Algorithm based fault tolerance with wavelets

Roman Andreev (1, 2, 3)

1 LJLL - Laboratoire Jacques-Louis Lions
2 MOKAPLAN - Méthodes numériques pour le problème de Monge-Kantorovich et Applications en sciences sociales
3 CEREMADE - CEntre de REcherches en MAthématiques de la DEcision

In wavelet Galerkin discretizations of partial differential equations the value of a coefficient of the discrete solution directly translates into its importance. In the context of a parallel iterative computation on faulty computational nodes we propose to use this information to distribute the coefficients in a way that minimizes the expected loss upon hard node failure, thus dramatically increasing the chances of approaching the discretization accuracy.

Mots clés: algorithm based; fault tolerance; wavelets; partial differential equations; parallel computation

Domaines: Analyse numérique [math.NA]; Performance et fiabilité [cs.PF]; Calcul parallèle, distribué et partagé [cs.DC]; Algorithme et structure de données [cs.DS]; Analyse numérique [cs.NA]

Roman Andreev. Algorithm based fault tolerance with wavelets. 2016. ⟨hal-01359755⟩