Topic 10: Parallel Numerical Algorithms

Since the early days of supercomputing, numerical routines have caused the highest demand for computing power anywhere, making their ecient paral-lelization one of the core methodical tasks in high-performance computing. And still, many of today's fastest computers in the world are mostly used for the solution of huge systems of equations as they arise in the simulation of complex large scale problems in engineering and science. Despite this long tradition, parallel numerical algorithms did not lose anything of their relevance. The ecient implementation of existing schemes on state-of-the-art parallel systems (such as clusters or hybrid systems), the challenges resulting from massively parallel systems, the design of easy-to-use portable software components, the recent endeavours to tackle optimization, control, and interactive steering scenarios, too – all this clearly shows that progress in computational science and engineering strongly depends on progress with parallel numerical algorithms. This crucial importance of parallel numerical algorithms certainly justifies to again having devoted a special workshop to this topic at Euro-Par, in addition to the discussion of special aspects of high-performance and grid computing in Topic 16. Overall, fourteen papers were submitted to our Topic. is the dominant male (as expected at Euro-Par), but three papers are authored by scientists from the United States, Australia, and China. Out of these fourteen submissions, eight were accepted as regular papers. Both devising new parallel algorithms for numerical tasks and adapting existing ones to state-oft-the-art parallel systems are vigorously flourishing and still developing fields of research. Hence, it is no surprise that the eight research articles presented in this section cover a wide range of topics arising in the various subdomains of parallel numerical algorithms. At the conference, the presentations were arranged into three sessions on PDE-Related Topics, ODE-or Particle-Related Topics, and Miscellaneous Topics. This structure also reflects in the following part of the conference's proceedings. In the PDE-Related Topics section, Raimondaš Ciegis addresses a parallel locally one-dimensional scheme for the numerical solution of three-dimensional parabolic problems with nonlocal boundary conditions. Vincent Heuveline and Andrea Walther discuss online checkpointing strategies for parallel adjoint computations as they occur in the optimization or control of time-dependent flow problems. Finally, fault tolerance is the topic of Hatem Ltaief, Marc Garbey, and