Modelado supported a Workshop and a Birds-of-a-Feather session at the International Supercomputing Conference, June 19-23, 2016 in Frankfurt, Germany.
Both meetings discussed dynamic runtimes.
The BoF presented recent progress on the Open Community Runtime (OCR) and the workshop followed up later in the week as a community building session on cooperation across the Atlantic. The complete set of presentations are archived through the highlighted links.
Abstract: Exascale and extreme-scale systems impose fundamental new requirements on software to target platforms with severe energy, data movement and resiliency constraints within and across nodes, and large degrees of homogeneous and heterogeneous parallelism and locality within a node. The Open Community Runtime (OCR) was created to foster discussions on how best to address these issues.
This BoF included a presentation of the current state of OCR and a discussion of next steps and collaboration opportunities and presented the motivation and current state of the various OCR implementations as well as concrete use cases. We discussed how to get started with OCR and engaged with the audience in discussing future directions and collaboration opportunities. OCR’s goal is indeed to build and foster a community around the concepts expressed in OCR to better accelerate innovations in the exascale software stack.
Workshop: Building a European/American Community for the Development of Dynamic Runtimes in Extreme-Scale Systems (pdf)
Abstract: Three main challenges face the high performance computing community on the road to Exascale:
1) Exploitation of extreme parallelism and the concurrent difficulty of programming,
2) Regular hardware and software failures of these very high part-count machines requiring adaption to frequent failures incidents, and
3) Severe limits on power consumption in order to fit within datacenters’ capacity and budgets.
Dynamic runtimes have emerged as tools to manage quickly changing circumstances. Several European and American projects have developed prototype systems that share some common characteristics, but there is no overall consensus on the exact needs and requirements of systems that will eventually be implemented in the extreme scale systems of the next decade. The workshop will bring together the developers of several systems to discuss the building of a collaborative community.
In this workshop the developers of the OmpSs programming model from the Barcelona Supercomputing Center, the StarPU system from INRIA, the Open Community Runtime from Rice University and University of Vienna and researchers from the American Department of Energy projects including HPX, DARMA and Legion addressed these challenges and identified common threads upon which to build future collaboration across the Atlantic.