 |
 |
Heat and Mass Transfer Tecnological Center of the Technical University of Catalonia
|
Papers:
|
What am I doing here:
|
Now I'm working with DPC. It's CFD research code with a plenty of physical models.
The code is for structured meshes 2D and 3D. It is limited parallel: it use non-parallel solvers.
Subdomains are computed separately and coupled only by special boundary conditions.
These boundaries are updated once per SIMPLE iteration. Somehow it works :)
At this moment I'm trying to make explicit parallel solvers work within this technology.
|
|
What I was doing here:
|
First half a year there were only training tasks and books to read.
At the second part of this period task even seemed to be interesting. I was doing parallel object-oriented realizations of direct and iterative solvers for implicit schemes.
I even achieved rather good acceleration about 24 times on 25 CPUs with things like GMRES...
Details are in reports below...
| |
Some old reports:
|
| |
|
|
Local cluster:
| Cluster on AMD Athlons and older
About 120 single-CPU nodes with AMD CPUs.
About 30 of these nodes are new 2.8 GHz with 2Gb of RAM. Others are old 900MHz with 512Mb of RAM
Nodes are common personal computers.
Cluster is built on three switches and hence consist of 3 parts.
Two switches connect old 900MHz nodes Third switch (most likely 1GBit) ñonnects new Athlons2.8. (And of course these switches are linked together)
|
|
