CHI@UC. Thanks to those who have reached out and helped shape CHI-in-a-Box to where it is today! Whether you're interested in setting up a Chameleon associate site via CHI-in-a-Box (in which case, do read this interview with Northwestern University, who set up the first associate site), or you want to use the infrastructure in a more independent use-case, please let us know!


Interactive Chameleon User Guide! We've been watching how more people are leveraging notebooks for teaching and learning and decided to convert some of our existing documentation and tutorials into an interactive form, via Jupyter notebooks. The first module we're publishing covers several of the advanced networking capabilities, such as BYOC (Bring Your Own Controller) and creating per-experiment isolated layer-2 networks. You can find it on, you guessed it, Trovi! We'll be building this out in the future with more examples and modules. Most examples leverage the Python API to Chameleon we announced last month and can hopefully give you more ideas about what is possible on the testbed.


Load balancer and orchestration cloud services now available at KVM@TACC! We have deployed and configured two additional powerful capabilities at the KVM site. All users now can configure load balancers across their KVM instances. HTTP, TCP, and UDP load balancers are supported, and you can configure both health checks and the host selection algorithm. This is a commonly-found technology in commercial cloud deployments and we hope it is useful both in your experimentation and teaching. Additionally, we have added support for orchestration at KVM. As on our bare metal clouds, orchestration on KVM allows you to conduct a group of interrelated pieces of cloud infrastructure as one unit. Like the load balancing capability, it is a feature shared by many cloud infrastructures and thus can be an interesting teaching aid. Over time we will also be working to adapt our existing complex appliances such that they can be deployed on KVM--this can make it easier to deploy virtual NFS and MPI clusters, for example. Stay tuned!


A "lighter" form of publishing experiment artifacts on Trovi. Trovi, Chameleon's sharing platform, already supports publishing your Jupyter notebook-based experiments to Zenodo, a service that provides long term storage and a durable DOI, which can be cited. Sometimes, though, you just want to share something more widely without formally publishing in this way. This is now possible on Trovi: in the "Share" menu, there is a new option to publish without a DOI. Artifacts published in this way can be discovered and launched by any Chameleon user, but are not additionally stored on Zenodo. Read more in our docs.


Ubuntu 20.04 appliance now available. Ubuntu 20.04 LTS was released last year and is now officially supported as a Chameleon base image. We also released an additional CUDA variant of the image with the CUDA 11 toolkit preinstalled. Enjoy!


Regular care and feeding of Chameleon. Aside from the above, we spent some time tracking down issues we sometimes see reported. Notably, there were issues with Jupyter notebooks breaking after being logged in for a while. Let us just say that we talked to Jupyter about this, and perverse incentives, and a few other philosophical concepts, and he's fortunately come around on this matter.


Until next time!

Add a comment

No comments