Requirements of engineering applications to obtain water currents behavior in real time and to provide data to environmental risk models led us to the grow the serial code into a distributed version.
Simulations could have been done employing the serial Tidal model but limiting the number of grid points, adjusting the cell size by the available computers.
To get a better quality result and still return the requested data in real time for the environmental applications we decided to develop a parallel version, the PTidal code.
Typical test over the Río de la Plata and adjacent area uses a grid with 1 km cell size, resulting in a mesh of 80.000 cells. A simulation over this mesh, running on a midrange workstation (20 MFlops) employs 2.5 hours for a 3 day modeling working with the serial code, while the distributed code running in a cluster of 6 midrange workstations could perform in 1/4 this time, 40 minutes.
Whereas a 12-day long storm surge simulation will require 10 hours using Tidal code, it will employ only 2.5 hours running with PTidal. We must take into account that, since storm surge simulations are based in meteorological predictions, several simulations should be performed to cover a range of different possible scenarios.
In order to distribute the serial model Tidal, a block DD method has been applied.
Two different approaches of the block domain decomposition have been studied: an explicit-implicit (or block wise implicit) version and a fully-implicit version.