56 / 101
Biophysics in the Understanding, Diagnosis, and Treatment of Infectious Diseases Poster Abstracts
51
5-POS
Board 5
JMS: Creating and Running Complex Computational Pipelines on High Performance
Computer Clusters
David Brown
, David Penkler, Thommas Musyoka, Özlem Tastan Bishop.
Rhodes University, Grahamstown, South Africa.
Modern computing has enabled research that was previously considered unfeasible. Parallel
algorithms have been developed to run over powerful multicore machines. For even more
computing power, these machines can be aggregated together into large high performance
computing (HPC) clusters. On these clusters, jobs can be spread out across a large number of
nodes instead of being executed on a single machine. This can substantially decrease the time
required to execute resource intensive modeling and simulation jobs – a common requirement in
the field of biophysics. It is also useful when a large number of much smaller jobs need to be
executed. Unfortunately, running jobs on a cluster involves a steep learning curve. Jobs must be
submitted via software systems known as resource managers. These systems are usually run via
the command line and require expertise that most researchers do not have. To solve this problem,
we have developed JMS (Job Management System), a web-based front-end to an HPC cluster.
JMS allows users to run, manage and monitor jobs via a user-friendly web interface. It also lets
users create new tools that can be pipelined together with existing tools to create complex
computational workflows. These workflows can be saved, versioned and reused as needed. All
tools, workflows and jobs can be shared with other users to create a highly collaborative work
environment. In addition, tools and workflows can be made public via external interfaces.
Although applicable to any field, JMS is currently being tailored toward structural bioinformatics
with the introduction of tools and workflows for homology modelling, docking studies, and
molecular dynamics. JMS has been open-sourced and is freely available at
https://github.com/RUBi-ZA/JMS.