Chameleon Changelog for January 2022

Dear Chameleon users,


It's been a hard-ware month (sung to the tune of that helpful Beatles ditty)! We've got some great updates about new hardware, some provided by an entirely new Chameleon site. Without further ado:


A brand new Chameleon site is online! We are very happy to welcome a new associate site to Chameleon, increasing the total amount and diversity of hardware available to all researchers. NCAR (National Center for Atmospheric Research) in Boulder, Colorado and is now the home of CHI@NCAR; this site evolved out of the IndySCC competition last year, and now will continue to serve the computer science research community by donating their unique and high-powered ARM nodes: the site is equipped with 5 Cavium ARM Thunder X2 nodes, each with 2 CPUs, a total of 64 cores (256 threads), and 128GB RAM. The site can be reached here: CHI@NCAR.


Node descriptions for are available in the resource catalogue. Enjoy, and join me in giving the site operators a hand for making this a reality!


Heavy-hitting AMD nodes installed at CHI@TACC! More new hardware? Indeed! CHI@TACC has installed 20 new nodes, each with two AMD EPYC 7763 and a total of 128 cores (256 threads), 256 GB RAM and 480GB SSD storage. These nodes are divided into two groups: the first 16 are ready for a future upgrade with Mellanox HDR 100 Gb/s InfiniBand (we will announce when this has arrived and is configured, supply chain availability is causing significant delays); the remaining 4 have been configured with two AMD MI100 GPUs, which is available today.


Increased storage and GPU opportunities at CHI@UC! These hardware updates are getting a bit much, no? At CHI@UC, several existing nodes have been upgraded with additional hardware. In particular, we've installed 8 SSDs (with a mixture of consumer and enterprise grade). Two nodes have been upgraded with 2x Intel 7.68TB PCIe4.0 SSDs, and two others now have 2x Corsair 2TB m.2 SSDs. This means that many experiments needing more storage capacity or performing experiments on storage itself are now possible. Finally, we installed another V100 GPU that arrived late to the party to one of the Phase 3 nodes, increasing total GPU node capacity.

CHI-in-a-Box stable release. During the rollout of the new associate sites, we've been hard at work on kicking the tires, working out the kinks, and putting a layer of polish (or clear-coat, I guess, to continue with the vaguely automotive metaphor) to CHI-in-a-Box, the packaging of the CHameleon Infrastructure (CHI), which is powering all Chameleon sites. Over the past month we've been improving on the installation process with feedback from our two new sites and documenting a lot more information and options. Notably, we've added support for "flat" networking, where CHI does not need management access to the network fabric and can use pre-allocated network segments instead. This does mean users of the site cannot provision their own isolated L2 networks, but can accommodate more restrictive host environments.

Add a comment

No comments