*** NEW: This workshop has been cancelled this instance.
AboutMessage-passing programming models have dominated high-performance computing (HPC) for the past quarter century. With the continued breakdown of the (uniform) communicating sequential processors abstract machine model, many researchers have questioned the continued viability of message passing as a model for direct, application-level interaction. Recent years have seen an explosion of new paradigms in programming models for distributed-memory computing, ranging from evolutionary to revolutionary. With this paradigm shift, it has become clear that raw performance is not the only concern in the design of a programming model; rather, for many applications the overall time-to-solution is just as important. The move to the next generation of HPC platforms presents a wider variety of challenges than ever before. Potential challenges to address include the increased need for asynchrony, increased heterogeneity and performance nonuniformity, decreased hardware reliability and increased failure rates, increased hardware diversity, and the maintainability of increasingly complex scientific code bases. Alternative programming models that address some or all of these challenges include task-based programming models (and a variety of higher-level programming models that generate task-based execution), distributed-memory actor models, extensions and augmentations to the existing message-passing paradigm (such as active messages), and partitioned global address space models (implemented at a broad range of abstraction levels).
This workshop proposes to promote a dialogue about the spectrum from evolutionary to revolutionary programming models and the relative importance of the concerns these new approaches address. Dialogues we propose to encourage include: the right level of abstraction for application developers to interact with; the importance of performance versus productivity when developing both rapid prototypes and production-scale applications; the relative importance of fault-tolerance and the potential for nonuniformity in future hardware; the feasibility of obtaining performance without machine-specific implementation (and the importance of doing so); and the viability of evolving or replacing current programming models in production-scale scientific applications. In addition to having a dialogue, a goal of this workshop is also to expose researchers who work on extensions or augmentations to message-passing models (the “evolutionary” crowd) to alternative programming models and to expose those who develop fundamentally different programming models (the “revolutionary” crowd) to the recent advances in message-passing programming models. The workshop seeks to garner a better understanding of key motivations for alternative programming models, key insights from advances in message passing (and how they can be applied to alternative models), and key opportunities for collaboration between researchers on both sides.
- Performance nonuniformity
- Fault-tolerance and resilience
- Performance portability: a posteriori adaptability (“one code for many machines”) in situ adaptability (reactive and dynamic runtime system solutions)
- Maintainability (particularly in the presence of these other challenges)
- Testability and debuggability (particularly in the presence of these other challenges)
- Experimental comparative results (particularly with emphasis on overall time-to-solution)
- Relevance of alternate programming models to particular scientific kernels
- Provision for power/energy/temperature management
SubmissionInterested authors are encouraged to submit full papers (8-10 pages) or short/position papers (4 pages) "sigconf" style in the ACM 2017 template. This page limit includes figures, tables and appendices but not references. Authors should submit their work at https://easychair.org/conferences/?conf=pmamp1.
Selected papers will be invited to submit revised and extended versions to a special edition of International Journal of Parallel Programming (IJPP).
- Paper submission: July 1, 2017
- Author Notification: July 30, 2017
- Camera-ready papers due: August 15, 2017
- Workshop Date: September 25, 2017
- Abhinav Bhatele, Lawrence Livermore National Laboratory
- Alex Aiken, Stanford
- Jesper Larsson Träff, TU Wien
- Karl Fuerlinger, LMU Munich
- Ron Brightwell, Sandia National Laboratory
- Zoran Budimlic, Rice University
- Hartmut Kaiser, Louisiana State University
- Sean Treichler, Nvidia
- Bryce Adelstein Lelbach, Lawrence Berkeley National Laboratory
- Martin Schulz, Lawrence Livermore National Laboratory
OrganizersDavid S Hollman, Sandia National Laboratory
Murali Emani, Lawrence Livermore National Laboratory
Contact firstname.lastname@example.org for any questions