In this letter, we write about SuperCloud from the perspective of a newbie user, Peter Fisher, and a pro, Lauren Milechin.
Dear ORCD Community,
Peter’s research in particle physics and dark matter has always led him to use high performance computing (HPC) and high throughput, low data systems (HTC) – with many independent jobs processing sequential data sets. He grew up professionally along with computing clusters, learning things “organically” – get some scripts for the new system from someone, hack away to make it work and go from there.
This week, Peter started using SuperCloud and had a very different experience – you have to work and think differently and the SuperCloud team helps you do that. SuperCloud bends toward machine learning, requiring that users know something about how the machine works and doing some front work to make sure the right numbers of CPU, GPU, and memory gets allocated to your task.
SuperCloud has a specific configuration so users can debug their workflows involving the launch of many jobs that pass messages to each other – a requirement for machine learning and other AI applications. The SuperCloud team has also demonstrated they can launch jobs using 32,000 cores in only 4 seconds.
The SuperCloud team is available through email to help get your code running and scale up. They also host weekly office hours to answer questions and help debug issues. Office hours with Lauren are drop-in and do not require an appointment.
SuperCloud is widely available to the MIT research community and you can request an account here and start splashing around. We highly recommend their Practical HPC online course, which takes some time (and requires that you pass their quizzes with a 70% score), but serves as an excellent introduction to High Performance Computing in general, and SuperCloud in particular; completing the course gets you a certificate, and also unlocks additional computational capacity for your account. The Practical HPC online course began as a live IAP class offered yearly. The live course incorporates the online material with hands-on activities and project work and has a for-credit option.
Peter and Lauren