Project description
Within this project we will investigate the impact on executing large and very large computationally intensive numerical simulations in a cloud environment in terms of performance, usability, and cost.
The problem in focus is the solution of linear systems of equations, arising from discrete partial differential equations. The solution methods are preconditioned Krylov subspace iterative solvers.
The test application originates from Geophysics. It models the behavior of a (visco-)elastic inhomogeneous material body under load. The discrete model, based on finite element discretization, is implemented in MPI/OpenMP, using toolboxes from the packages deal.ii and Trilinos.
Project Goals
Through this project we intend to address the following questions:
- What is the suitability of cloud environment for compute-intensive scientific applications, compared to classical cluster environment?
- When is it cost effective to run compute-intensive scientific computing applications in a cloud?
- Evalueste and give recommendations what is the best suitable setup for the considered type of application -
bare-metal, virtual servers or Containers-based resources,
where
- bare metal cloud servers are dedicated servers that are not shared;
- virtual servers are instances running on hypervisors, providing the flexibility to quickly scale the application up and;
- containers are light weight kernels thst create minimum load on the infrastructure compare to the virtual machines, for example Docker containers that automate the deployment of applications inside software containers. - Is it feasibile (and when) to provide large-scale computational applications as a service?
Back
Last changed on April 17, 2015
Mail to: Maya dot Neytcheva "at" it dot uu dot se "