Performance Analysis of a Quantum Monte Carlo Application on Multiple Hardware Architectures Using the HPX Runtime

Wei, Weile; Chatterjee, Arghya; Huck, Kevin; Hernández, Óscar; Kaiser, Hartmut

doi:10.1109/scala51936.2020.00015

Cited by 6 publications

(2 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…al. [4] reported that DCA++ with HPX user-level [11] threading support achieves a 20% speedup over the original C++ threading (kernel-level) due to faster context switching in HPX threading.…”

Section: Concurrency Overlappingmentioning

confidence: 99%

See 1 more Smart Citation

Memory Reduction using a Ring Abstraction over GPU RDMA for Distributed Quantum Monte Carlo Solver

Wei

D'Azevedo

Huck

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Scienti c applications that run on leadership computing facilities often face the challenge of being unable to t leading science cases onto accelerator devices due to memory constraints (memorybound applications). In this work, the authors studied one such US Department of Energy missioncritical condensed matter physics application, Dynamical Cluster Approximation (DCA++), and this paper discusses how device memory-bound challenges were successfully reduced by proposing an e ective "all-to-all" communication method-a ring communication algorithm. This implementation takes advantage of acceleration on GPUs and remote direct memory access (RDMA) for fast data exchange between GPUs. Additionally, the ring algorithm was optimized with sub-ring communicators and multi-threaded support to further reduce communication overhead and expose more concurrency, respectively. The computation and communication were also analyzed by using the Autonomic Performance Environment for Exascale (APEX) pro ling tool, and this paper further discusses the performance trade-o for the ring algorithm implementation. The memory analysis on the ring algorithm shows that the allocation size for the authors' most memory-intensive data structure per GPU is now reduced to 1/𝑝 of the original size, where 𝑝 is the number of GPUs in the ring communicator. The communication analysis suggests that the distributed Quantum Monte Carlo execution time grows linearly as sub-ring size increases, and the cost of messages passing through the network interface connector could be a limiting factor.

show abstract

Section: Concurrency Overlappingmentioning

confidence: 99%

“…Dynamical Cluster Approximation (DCA++) 2 is a high-performance research software application [1,2,3,4] that provides a modern C++ implementation to solve quantum many-body problems. DCA++ implements a quantum cluster method with a Quantum Monte Carlo (QMC) kernel for modeling strongly correlated electron systems.…”

Section: Introductionmentioning

confidence: 99%