HPC4E presents three papers at ICCS 2017

HPC4E researchers Victor Martinez (Ph.D. student at UFRGS) and Maciej Paszyński (affiliated professor at AGH University of Science and Technology and visiting researcher at LNCC) presented three papers in the International Conference on Computational Science (ICCS) held on Zürich (Switzerland) from 12 to 14 June 2017.

Martínez (UFRGS) presented two papers, available by following these links:

  • Using Power Demand and Residual Load Imbalance in the Load Balancing to Save Energy of Parallel Systems
    Power consumption of the High Performance Computing (HPC) systems is an increasing concern as large-scale systems grow in size and, consequently, consume more energy. In response to this challenge, we have develop and evaluate new energy-aware load balancers to reduce the average power demand and save energy of parallel systems when scientific applications with imbalanced load are executed. Our load balancers combine dynamic load balancing with DVFS techniques in order to reduce the clock frequency of underloaded computing cores which experience some residual imbalance even after tasks are remapped. The results show that our load balancers present power reductions of 7.5% in average with the fine-grained variant that performs per-core DVFS, and of 18.75% with the coarse-grained variant that performs per-chip DVFS over real applications.
    Download full text (PDF)
  • Performance Improvement of Stencil Computations for Multi-core Architectures based on Machine Learning
    Stencil computations are the basis to solve many problems related to Partial Differential Equations (PDEs). Obtaining the best performance with such numerical kernels is a major issue as many critical parameters (architectural features, compiler flags, memory policies, multithreading strategies) must be finely tuned. In this context, auto-tuning methods have been extensively used to improve the overall performance. However, the complexity of current architectures and the large number of optimizations to consider reduce the efficiency of this approach. This paper focuses on the use of Machine Learning to predict the performance of stencil kernels on multi-core architectures. Low-level hardware counters (e.g. cache-misses and TLB misses) on a limited number of executions are used to build our predictive model. We have considered two different kernels (7-point Jacobi and seismic wave modelling) to demonstrate the effectiveness of our approach. Our results show that performance can be predicted and that the best input configuration for stencil problems can be obtained by simulations of hardware counters and performance measurements.
    Download full text (PDF)
    Download presentation (PDF)

On the other hand, Paszyński (LNCC) presented a third paper:

  • A wrapper around parallel MUMPS solver to reduce its memory usage and execution time for finite element method computations

    In this paper, we present a wrapper around MUMPS solver, called Hierarchical Solver Wrapper (HSW), that is tailored to domain decomposition-based parallel finite element method computations on distributed memory systems. It offers the same interface as parallel MUMPS with matrix entries in coordinate format provided in a distributed fashion among multiple processors. The algorithm implemented by the wrapper utilizes multiple sequential instances of MUMPS solver to compute Schur complements over subdomains. Next, it deallocates sequential MUMPS solvers and LU factors, and it calls the parallel MUMPS solver feeded with the Schur complements, stored in distributed manner. In the backward substitution stage it recomputes the local LU factors before solving the local problems. The wrapper has been tested with three-dimensional isogeometric analysis computations, and we show it reduces both the memory usage and the execution time, in comparison with a single parallel MUMPS call.
    Download full text (PDF)

The ICCS is an annual conference that brings together researchers and scientists from mathematics and computer science as basic computing disciplines, researchers from various application areas who are pioneering computational methods in sciences such as physics, chemistry, life sciences, and engineering, as well as in arts and humanitarian fields, to discuss problems and solutions in the area, to identify new issues, and to shape future directions for research, as explained in the event's website. The theme for ICCS 2017 was “The Art of Computational Science. Bridging Gaps - Forming Alloys”, to highlight the role of computational science as a broad, multidisciplinary science tackling problems and creating synergies with other fields.

Further information:

Link icon @iccs_conf | #ICS2017