Editorial for the special issue on operating systems and programming systems for HPC
- PDF / 470,159 Bytes
- 2 Pages / 595.276 x 790.866 pts Page_size
- 98 Downloads / 234 Views
EDITORIAL
Editorial for the special issue on operating systems and programming systems for HPC Xiaobing Feng1 · Minyi Guo2 Published online: 13 November 2020 © China Computer Federation (CCF) 2020
With the coming of exascale computing era, programming systems and operating systems (including runtime systems) are facing several challenges. In aspect of architecture, increasing deeper level of parallelism, heterogeneity, and the adoption of diverse domain specific accelerators raise the urgent need for programmability, performance optimization and portability. On the other side, big data analytics and machine learning applications demand to be ported and optimized on modern HPC systems. This issue focuses on the novel ideas, methods, as well as efforts of system software development for resolving the above challenges, and to fill the gap between applications and the underlying hardware systems. We have eight invited papers selected for this special issue based on a peer-review procedure, which cover a number of different aspects that relate to programming systems, operating or runtime systems challenges mentioned above. The first part of the special issue focuses on the improvements of programming systems for contemporary large scale HPC systems. We have four papers that discuss programming system innovations covering traditional HPC applications and deep learning area, tackling inter-node parallel scalability and intra-node processor heterogeneity, addressing user programmability and performance challenges. • The first paper written by Li Chen et al. presents
AceMesh, a task-based data-driven programming language targeting legacy MPI applications. The new language features not only relieve the programmer from
* Xiaobing Feng [email protected] Minyi Guo [email protected] 1
State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China
Shanghai Jiaotong University, Shanghai, China
2
tedious refactoring efforts but also provide possibility for structured execution of complex task graphs, data locality exploitation, and less runtime overhead. Related compiling and runtime optimizations are also presented, and evaluation on two supercomputing platforms shows its performance superiority to existing programming models. • The second paper written by Libo Zhang et al. proposes an automatic mapping technique for OpenACC kernel codes on heterogeneous, deeply fused many-core architecture. Static compiling analysis is integrated with dynamic feedbacks. Experimental results show that the approach gets similar performance to the manual annotated approach. • The third paper written by Zihan Liu et al. focuses on how to exploit the performance potential of deep learning inference accelerators in a compiler tool chain. It introduces an operator level SDK for a specialized hardware accelerator (Cambricon) and proposes a middle layer compiler tool-chain. The tool chain not only provides enough abstraction level, but also exposes major optimization knobs. Experiment
Data Loading...
