Getting Scientists On Board with Cloud Portals

Abstract

Cloud resources already exist that enable doing science on massive datasets and computationally large problems. There is also the pressure for collaboration and replicability, which clouds are already strong with. It requires some adjustment by scientists to learn cloud tools. Early adopters are willing to self-start, but most scientists need a push to spend precious science time in training up. We find two main approaches to tackle this. The (easy) technical solution is to hide the cloud coding while touting its capabilities, and the (hard) people solution is to devote time to workshops and training. JupyterHub cloud portals (like HelioCloud and Helio-Lite) make workflows nearly like the conventional 'on my laptop' experience, PyHC package devs and portals like Heliodata embed cloud APIs, and data portals like Heliodata include code stubs for accessing cloud data. Meanwhile, HelioCloud/Lite provides copious tutorials and videos (oft ignored) with hands-on demos and workshops that have proven effective. Yet currently we are still reliant on external pressures to motivate scientists: that there are science problems only solvable with cloud resources, that the push for stronger replicability will motivate users to adopt better practices. We share our experiences to spark discussion.