Time | Event |
---|---|
Full Day |
|
13:00-17:00 |
Time | Event |
---|---|
08:30-09:00 |
Opening Remarks (Pavan Balaji, Bill Gropp, Rajeev Thakur, Kathryn Mohror) |
09:00-10:00 |
Keynote Session Session Chair: Bill Gropp Rick Stevens |
10.00-10.30 |
Morning Break |
10.30-12.00 |
Paper Session 1: Collectives Session Chair: Rajeev Thakur Practical, Linear-time, Fully Distributed Algorithms for Irregular Gather and Scatter Enabling Hierarchy-aware MPI Collectives in Dynamically Changing Topologies Transforming Blocking MPI Collectives to Non-blocking and Persistent Operations |
12.00-13.30 |
Lunch Break |
13:30-14:30 |
Paper Session 2: Persistent Collectives Session Chair: Jesper Larsson Träff Planning for Performance: Persistent Collective Operations for MPI Offloaded MPI Persistent Collectives using Persistent Generalized Request Interface |
14:30-15:00 |
Industry Session Maria Garzaran How Intel is Enabling MPI for Current and Future Architectures |
15:00-15:30 |
Afternoon Break |
15:30-16:15 |
Poster Presentations |
16:15-18:00 |
Poster Session and Reception |
Time | Event |
---|---|
09:00-10:00 |
Keynote Session Session Chair: Kathryn Mohror Michela Taufer |
10:00-10:30 |
Morning Break |
10:30-12:00 |
Paper Session 3: Tools and Simulation Session Chair: Martin Schulz Verification of MPI Programs using CIVL Using Software-Based Performance Counters to Expose Low-Level Open MPI Performance Information Characterizing MPI Matching via Trace-based Simulation |
12:00-13:30 |
Lunch Break |
13:30-15:00 |
Paper Session 4: Memory and Topology Session Chair: Antonio Peña A Hierarchical Model to Manage Hardware Topology in MPI Applications Enhanced Memory Management for Scalable MPI Intra-node Communication on Many-core Processor Improving the memory access locality of hybrid MPI applications |
15:00-15:30 |
Afternoon Break |
15:30-17:00 |
Panel Session: MPI on Post-Exascale Systems Panelists: Bill Gropp (Moderator), Atsushi Hori, Martin Schulz, and Michela Taufer |
18:00-22:00 |
Time | Event |
---|---|
09:00-10:00 |
Keynote Session Session Chair: Pavan Balaji Ron Brightwell |
10:00-10:30 |
Morning Break |
10:30-12:00 |
Paper Session 5: MPI Usage and Infrastructure Session Chair: Rolf Rabenseifner Notified Access in Coarray Fortran What does fault tolerant Deep Learning need from MPI? Ralph Castain, David Solt, Joshua Hursey, and Aurelien Bouteiller |
12:00-13:30 |
Lunch Break |
13:30-15:00 |
Paper Session 6: Best Papers Session Chair: Pavan Balaji MPI Windows on Storage for HPC Applications MPI Performance Engineering with the MPI Tool Interface: the Integration of MVAPICH and TAU |
15:00-15:30 |
Closing |
Despite claims that MPI would never be effective for first petascale and now exascale systems, MPI is being used effectively on petascale systems and remains the internode programming model for the upcoming pre-exascale systems. Furthermore, MPI is likely to be a key part of the programming environment for exascale systems. This panel will address the question of whether MPI can continue to be the programming model for extreme-scale systems, or whether post-exascale systems will require either significant additions to MPI or need new programming systems to address both the capabilities of post-exascale systems as well as changing needs of applications. What should change in MPI? What changes could broaden the applicability of MPI? How would those changes impact the use of MPI on average-sized parallel computers?
The EuroMPI/USA 2017 proceedings can be obtained from the ACM Digital Library.