Topic outline
-
Most HPC systems are clusters of shared memory nodes. To use such systems efficiently both memory consumption and communication time has to be optimized. Therefore, hybrid programming may combine the distributed memory parallelization on the node interconnect (e.g., with MPI) with the shared memory parallelization inside of each node (e.g., with OpenMP or MPI-3.0 shared memory). This course analyzes the strengths and weaknesses of several parallel programming models on clusters of SMP nodes. Multi-socket-multi-core systems in highly parallel environments are given special consideration. MPI-3.0 has introduced a new shared memory programming interface, which can be combined with inter-node MPI communication. It can be used for direct neighbor accesses similar to OpenMP or for direct halo copies, and enables new hybrid programming models. These models are compared with various hybrid MPI+OpenMP approaches and pure MPI. Numerous case studies and micro-benchmarks demonstrate the performance-related aspects of hybrid programming.
Hands-on sessions are included on all days. Tools for hybrid programming such as thread/process placement support and performance analysis are presented in a "how-to" section. This course provides scientific training in Computational Science and, in addition, the scientific exchange of the participants among themselves.
This course is a joint training event of EuroCC@GCS and EuroCC-Austria, the German and Austrian National Competence Centres for High-Performance Computing. It is organized by the HLRS in cooperation with the VSC Research Center, TU Wien and NHR@FAU.
Agenda & Content
1st day – 23 January 2024
08:45 Join in
09:00 Welcome
09:03 Exascale Supercomputing is coming to Stuttgart: Hunter and Herder
Prof. Michael Resch (Director, HLRS)
09:30 Hunter's hardware architecture and its programming models
Dr. Christian Simmendinger (HPE) and Igor Pasichnyk (AMD)
10:15 Break
10:30 Introduction to the general MPI+X course with exercises on existing Hawk cluster
11:15 Programming Models
11:20 - MPI + OpenMP
11:50 Practical (how to compile and start)
12:30 Lunch
14:00 - continue: MPI + OpenMP
14:45 Break
15:00 - continue: MPI + OpenMP
15:45 Practical (how to do pinning)
16:15 Q & A
16:30 End of first day2nd day – 24 January 2024
08:45 Join in
09:00 - continue: MPI + OpenMP
09:00 Practical (hybrid through OpenMP parallelization)
10:30 Break
10:45 - Overlapping Communication and Computation
11:15 Practical (taskloops)
12:15 - MPI + OpenMP Conclusions
12:30 Lunch
14:00 - MPI + Accelerators
15:00 Break
15:15 - MPI + Accelerators (continued)
16:15 Q & A
16:30 End of second day3rd day – 25 January 2024
08:45 Join in
09:00 Programming Models (continued)
09:05 - MPI + MPI-3.0 Shared Memory
10:00 Break
10:15 - MPI Memory Models and Synchronization
11:00 Break
11:15 - Pure MPI
11:35 - Recap - MPI Virtual Topologies
12:05 - Topology Optimization
12:30 Lunch
14:00 - Topology Optimization, continued
14:10 Conclusions
14:25 Practical (replicated data)
15:45 Q & A, Feedback
16:00 Q & A
16:30 End of third day (course)Date: Tuesday, January 23, 2024, 08:45 - Thursday, January 25, 2024, 16:30 Location: HLRS, Room 0.439 / Rühle Saal, University of Stuttgart, Nobelstr. 19, D-70569 Stuttgart, Germany
Lecturers:
Rolf Rabenseifner (HLRS), Claudia Blaas-Schenner (VSC Team, TU Wien), Georg Hager (RRZE)
Course material (here ☺):
http://tiny.cc/MPIX-HLRS