Stanford University HomeContactStanford University Home
Center for Integrated Turbulence Simulations Advanced Simulation and Computing ASC logo
 About CITS Research Simulation Publications Software
Research Projects
RANS
LES
Integration
Merrimac



 CITS Home

Research Projects

Simulation Framework | Software Engineering | Physics Modeling | Verification & Validation

Software Engineering

Scalability study for CITS LES Code. CDP Scalability study for CITS RANS Code. TFLO
Scalability study for CITS LES Code. CDP Scalability study for CITS RANS Code. TFLO


Scalability has been extensively pursued at the Center. Tests have been performed on several platforms; dedicated ALC runs (1,800 CPUs) have demonstrated nearly perfect scalability for TFLO. CDP demonstrated good scalability to 1350 processors. Code I/O has also received considerable attention; both CDP and TFLO use native MPI I/O procedures and have demonstrated I/O rates of 150Mb/sec or greater on a variety of problem sizes. TFLO now uses SubArray types to avoid multiple small requests when writing blocks, and demonstrated writing at 150 MB/s on ALC.

In CDP, after preprocessing, refinement or load-balancing, a new set of restart files must be written based on the data on several processors. CDP uses a coloring algorithm to produce sub-communicators, and then collective writes from Indexed type

On the streaming side, a new hierarchical programming language that makes it significantly easier to develop applications for exposed-communication architectures has been developed. Also a detailed architectural design of a stream processor was completed.



Stanford Seal

Back to top

Copyright © 2004 Center for Integrated Turbulence Simulations, Stanford University. All rights reserved.