High Performance Computing in Physics

8 Feb 2011, 08:00 → 9 Feb 2011, 18:00 Asia/Kolkata

AG-66 (Colaba Campus)

This program aims to bring together a community of people who intensively use high performance computing (HPC) to address interesting questions in science. The community of such people in India has grown over the last few years, and the growth is accelerating. The workshop aims to foster a discussion on advances in the HPC landscape and how these advances in HPC can be put to use in research. We aim to pay close attention to computational techniques, HPC requirements, computational performance applications, data management strategies, and physical facility challenges in hosting petaflop computing systems.

Tuesday, 8 February

08:30 → 09:00 Registration and Inauguration

09:00 → 09:30 Challenges for Lattice QCD in the Multipetaflops Age

09:00 → 10:40 Session I

09:30 → 10:00 Quantum Monte Carlo studies of a valence-bond-solid state in two dimensions

10:00 → 10:30 Jamming and Void Space in Particle Packing: from Glasses to Globular Proteins

11:00 → 11:30 QCD Critical Point Search from Supercomputers

11:00 → 12:30 Session 2

11:30 → 12:00 Computation in Astrophysics: present and future challenges

12:00 → 12:30 Symplectic Integrators and Shadow Hamiltonians

14:00 → 15:30 Session 3

14:00 → 14:30 Super-scale Physics Applications on Large HPC Systems

14:30 → 15:00 Revolutionizing HPC with GPUs

15:00 → 15:30 Tape takes on Mass Storage

16:00 → 17:30 Session 4

16:00 → 16:30 Top 7 Challenges of Today's HPC Centers

16:30 → 17:00 To be announced

Wednesday, 9 February

09:00 → 10:30 Parallel Computation for Lattice QCD

09:00 Parallel Computation for Lattice QCD

10:45 → 12:15 Quantum Monte Carlo Algorithms for Spin Systems

10:45 Quantum Monte Carlo Algorithms for Spin Systems

13:15 → 14:15 Cray Programming Environment : The performance-productivity bridge at extreme scale

13:15 Programming Languages for Peta-scale Computing

14:15 → 15:45 Finding selected eigenvalues of the lattice Dirac operator using the Lanczos algorithm with selective re-orthogonalization

An implementation and modification of the LANSO (Lanczos Algorithm with Selective re-Orthogonalization) of Parlett and Scott, for which we just use the (Hermitean) lattice Dirac operator as an example of a useful large sparse (Hermitean) operator. The talk would really be about Krylov spaces, the Lanczos algorithm in floating-point arithmetic, and how to avoid spurious repeated eigenvalues. It would also touch upon the Kaniel-Paige theory of convergence of Ritz pairs to actual eigenpairs, and some possibly new results on the rate of convergence of inner eigenvalues. The work is being done in collaboration with
Chris Johnson (EPCC) using the UK's Cray XT5h system Hector at EPCC.

14:15 Finding selected eigenvalues of the lattice Dirac operator using the Lanczos algorithm with selective re-orthogonalization

Slides Kennedy.pdf.pdf

16:00 → 17:30 Numerical Methods for Compressible Flows