Wednesday, March 28, 2012

Workflows and Interoperability

This afternoon, I had a chance to drop in on the workflows session at the Community Forum. Workflows – or rather modular, interchangeable, and reusable workflows – are a hot topic, because they allow scientists the opportunity to benefit from the previous efforts of others, cutting down research times and costs and increasing productivity. Martin Junghans from KIT, the state University of Baden-Württemburg, explained how WisNetGrid enabled creative workflows: "Scientific workflows reflect the creativity of scientists," he posited, "and complex workflows become inevitable. So our goal is to allow scientists reuse existing methods and algorithms as part of the workflow functionality." The way to do this, it turns out, is semantically.

Semantic descriptions of complex workflows allow users to quickly decide whether a service or workflow is appropriate or not. By allowing jobs to be tagged with metadata, the WisNetGrid facilitates not only easy searching of appropriate tools, but automatic service discovery and support for semi-automatic discovery of workflows.

Vladimir Korkhov then described how SHIWA (Sharing Interoperatibility Workflows for large-scale scientific simulations on Available distributed computing infrastructures), could be used to analyse neuroimaging data. Again, the interoperability is key: "Sharing and re-use of workflows are needed to promote scientific collaboration," he concluded.

Stefan Bozic from the Karlsruhe Technical Institute followed with an illuminating (pardon the pun) talk on simulating organic light emitting diodes. Here, UNICORE provides the integration infrastructure, providing a UNiform Interface to COmputing REsources. Implementing standards from the Open Grid Forum (OGF), it provides seamless and secure access to distributed grid resources. For the application interface, GridBeans provides a way in to HPC resources, even for non-experts. GridBeans decouple applications from the ever-changing underlying grid, whatever that might be in terms of infrastructure. But an important point is that different simulation can re-use the same bean.

For the task in hand, structural data format is the open standard Chemical Markup Language (CML), while OpenMolGrid library provides the means for data flow management. Currently, the applications being used for the actual simulation are DL_Poly and BigDFT (this is essentially a chemistry problem, after all) as well as GridBeans. This was an extremely interesting presentation, and I'd recommend to take a look at the slides for a full taste of the talk.

Well, that about wraps it up for an interesting 90 minute session. I should also mention, the gala dinner was wonderful - a real chance for people in grid to get to know each other. Can't wait for the talks in the morning ( I can't accurately say 'tomorrow' any more!)

No comments: