Some Progresses on Krylov Kinear Solvers: Block-variants and Resilience
In this talk we will discuss two recent research activities related to parallel numerical linear algebra techniques for the iterative solution of large sparse linear systems. The first one is related to the design of a block-GMRES variants for the solution of multiple right-hand sides.
It addresses the problems related to spectral augmentation at restart and the partial convergence of some linear combinations of the right-hand sides. In the second part of the presentation we will describe some numerically based resilient techniques based on recover-restart policies to overcome the problem related to data lost either become of memory corruption or node crashes in a HPC context.
We will discuss the robustness of the proposed techniques in CG and GMRES context when the fault rate and the amount of data lost vary. Those activities are part of the Integrated Project Lab: C2S@Exa that will be shortly presented in the introduction as a possible framework for future scientific collaborations between Inria and ANL in the JLESC initiative (Inria-UIUC-ANL-BSC Joint laboratory on Extreme Scale Computing).
2009-Present: Senior Scientist at Inria, HiePACS Project Leader
2005-2005: Full Professor at INPT in Applied Math
1993-2005: Senior Scientist then Deputy Project Leader at CERFACS