Program

Please note: some sessions of EuroMPI 2019 may be video-recorded, and video recordings may be made in the keynote and panel sessions and common areas as well. By attending the keynote and panel sessions, you are
granting EuroMPI 2019 permission to record your image, comments, and questions during sessions.

 

September 10
(additional registration required)

09:00 – 12:00
Tutorial
Konstantinos Parasyris (Barcelona Supercomputing Center)
Multilevel Checkpointing for MPI Applications
 LEE C 104
12:00 - 14:00 Catered lunch CLA Glashalle
14:00 - 17:00
Tutorial
Tal Ben Nun (ETH Zurich)
Reproducible Parallel and Distributed Deep Learning with Deep500
 LEE C 104
September 11
08:45 - 09:00 Opening Remarks, Torsten Hoefler and Jesper Larsson Traeff HG E 1.2
09:00 - 10:00
Keynote - chaired by Torsten Hoefler
Brian Van Essen
Scalable Deep Learning for Scientific Computing
 HG E 1.2
10:00 - 10:30 Coffee Break HG E Nord Foyer
10:30 - 12:00
Paper Session 1: Message-Passing and PGAS Interfaces: Part 1 - chaired by Bill Gropp
P. V. Bangalore, R. Rabenseifner, D. Holmes, J. Jaeger, G. Mercier, C. Blaas-Schenner, A. Skjellum.
Exposition, Clarification, and Expansion of MPI Semantic Terms and Conventions: Is a nonblocking MPI function permitted to block?
S. Rivas-Gomez, A. Fanfarillo, S. Narasimhamurthy, S. Markidis.
Persistent Coarrays: Integrating MPI Storage Windows in Coarray Fortran
Gustavo Alonso
Invited talk: Building a large scale database on top of MPI
HG E 1.2
12:00 - 13:30 Catered Lunch HG E Nord Foyer
13:30 - 15:00
Paper Session 2: Message-Passing and PGAS Interfaces: Part 2 - chaired by Daniel Holmes
B. Elis, D. Yang, M. Schulz.
QMPI: A Next Generation MPI Profiling Interface for Modern HPC Platforms
M. Sergent, C. T. Aitkaci, P. Lemarinier, G. Papaure.
Efficient Notifications for MPI One-Sided Applications
C. Niethammer, R. Rabenseifner.
An MPI interface for application and hardware aware Cartesian topology optimization
HG E 1.2
15:00 - 15:30 Coffee Break HG E Nord Foyer
15:30 - 16:15 Poster Presentations HG E 1.2
16:15 - 18:00 Poster Session with drinks and snacks HG E Nord Foyer
September 12
09:00 - 10:00
Keynote - chaired by Stefano Markidis
Mathieu Luisier
Massively Parallel Computer Aided Design of Nano- Transistors: When Physics Lets You Down
 HG E 1.2
10:00 - 10:30 Coffee Break HG E Nord Foyer
10:30 - 12:00
Paper Session 3: Applications of and for MPI - chaired by Stefano Markidis
A. Castello, M. F. Dolz, E. S. Quintana-Orti, Duato.
Analysis of Model Parallelism for Distributed Neural Networks
K. Dichev, D. Nikolopoulos.
Implementing Efficient Message Logging Protocols as MPI Application Extensions
A. Y. Polyakov, B. I. Karasev, J. Hursey, J. Ladd, M. Brinskii, E. Shipunova.
A Performance Analysis And Optimization Of PMIx-based HPC Software Stacks
HG E 1.2
12:00 - 13:30 Catered Lunch HG E Nord Foyer
13:30 - 15:00
Paper Session 4: Performance, Implementations, Hardware: Part 1 - chaired by Martin Schulz
J. Besnard, J. Jaeger, A. D. Malony, S. Shende, H. Taboada, M. Perache, P. Carribault.
Mixing Ranks, Tasks, Progress and Non-Blocking Collectives
N. Islam, G. Zheng, S. Sur, A. Langer, M. Garzaran.
Minimizing the Usage of Hardware Counters for Collective Communication using Triggered Operations
S. Levy, K. B. Ferreira.
Evaluating Tradeoffs Between MPI Message Matching Offload Hardware Capacity and Performance
HG E 1.2
15:00 - 15:30 Coffee Break HG E Nord Foyer
15:30 - 17:00

Panel chaired by Bill Gropp
Experts: Maria Garzaran, Stefano Markidis, Ron Brightwell

Designing MPI in 2020

MPI was designed over 25 years ago. In that era, computers required multiple chips for each processor; the ratio of memory latency to processor clock speed was far different, as was the ratio of floating point performance to memory and interconnect bandwidth. There has also been a tremendous change in how software is designed and used, with backward compatibility no longer required in many cases, and the lifetime of much (but not all) software now measured in months or years instead of decades. What would MPI look like if it were designed today? This panel will discuss some of the trends of current parallel computer architecture and software and how they might shape the design of a parallel programming system.

 

 HG E 3
17:00 - 17:15 EuroMPI 2020 HG E 3
17:15 - 18:00 Poster Session HG E Nord Foyer
18:00 - 18:15 Group photo Polyterrace
18:15 - 22:00
Social Event

We will take a relaxing walk from the ETH to the old town of Zurich. There, Mr. Sepp Wimmer from the Zunfthaus Zur Waag greets us with an aperitif at the wine fountain or in the cloister of the Münsterhof (depending on the weather). Afterwards we will have dinner at the Zunfthaus Zur Waag with many Swiss specialties.

 From Polyterrace to Zunfthaus Zur Waag
September 13
09:00 - 10:00
Keynote - chaired by Jesper Larsson Traeff
Ron Brightwell
Opportunities and Challenges for Accelerated Network Interfaces in HPC
 HG E 1.2
10:00 - 10:30 Coffee Break HG E Nord Foyer
10:30 - 12:00
Paper Session 5: Performance, Implementations, Hardware: Part 2 - chaired by Ron Brightwell
W. P. Marts, M. G. F. Dosanjh, W. Schonbein, R. E. Grant, P. G. Bridges.
MPI Tag Matching Performance on ConnectX and ARM
D. Zhong, A. Bouteiller, X. Luo, G. Bosilca.
Runtime Level Failure Detection and Propagation in HPC Systems
Martin Schulz
Invited Talk: MPI 4.0 is coming! What’s next?
HG E 1.2
12:00 - 13:30 Catered Lunch HG E Nord Foyer
13:30 - 14:15
Edmund Preiss (Intel)
Software Development Tools for distributed and cross platform HPC environments
 HG E 1.2
14:15 - 14:45 Closing Remarks HG E 1.2