Poster

Yagmur Kati (HU Berlin)

When a workflow finishes, most of its context disappears: log files, monitoring data, folders on somebody's machine. The publication stays visible, the steps behind it do not, and sharing the code is rarely enough to repeat a computation. The FONDA-VIVO platform (hu.berlin/vivo) is being developed as a linked information system for scientific workflows, so that a workflow run stops being an isolated technical event and becomes a structured research object that can be searched, compared and reused. A collector on the FONDA cluster records each run automatically as it executes: the code version, input data, software environment and machines, together with runtime, processor, CPU, GPU and memory use, electricity measured with Kepler, and CO2 estimated from the carbon intensity of the German electricity mix at that moment. Metadata is kept per workflow stage as well as per run, and each record is linked to the publication, the workflow code and the trace archive. This makes versions of a workflow comparable: which implementation runs faster, which step demands the most resources, and where energy or CO2 can be reduced.