Parallel cloud solution of large algebraic multivalued systems.

The present paper deals with the resolution on cloud architecture of synchronous and asynchronous iterative parallel algorithms of stationary or evolution variational inequations formulated by a multivalued model. The performances of synchronous and asynchronous iterative parallel methods are compared with previous ones obtained on cluster or when grid architecture is used. Thanks to the properties of the algebraic systems resulting from problem discretization we are able to analyze the behavior of the iterative algorithm in particular the convergence and the speed of convergence. The implementation of the studied methods on cloud architecture is described. Then we present various applications in particular the solidification of steel in continuous casting, the cavity pressure calculation described by a problem subject to unilateral constraints and finally a financial problem modeled by American option pricing. • Implementation of parallel synchronous and asynchronous algorithms on cloud computing for multivalued problems is presented. • Three distinct applications are presented and solved by the studied method. • Analysis of parallel methods applied to the target problems is recalled. • Implementation of algorithms on cloud architecture is described. • Behavior of both numerical simulation on cloud is compared with cluster and grid simulation. [ABSTRACT FROM AUTHOR]