Jeff Geerling

Early in 2023 I worked with YouTuber NetworkChuck to set up a Python-based system aggregating vote data from 100 rooms for MrBeast's "Ages 1-100 Compete for $500,000" video. We wrote an open-source application to manage the state of the challenge, and worked with various producers and technicians to install our custom voting and lighting solution on set.

Contributed a variety of patches and improvements to Ampere Computing's open source repositories for benchmarking and exploring the arm64-based Ampere Altra and Altra Max CPUs using tools like HPL, Steam, and GPU-accelerated applications (like LLaMA, games, and graphics rendering libraries).

I flew from the US to Canada, and enlisted the help of a homing pigeon, to answer the age-old question: Is a pigeon faster than the Internet?

I explored using Timebeat's soon-to-be-open-source software for time synchronization and PTP alongside their new 'Time Card mini' PCIe card, which straps a high-accuracy GPS clock and an OCXO (Oven Controlled Oscillator) to a Raspberry Pi Compute Module 4.

I created a repository to aggregate SBC (Single-Board Computer) review data and testing methodology for the various hardware devices I test. The goal is to be completely transparent with how I test devices, and to share all the data from my testing which may have not made it to a public repository before.

I built the automated Top500 benchmark to assist people running the HPL, or 'High Performance Linpack' benchmark on their systems—whether a single node or tens or hundreds! I wanted an easy way to compare systems against each other, and all I've found online are hand-tuned scripts or tutorials made just for one platform. This Ansible playbook works on any system that can run Ubuntu or Debian, and even works in clustered environments, so you can easily measure how many Gigaflops, Teraflops—or even Exaflops!—you can get out of your system!

I built an open source Ansible automation project to go along with my Ceph distributed storage exploration on a new 6-node Raspberry Pi Mini ITX motherboard, the DeskPi Super6c. I also published a YouTube video guiding people through both the hardware and the software setup: https://www.youtube.com/watch?v=ecdm3oA-QdQ

A couple years after the Raspberry Pi Compute Module 4 launched, the open source community finally got one of the less-flashy features working: PTP hardware timestamping. This makes the Compute Module 4 a very powerful timekeeping device, as it can behave as a network time server with the highest level of precision, using the Precision Timing Protocol and devices like GPS receivers and TXCOs (Temperature-Controlled Oscillators) or chip-scale atomic clocks to control timing over Ethernet networks.