Nathan Faggian https://2016.pycon-au.org/schedule/168/view_talk Bushfires are one of the most destructive forces in our environment. A major responsibility of the Bureau of Meteorology is to provide information on weather conditions that could make bushfires difficult to fight. Fire agencies then use simulators of fire behaviour to inform finer grained warnings to protect the Australian public. Objectively choosing the best performing simulators, or indeed the best versions of the same simulator, with respect to end-user requirements is a difficult task that requires a large amount of computation.
The introduction of new tools like Docker, Dask and Jupyter notebooks make the task of large scale computation much easier! The Bureau of Meteorology is looking at these technologies to help modernise its legacy workflows and processes, one example is our recent work in the area of fire simulators.
This talk will work through the approach developed for a reproducible environment (using Docker) and then how we form workflows (jupyter notebooks) that execute fire simulators on a small compute cluster of both windows and linux hosts (Dask, Distributed). Finally, we will work through an example of scaling up an experiment and demonstrate a process that is easy to replicate and follow for fellow scientists.