Engineering a better, faster, cheaper mousetrap is a really hard problem. We present a case study on building a cheaper computational mouse trap, UW-Biostatistics infrastructure done on a shoestring budget. Additionally, we show how closely collaborating with researchers in experiment design helps determine use case needs and results in WIN-WINs for researchers and computing staff.

LIGO recently announced the first ever detection of gravitational waves as predicted by Albert Einstein's Theory of Relativity. However, before the announcement could be made public, LIGO scientists needed to do a rigorous analysis of the data. As a part of this analysis, a lot of computational pipelines were executed to analyze the data collected from the instruments. One of the major pipelines employed by LIGO to measure the statistical significance of data needed for discovery was the Advance LIGO pyCBC pipeline. This workflow is exclusively managed by Pegasus and used by LIGO to execute on LIGO Data Grid, Open Science Grid, and XSEDE clusters. The Pegasus group has been working closely with LIGO scientists since 2001. The relationship has been mutually beneficial, with a lot of advances in Pegasus occurring as a result of feedback from LIGO and their requirements. The talk will give an overview of the pipeline, the challenges LIGO encountered as they scaled up, and how Pegasus helped address them.

The Waisman Center at UW-Madison is an on-campus, multi-departmental institution focused on the investigation of developmental disabilities. DIPA is a Python and HTCondor-based pipeline in development at the Waisman Center Brain Imaging Core for the processing and analysis of diffusion-weighted MRI scans (dMRI). These scans are useful tools in determining the integrity and structure of brain white matter (the "highways" of the brain). At the Waisman Center, dMRI scans are obtained from a wide range of individuals, ranging from infants to participants in clinical trials or longitudinal studies. The goal of DIPA is to provide imaging scientists a way to process images that is efficient, easy, and consistent, but flexible enough for the wide variety of studies and analyses. Currently DIPA uses a Python program to establish a set of HTCondor submit files and DAGs which are then also submitted to an HTCondor pool and monitored. In development are improvements that will allow dynamic workflow adjustments based on quality control metrics. Pegasus integration is also a goal for DIPA, but it is currently only at an early stage.