Optimizing Nextflow-based Software on Shared HPC Resources: A Case Study with make_lastz_chains

Nextflow is a widely adopted workflow manager in the bioinformatics community, known for its scalability, portability, and reproducibility. However, on shared HPC clusters that use the Slurm job scheduler and the Fairshare score to record historical resource usage and determine current job queuing positions, individual Nextflow job submissions negatively impact user Fairshare scores and lead to extended queue wait times. In this paper, we share practical observations from supporting researchers running the Hiller Lab make_lastz_chains pipeline, which uses Nextflow to orchestrate genome alignment with LASTZ and related UCSC tools, on the Arizona State University supercomputers. We identify key challenges and solutions regarding scheduling, Fairshare impact, and capturing Slurm errors. We would like to share these observations and practical considerations with researchers, RSEs, and HPC system administrators in order to improve management of Nextflow workflows on shared HPC resources, foster more efficient resource utilization, and a smoother user experience.