Hello, My name is Erin Shaw and I have been serving as an ACI-REF for almost three months at USC’s Center for High-Performance Computing (HPC). I have been focusing on outreach and as a result we will be collaborating with a group that provides an HPC service to the medical research community on USC’s health sciences campus across town. They have access to a mailing list and workshop space and have asked us to give a three-hour brand-new-user workshop. I expect this will result in about two hours of hands-on training.
On the surface it seems easy — novice material, some of it well-covered online, super-smart population. However, we currently hold introductory workshops on Linux and HPC that run two hours each, and there seems to be barely enough time to scratch the surface of either given the workshop format. To make matters more challenging, we’ve been told to expect 80 participants. For comparison, our regular workshop attendance is around 20 and participants already have (and have tested) HPC accounts. So how do we create a tutorial that will give complete novices an authentic hands-on experience and leave them feeling optimistic about high-performance computing?
Let’s start with outcomes. They are basic: At the end of the workshop, participants will Understand why/when to use HPC; Learn about USC’s HPC cluster; Connect to and transfer files to our HPC network; Edit a file; and Run a file on the cluster. But to make these happen, users — mainly graduate students and a small number of faculty and staff — who have never before seen a command line and may not even have an account, are going to have to learn Linux and understand at least a little about scheduling.
TACC has a good 90-miute new-user training video for XSEDE that contains some introductory slides that can be used together with our current materials. Since we have only two hours for a hands-on workshop, the most pressing question is how to streamline the busywork. Clearly, software like X-Win32 will have to be downloaded prior to the workshop, but setting up an sftp connection can be confusing, so we’re looking at, what, 15 or 20 minutes to connect and transfer files? Linux is a much bigger and more fundamental time sink so what gets taught will have to be strictly need-to-know. But, of course, there’s a lot to need to know. And editing has to be easy so, Nano, then?
Then there is the question of pedagogy: How can we elevate the tutorial so that the take-away is more than a sum of discrete steps? What’s going to have someone leave the workshop thinking HPC is awesome and easily learnable? We need a killer example to pull it together. I found a 2015 article about how Fujitsu is enabling customers to use their own data to make HPC technology more accessible, and like the idea of creating an example with a medical theme. I am sure my ACI-REF colleagues have had to re-work courses before, for different schedules and populations, and I would be interested to learn what worked and what didn’t.