2016-02-22

An asynchronous One-shot method for HPC

Dr. Torsten Bosse, Argonne National Laboratory, U.S.A.

31 Mar 2016, 17:00–18:30; Location: S4|10-1

The One-shot method for design optimization problems has been successfully implemented for various physical applications. To this end, a slowly convergent primal fixed-point iteration of the underlying state equation is augmented by an adjoint iteration and a corresponding preconditioned design update. Besides a suitable choice for the preconditioner, a 'correct' sequencing of the steps of the three iterations is most important to allow for an efficient optimization method.

The steps of the three iterations can be performed in various ways, e.g., in parallel, sequential, or combinations thereof. Naturally, each of the steps/tasks has to be allocated to a certain amount of computational resources on HPC platforms.  This leads to the problem that either parts of the resources need to be reallocated to another task or that some of them are idle for a certain amount of time and, thus, performance loss. Therefore, we will motivate and present an asynchronous, parallel One-shot updating scheme with load-balancing in this talk.

The method itself is a generalization of all previously presented One-shot methods. Its local convergence can be enforced by the choice of the design update preconditioner and a suitable load-balance. For both of them, we will provide some ideas how they should be chosen. Furthermore, we will discuss how this approach can be implemented and give an outline of some problems that might arise in doing so.

Category: CE Seminar

Contact

Technische Universität Darmstadt

Graduate School CE
Dolivostraße 15
D-64293 Darmstadt

Phone+49 6151/16-24401
Fax -24404
OfficeS4|10-322

to assistants' office

Open BSc/MSc Theses

Show a list of open BSc/MSc topics at GSC CE.

 Print |  Impressum |  Sitemap |  Search |  Contact |  Privacy Policy
zum Seitenanfangzum Seitenanfang