DL_POLY - A performance overview analysing, understanding and exploiting available HPC technology

Abstract

This paper considers the performance attributes of the molecular simulation code, DL_POLY, as measured and analysed over the past two decades. Following a brief overview of HPC technology, and the performance improvements over that period, we define the benchmark casesfor both DL_POLY Classic and DL_POLY 3 & 4used in generating a broad overview of performance across well over 100 HPC systems, from the Cray T3E/1200 to today's Intel Skylake clusters and accelerator technology offerings from Intel's Xeon Phi co-processor family. Consideration is given to the tools that have proved helpful in analysing the code's performance. With a more rigorous analysis of performance on recent systems, we discuss the optimum choice of processor and interconnect, and present power measurements when running the code, comparing these to measurements for other community codes.

DL_FFLUX: A Parallel, Quantum Chemical Topology Force Field

“…The reason that DL_FFLUX scales worse than DL_POLY in the L ′ = 0 case is once again due to the redundant work when predicting point charges. Note that the scalability of the MPI in various versions of DL_POLY has been tested extensively in the past. , …”

“…Note that the scalability of the MPI in various versions of DL_POLY has been tested extensively in the past. 56,57 Profiling also gives us some insight into the percentage of the total runtime spent in DL_FFLUX and DL_POLY routines (as well as MPI-specific routines), which is presented in Figure 16 as pie charts. For the sake of clarity, just the profiles for N p = 1, 8, and 36 are shown; the rest of the pie charts appear in Figure S9 as well as a more detailed "sample" breakdown in Figure S10 showing the relative timings of the top 5 most costly subroutines.…”

Interatomic potentials for oxide glasses: Past, present, and future

“…The PIM approach has not been widely applied for the simulation of oxide glasses due to its complex nature which makes it very complicated to be parameterized and it has not been extended to many elements, lacking in transferability. Furthermore, the majority of the most used MD programs (Dl_Poly [150], LAMMPS [151], …) do not allow to use this model. To our knowledge, the applications were limited to the simulation of SiO 2 [58,152], GeO 2 [153], B 2 O 3 [154] , and Na-borosilicate [149] amorphous systems leading to a good computation of vibrational spectra and allowing to stabilize the formation of boroxol rings in B 2 O 3 , producing ~33% of rings, better than other force fields.…”

On the performance of a highly-scalable Computational Fluid Dynamics code on AMD, ARM and Intel processors

“…However, improvements in code performance will of course depend on the end-user application, e.g. a classical molecular dynamics code may be compute-bound [7] whilst CFD codes are typically memory-bound [8]. This calls for heterogeneous HPC systems that comprise a variety of chip architectures, allowing users to maximise the performance of their specific computing applications Gray et al [9].…”