Scalable and high-performance support for PGAS and hybrid MPI+PGAS programming models on modern clusters: opportunities and challenges

Date:

2 October 2013

Venue:

Room 6301 The James Clerk Maxwell Building

The King's Buildings

The University of Edinburgh

See http://www.ed.ac.uk/maps/

Registration:

To register, send your contact details to us by email

Programme:

The tutorial will run from 1:30pm to 5:00pm (including a coffee break from 3:00-3:30pm.)

The outline of the tutorial and what will be discussed is as follows:

1: Overview of the Modern HPC System Architectures

Multi-core Processors

High Performance Interconnects (InfiniBand, 10GigE/iWARP and  RDMA over Converged Enhanced Ethernet (RoCE))

Heterogeneity with Accelerators (GPUs) and Coprocessors (Xeon Phis)

2:  Introduction to Partitioned Global Address Space Models

Language-based Models: Case Study with UPC

Libraries-based Models: Case Study with OpenSHMEM

3: Overview of MPI+PGAS Hybrid Programming Models and Benefits

4: Designing Scalable and High Performance Support for UPC and OpenSHMEM on Modern Clusters

Multi-core Nodes

Clusters with High Performance Interconnects

Heterogeneous Clusters with GPUs and Xeon Phis

5: Challenges in Designing Unified Runtime for Hybrid MPI+PGAS  Programming Models

6: Application-level Case Studies and Performance Evaluation

7: Opportunities for Future Extensions and Enhancements

8: Conclusion and Q&A