Optimizations to enhance sustainability of MPI applications

Ultrascale computing systems are likely to reach speeds of two or three orders of magnitude greater than today's computing systems. However, to achieve this level of performance, we need to design and implement more sustainable solutions for ultra-scale computing systems, at both the hardware and software levels, while understanding sustainability in a holistic manner in order to address challenges in economy-of-scale, agile elastic scalability, heterogeneity, programmability, fault resilience, energy efficiency, and storage. Some solutions could be integrated into MPI, but others should be devised as higher level concepts, less general, but adapted to applicative domains, possibly as programming patterns or libraries. In this paper, we layout some proposals to extend MPI to cover major relevant domains in a move towards sustainability, including: MPI programming optimizations and programming models, resilience, data management, and their usage for applications.

[1]  Thomas Hérault,et al.  An evaluation of User-Level Failure Mitigation support in MPI , 2012, Computing.

[2]  Polyvios Pratikakis,et al.  DRASync: distributed region-based memory allocation and synchronization , 2013, EuroMPI.

[3]  Martin Schulz,et al.  The myrmics memory allocator: hierarchical,message-passing allocation for global address spaces , 2012, ISMM '12.

[4]  Jesús Carretero,et al.  A Scalable Message Passing Interface Implementation of an Ad-Hoc Parallel I/o system , 2010, Int. J. High Perform. Comput. Appl..

[5]  Jesús Carretero,et al.  FLEX-MPI: An MPI Extension for Supporting Dynamic Load Balancing on Heterogeneous Non-dedicated Systems , 2013, Euro-Par.

[6]  Qian Zhang,et al.  Accelerated Conformational Entropy Calculations Using Graphic Processing Units , 2013, J. Chem. Inf. Model..

[7]  Jesús Carretero,et al.  Dynamic-CoMPI: dynamic optimization techniques for MPI parallel applications , 2010, The Journal of Supercomputing.

[8]  Thomas Hérault,et al.  Assessing the Impact of ABFT and Checkpoint Composite Strategies , 2014, 2014 IEEE International Parallel & Distributed Processing Symposium Workshops.

[9]  Alexey L. Lastovetsky,et al.  Two Algorithms of Irregular Scatter/Gather Operations for Heterogeneous Platforms , 2010, EuroMPI.

[10]  Thomas Fahringer,et al.  A multi-objective auto-tuning framework for parallel codes , 2012, 2012 International Conference for High Performance Computing, Networking, Storage and Analysis.

[11]  Alexey L. Lastovetsky,et al.  Hierarchical approach to optimization of parallel matrix multiplication on large-scale platforms , 2015, The Journal of Supercomputing.